MONTE CARLO AGENT-BASED HOSPITAL EVACUATION

SIMULATIONS.
THE PRINCIPLES OF PERFORMANCE-BASED INCLUSIVE
DESIGN
A. Tinaburri 1, F.A. Ponziani 1

1 CNVVF. Largo Santa Barbara 2, Rome, 00178, Italy

E-mail address: [email protected]

ABSTRACT
In emergency, while the majority of the occupants recognise the risk and start the evacuation process
autonomously, the people who need assistance do not necessarily act in that way. Trained and skilled
staff are required to rescue those needing help, avoiding the risk of discriminatory response and
failure. A few modelling approaches are available to assess the egress of people who need assistance
to evacuate, since both physical and cognitive aspects should be considered. In this study, an agent-
based approach has been adopted with a focus on the behavioural rules assigned to the agents
depending on their characteristics and goals. A unified framework has been proposed to establish a
standard codification of the occupant profiles: based on their mobility and way-finding abilities,
occupants are classified into a basic set of five categories. For each occupant category, the mobility
device and the staff assistance eventually required are specified. Apart from occupant characteristics,
it is necessary to define the service discipline. consisting of three components: the staff skills and
consistency, the scheduling policy, and the mobility device eventually required to relocate the
assisted occupant. The movement of people as groups is also considered. The inclusive approach
proposed in this study has been implemented using the Pathfinder software and applied to the
analysis of the assisted evacuation scenario of a hospital ward. In order to demonstrate the ability to
simulate complex evacuation plans, the horizontal relocation of the In-patients to refuge areas
located on the same floor is combined with the vertical transfer of one In-patient using a firefighters
lift. The stochastics variables representing the occupant characteristics and the service discipline
have been described by probability distributions, including both autonomous and assisted profiles.
The Monte Carlo methods provide the means to address the parameter uncertainty in probabilistic
risk analysis. Based on the Central Limit Theorem (CLT), the number of trials required for a specified
accuracy in the evacuation time modelling is obtained using a predictor-corrector scheme applied to
the worked case. The principles of performance-based inclusive design are thus established, with
proper consideration of the occupants who need assistance, the assisting staff role and the service
discipline. This approach can be considered a generalization of the conventional Required Safe
Escape Time (RSET) evolving into the Required Safe and inclusive Escape Time (RiSET).
Keywords: People with disabilities; Mobility impairment; Horizontal assisted evacuation; Monte Carlo
simulation, Human behaviour; Evacuation modelling; Inclusive design; Convergence criteria

1. Introduction
Life safety in buildings and its technical requirements respond to a wider range of concerns than
just fire hazards, requiring an ordered and controlled movement of people in emergency conditions
and planning in advance where people can be safely located [1-3]. Protection of occupants is achieved
by the combination of prevention, protection, egress, and other measures [4-5]. When deciding the
strategy of evacuation, all forms of escape routes should be considered, including the use of refuges
for people with disabilities or lifts specifically designed to provide protection from the immediate
danger from fire [6]. The number of and skills and training of the staff assisting the occupants in the
1
evacuation scenario should also be considered, especially if someone cannot autonomously reach a
place of safety [7-10]. Life safety goals and objectives shall be met with due consideration for the
occupancy functional requirements. Typical emergency procedures aim to avoid the need for
simultaneous evacuation, especially in buildings which accommodate occupants who are mostly
incapable of self-preservation, due to their age or physical/cognitive disability, or are confined in
locked rooms or wards in places of “lawful custody” (e.g., prisons, police stations and specialized
psychiatric hospitals). When required by the evacuation plan or prescribed by national or local codes
[1,4-5], refuge areas are provided to limit the number of people impacted and reduce the distress
that might be caused to vulnerable occupants, serving as temporary locations that provide a place of
relative safety to a predefined number of occupants. A refuge area might be another building
connected by a bridge, a compartment of a subdivided story, a protected lobby or stairway or a
corridor complying with the accessibility requirements foreseen in the national building regulations
and standards [11-12]. Fire safety codes provide prescriptive guidance concerning the means of
escape and the evacuation strategy. Provision of adequate escape routes and refuge areas equipped
with means of communication to a manned location, staff reaction and preparedness, notification and
communications systems to alert the occupants are key elements to be considered, especially in
occupancies such as health and day care facilities, where there are likely to be many persons to be
assisted in an emergency situation [13-14].
A full review of evacuation models is given for instance by Kuligowski et al. [15-16], Vermuyten et
al. [17], Ronchi et al. [19-20] and unveils that most published studies are focused on the autonomous
evacuation in building or transportation. Few studies deal with the numerical simulations or
experiments of assisted evacuation scenarios, mainly concerning hospitals: Tsuchiya et al. [21-23],
De-Ching et al. [24], Golmohammadi and Shimshak [25], Ze-min et al. [26], Alonso et al. [27-28], Hunt
et al. [29-32], Yokouchi et al. [33], Ursetta et al. [34], Rahouti et al. [7].
This paper presents three main contributions. First it reports a standard codification of the
occupants, including those who need assistance, suitable to be implemented in agent-based models
(ABM) for evacuation developed over recent years. Based on their mobility and way-finding abilities,
occupants are classified into a basic set of five categories, considering the general disabilities
categories proposed by the NFPA DARAC guide [35]. Each category may originate a variety of
occupant profiles which are applied to specific groups of building occupants. Assisted evacuation is
modelled as a queueing process, with a prescribed service discipline, consisting of three components:
the staff skills and consistency, the scheduling policy, and the mobility devices eventually required to
relocate the assisted occupant. The relocation tasks are executed according to the scheduling policy,
by one or more assisting operators having the necessary skills, using a mobility device if needed.
The principles introduced in Section 2 are applied in Section 3 to the simulation of the horizontal
assisted evacuation of a hospital ward combined with the vertical transfer of one in-patient using a
firefighter elevator, showing the potential to assess complex evacuation strategies. As most input
parameters are represented by stochastic variables, a single scenario may produce a distribution of
possible outcomes (e.g., the evacuation time). It is therefore critical to enable the designer to make
informed decisions on when to terminate the simulations. Researchers have proposed various
methods for judging when enough simulations have been run [36-39]. The second contribution is a
prediction-correction convergence scheme suitable to be implemented in existing agent-based
models, to obtain an estimate of the total number of repeated runs required to obtain a specified
accuracy in evaluating the evacuation time (and the relocation times) distribution. Besides the mean
and the standard deviation, the method also includes the proportion of the simulations concluded
with all occupants relocated or exited, being a measure of system performance. The application of
the convergence scheme to the worked case provides an example of its use and strength.
The third contribution concerns the generalization of the conventional Required Safe Escape Time
(RSET) to include the occupants who need assistance and their service discipline, introducing the
Required Safe and inclusive Escape Time (RiSET) concept. While the RSET calculation is traditionally
based on deterministic methods, RiSET requires a probabilistic risk assessment. The 99th percentile
evacuation time prediction is proposed as a key parameter to calculate one characteristic RiSET value
to be used in performance-based design.
This approach will well serve the fire protection engineering professional by providing guidance
on the quantitative human behaviour analysis needed in the performance-based design, relating
theory to practice.

2. Method
2.1 Occupant profiles and evacuation capabilities
The Protective Action Decision Model (PADM) provides a theoretical framework that describes
the information flow and decision-making that influences protective actions taken by occupants in
response to an emergency scenario [40]. In case of a fire, the perceptions of the physical cues (e.g.,
the sight of smoke or the heat exposure or the asphyxiant or toxic gases inhalation) as well as
information from the social environment (e.g., from people inside the building or emergency
messages or warnings) must be processed by each occupant to assess the threat and eventually start
the evacuation [41].
The selection of the occupant profiles to be used in the evacuation modelling is a critical task and
should provide an accurate reflection of the expected population of building users. Occupant
characteristics include factors such as gender, age, physical/cognitive/sensory capabilities,
familiarity with the building, participation in emergency training, social and cultural roles, presence
of others and commitment to activities [42]. Four basic people characteristics have been identified in
the Life Safety Code © [1] that can affect a fire safety system’s ability to meet life safety objectives:
sensibility to physical cues, reactivity (ability to interpret cues correctly and take appropriate action),
mobility and susceptibility to products of combustion. Individual physical and mental capabilities must
be combined with social and contextual factors like alertness, the condition of being alone or with
others, familiarity with the building and training. Separated group members are likely to attempt to
re-establish their unity before moving towards the exit and their speed of movement is often dictated
by that of the slowest member while attempting to stay together in proximity [43-44].
Evacuation capability is defined in [1] as the ability of occupants, residents, and staff as a group
either to evacuate a building or to relocate from the point of occupancy to a point of safety. It is a
function of both the ability of the occupants to evacuate and the assistance provided by the staff, if
any. It is determined using the procedure acceptable to the Authority Having Jurisdiction (AHJ) with
the application of the standard NFPA 101A [45] or evaluated “experimentally” by a program of drills
performed by persons approved by or acceptable to the AHJ.
If an occupant cannot reach the public way or an area of refuge with minimal intervention from
staff members, such as a verbal or a visual (e.g., sign language) communication, classification as
incapable of self-preservation [46-47] should be considered and active staff/emergency response
assistance during the evacuation or relocation activities should be included in the emergency plan.
Examples of direct intervention by staff members include carrying an occupant, pushing an occupant
outside in a wheelchair or bed or stretcher, and guiding an occupant by direct hand-holding or
continued bodily contact, as detailed in the NFPA DARAC guide [35] that outlines the four elements
of evacuation information that occupants need: notification, way finding, use of the way, and
assistance.
Occupant disabilities can be classified according to the general categories - mobility impairments,
visual impairments, hearing impairments, speech impairments, cognitive impairments - reported in
Table 1, derived from NFPA DARAC guide [35]. It is not uncommon for people to have multiple
disabilities, combining for instance mobility impairment with cognitive or sensory deficit. A similar
approach has been proposed in Italy by an expert panel (Serra [48]), inspired by the International
Classification of Functioning, disability and health (ICF) [49], developed by the World Health
Organization (WHO).
Table 1: Disabilities classification derived from NFPA DARAC Guide (2016) [35].
General category Examples of mobility devices required
Mobility Ambulatory mobility Canes, crutches, walkers
Wheelchair users Power-driven or manually operated wheelchair
Respiratory Depending on the case
Blind or Low vision Canes, service animals.
Deaf or Hard of hearing
Speech disabilities
Cognitive disabilities Depending on the case
Temporary disabilities Depending on the case

The disabilities classification reported in Table 1 is reformulated as shown in Table 2 for

application in evacuation agent-based computer models. Mobility is combined with way-finding
ability to obtain a basic set of five occupant categories establishing a standard codification of
evacuation capabilities. For each occupant category, the mobility device and the staff assistance
eventually required are specified [95]. The Autonomous category applies to occupants having full way
finding capability and ability to independently walk on even and uneven surfaces and negotiate
stairs. Depending on the skills and training received, this category might assist other occupant
categories. The Autonomous with mobility devices category refers to occupants having full way finding
capability but impaired in their movements by the necessity to use a mobility device. Type a) applies
to those occupants that can move/walk independently through an accessible route, at least for
relocation on the same floor. Type b) is reserved to those occupants that may also be able without
supervision to negotiate stairs with the use of a one-handed device. The Autonomous requiring
assistance in way finding or notification category refers to occupants requiring assistance. Type a)
applies to those occupants able to walk but requiring assistance in way finding or walking, due to
their age or sensory impairments or unhealthy conditions. Type b) is reserved to those occupants
requiring assistance only needs to be notified of the emergency. The last two categories – Not
autonomous – apply to the assisted evacuation of patients transferrable only using a mobility device
or a bed/incubator. In both cases, Type a) applies to patients transferrable only an accessible route
(for relocation on the same floor). Type b) is reserved to patients transferrable on stairs. In the case
of a patient transferred using a mobility device, it refers to the use of an emergency travel device or a
firefighters lift (e.g., complying with the standard EN 81-72 [50], clause 5.2.3) accessible for a
wheelchair or a stretcher (e.g., types 3 to 5 according to the standard EN 81-70 [50]). In the case of a
bed, the patient is transferrable on stairs only by means of a firefighter lift, with adequate accessibility
(e.g., type 5 according to the standard EN 81-70 [51]).
Anthropometric data may be also considered to introduce further distinction related to gender or
age or body shape, being mainly reserved to characterize the autonomous profiles.
The term meta-communication has been introduced by Ponziani et al. [5152] to identify the
interaction (e.g., the set of actions and verbal and visual communication) that is necessary to establish
with the assisted occupant in order to include people with disabilities in the evacuation process. As
the meta-communication may require specific abilities and training to the care giver, it is necessary
to distinguish in the simulation (and in the evacuation instructions) the roles of staff employees and
emergency responders, depending on their skills and the characteristics of the assisted people. When
a link is established, a group movement scheme shall be considered in the evacuation modelling, with
the care giver acting as a leader with the responsibility to select the travel path.
The degree of impairment should be considered together with building environmental factors; for
instance the locomotive ability of an individual can be enough to move effectively along corridors or
limited inclination ramps (e.g., Type a) sub-category) but inadequate to descend a stairway (e.g., Type
b) sub-category). Hence the evacuation capability assessment shall be properly conducted taking into
account both the specific occupancy and the population investigated.
Table 2: Occupant evacuation capabilities basic categories
Mobility and Mobility Assistance Examples
way finding capabilities device
category
1. Autonomous  Staff/Emergency
Full way finding capability and ability to response teams
independently walk on even and uneven  Walking patients
surfaces and negotiate stairs. (priority classification
Depending on the skills and training, may
level 4 1)
assist other categories.
2. Autonomous with mobility devices Cane,  Temporary or
Full way finding capability. crutch, permanent disabilities
Type a): move/walk independently walker,
through an accessible route (at least for wheelchair
relocation on the same floor).
Type b): with the use of a one-handed
device may also be able to negotiate stairs
without supervision.
3. Autonomous requiring assistance in 1 or 2  Blind or Low vision
way finding or notification operators for persons
Type a): able to walk on even and uneven each  Cognitive disabilities
surfaces and negotiate stairs only with the autonomous  Children
assistance of another person in way finding walking  Deaf or Hard of hearing
or walking. occupant
Type b): able to walk on even and uneven (only to be notified of
surfaces and negotiate stairs but requiring the emergency)
assistance only to be notified of the  Walking patients
emergency. (priority classification
level 3 1)
4. Non autonomous Wheelchair, 1 to 4  Non autonomous
- mobility device required stretcher, operators for patients
Type a): transferrable only on a rescue sheet, each (priority classification
wheelchair, a stretcher or a rescue sheet emergency assisted level 2 1)
through an accessible route (for relocation stair travel person
on the same floor). device
Type b): transferrable on stairs with
emergency travel devices or by means of a
firefighters lift (e.g., complying with EN 81-
72 [50], clause 5.2.3) accessible for a
wheelchair or stretcher (e.g., types 3 to 5
according to EN 81-70 [51]).
5. Not autonomous – Transferrable only Bed, incubator 1 to 2  Critical patients
with beds or incubators operators for (priority classification
Type a): transferrable only with a bed or each assisted level 1 1)
incubator through an accessible route (for person
relocation on the same floor).
Type b): transferrable on stairs only by
means of a firefighters lift (e.g., complying
with EN 81-72 [50], clause 5.2.3) with
adequate accessibility (e.g., type 5 according
to EN 81-70 [51]).
1 Patients priority classification according to the National Association for Home & Care Hospice [53].
2.2 Basic occupant profiles
Each of the five general categories reported in Table 2 may originate several occupant profiles,
depending on the occupancy considered. A basic set of autonomous profiles, including mobility
impaired people, is given in Table 3, with the key parameters required to describe the individual
horizontal evacuation capabilities in ABM simulations. For the sake of simplicity, no gender or age
differentiation is here considered, even if it could be necessary at least for the Active staff and the
Emergency responders, which are the profiles in charge of the assistance tasks. It might be possible
to use the gender and/or age split when the specifics of the model scenario are known. A
comprehensive literature review of unassisted movement speeds for people with disabilities is
available in the SFPE Handbook of Fire Protection Engineering [16], in Appendix G of ISO/TR 16738
[54], and in Geoerg et al. [55].
Table 3: Basic evacuation capabilities for autonomous occupants, including mobility impairments
Autonomous Unrestricted walking speed (m/s) Social grouping/ Remarks
occupant profile (on level terrain, straight-line) Role
Distribution law
Type μ σ Min Max
Active staff Normal 1 1.35 0.25 μ -2.8σ μ +2.8σ Individual or
(in each fire compartment) assistance team
member Familiar &
Emergency response Assumed equal to Active staff Individual or Trained
(in the emergency control center) assistance team
member
Generic autonomous Normal 2 1.20 0.20 μ -3.0σ μ +3.0σ Individual or Uncertain
occupant groups, eventually
familiarity &
linked to one
Not Trained
assisted profile
Assumed equal to generic autonomous
Worker (not in charge of Individual or Familiar &
evacuation assistance)occupant with co-workers Trained
Autonomous in-patient Normal 3 0.95 0.32 μ -2.2σ μ +2.2σ
(in hospital and care homes)

Crutches Normal 3 0.94 0.30 μ -1.0σ μ +1.4σ

but mobility impaired 4

Walking stick Normal 3 0.81 0.38 μ -1.4σ μ +2.0σ Individual

Autonomous

Uncertain
or linked to
Rollator or Normal 3 0.57 0.29 μ -1.6σ μ +1.6σ autonomous familiarity &
walking frames Not Trained
occupants
Electric Constant 3 0.89
wheelchair
Manual Normal 3 0.69 0.35 μ -1.6σ μ +1.9σ
wheelchair
1 Based on Alonso and Ronchi [28] averaged data for health care staff members. Data differentiated for gender
are available in the IMO guidelines [56], which assumes a uniform distribution of velocities in the range 0.93-
1.55 m/s for female and 0.93-1.55 m/s for male members of the crew.
2 Based on Fruin [56] averaged data (all age classes and gender); similar values are reported in Boyce et al. [58]

A constant speed of 1.19 m/s is proposed in the SFPE Engineering Guide to Human Behavior in Fire [42].
3 Based on Boyce et al. data [ 58]. Other sources of data are available in literature (e.g., Miyazaki et al. [59]).
4 A simplified approach is proposed by Alonso et al. [27-28] with a unique profile, assuming a uniform

distribution of velocities in the range 0.84-1.40 m/s.

 It is convenient to categorise the “rescuers”, through the assignment to different “emergency”
teams, the Active staff and the Emergency response personnel, to prescribe specific set of rules
concerning the use of the means of escape (e.g., elevators) or specialised skills. Following Gwynne et
al. [60-61], the Active staff profile is here used to identify those employees having a procedural role
in case of emergency, rather than simply being responsible for evacuating as quickly as possible. It is
assumed that the Active staff are already assembled in the corresponding compartment and are
prepared for performing the assigned evacuation tasks [27-28, 60-61]. The Emergency response
profile is reserved to occupancies protected by an emergency control center, which is usually
provided in all buildings designed for phased evacuation and in large and complex buildings [13, 62].
The emergency responders are specialized operators who are trained in the building emergency
management systems and procedures, supervising the protected activity from their control center.
These operators are not firefighters of the local fire and rescue service. A basic set of assisted profiles,
including autonomous occupants requiring support in way finding and those who need the aid of
mobility devices, is given in Table 4. The assisted ambulant profile applies to occupants with
cognitive or sensory impairments requiring support, even only for emergency notification, but no
mobility aid. It is assumed that a constant number of assisting operators (but variable for each
assisted profile) performs both the preparation phase, discussed in the following section 2.3, and the
transportation phase, using the prescribed mobility device.

Table 4: Basic assisted occupant profiles and travel speed in the horizontal route
Assisted occupant profiles Assisted travel speed (m/s) Number of
(on level terrain, straight-line) assisting
Distribution law operators
Type  σ Min Max
Assisted ambulant 1 Normal 0,71 0.34 μ -1.7σ μ +1.8σ 1 operator 3
Assisted transported on a wheelchair 2 Normal 0,63 0,04 μ -3.0σ μ +3.0σ 1 operator 3
Assisted transported on a carry or evac Uniform 1,34 1,75 1 operator 3
chair 4
Assisted transported on a bed 2 Normal 0,40 0,04 μ -3.0σ μ +3.0σ 2 operators
Assisted transported with hand-held Uniform 0,52 1,23 2 operators
rescue sheet 4
Assisted transported with a hand-held Uniform 0,91 1,23 4 operators 5
stretcher 4
1 Based on Boyce et al. data [58].
2 Based on Alonso et al. data [27-28] (minimum and maximum values assumed). Data for evacuation and carry
chairs transportation are available in Hunt et al. [30-31].
3 An additional operator may be needed to prepare the patient for transportation or assist along the travel path.
4 Based on Hunt et al. [29-31] overall data, discarding gender differentiation.
5 Could be reduced to two operators only to execute the task to prepare the patient for transportation.

Assisted evacuation is usually considered in health care evacuations and is modelled as a queueing
process where several “clients” (the occupants who need assistance) request the service of one or
more assisting operators having the necessary skills to help [7, 28, 63]. It is therefore necessary to
define the service discipline, consisting of three components: the number and skills of assisting
operators, the scheduling policy, and the mobility devices eventually required to relocate the assisted
occupant. Various scheduling policies can be adopted but usually the following two suffice:
- priority: assisted occupants with the highest-ranking need are served first;
- distance: assisted occupants closest to a free assisting member, having the necessary skills, are
served first.
In the first case it is necessary to establish an evacuation order list. In both cases, if more than one
assisting operator is needed, the assisted occupant shall wait for the whole team to assemble.
2.3 Pre-travel activity times and preparation times
The pre-travel time or pre-travel activity time (PTAT) is defined in international standards and
technical documents [42, 54, 64]as the interval between the time at which a warning of a fire is given
and the time at which the first move is made by an occupant towards the exit. The PTAT consists of
two components: recognition time and response time. The recognition time is the interval between the
time at which an alarm warning of a fire is given and the first observable response to the warning.
Depending on the fire scenario, occupants might be aware of various fire cues before or after a
warning is given. The response time is the interval between the time at which the first observable
response to the event occurs and the time at which the movement begins towards an exit or a safe
location. The provision of reliable data on the pre-travel activity times expected in various situations
and their incorporation into evacuation agent-based models is an important requirement for the
assessment of evacuation time. Guidance on default values is given in Annex E of PD 7974-6 [64],
which updates Annex E of ISO/TR 16738 [54]. The conventional definition of PTAT is applicable only
to the autonomous occupant categories not requiring assistance in evacuation (self-evacuation,
categories 1 and 2 in Table 1). For the emergency teams (Active staff and Emergency response
profiles), a different PTAT definition is proposed in literature [65-66] as the time elapsed between
the warning of fire being given (e.g., the alarm) and positive evacuation activities by staff. This
differentiation is important as this time relates to their participation in the procedure rather than
only their self-evacuation; it therefore directly relates to the time for staff to interpret the nature of
the event and commence their response [67]. PTAT can be represented by a log-normal statistical
distribution [27-28, 64]. For well-managed cases (denoted by M1 in PD 7479-6 [64] and ISO/TR
16738 [54]) such as the scenario investigated in Section 3, the minimum PTAT for autonomous
occupants in engineering application is in the order of 30 s. Active staff’s PTAT mean and standard
deviation values are taken from Alonso et al. [27-28], for health care occupancies. The maximum value
is reported in Gwynne et al. [60-61]. If the occupancy is protected by an emergency control center
permanently staffed by Emergence responders, shorter PTAT parameters are expected for this
profile than those assigned to the Active staff, due to their roles and responsibility [42]. Emergency
responders are assumed in this study to move within 60 s upon receiving the alarm notification. All
the other occupants having autonomous evacuation capabilities are specified to start their movement
within 120 s (ISO/TR 16738 [54] for awake and unfamiliar in medical care occupancy). The PTAT
adopted in this study for autonomous occupant profiles are given in Table 5. Active staff have higher
pre-evacuation statistical parameters than Other autonomous profiles, having uncertain familiarity
and training. This could occur by several reasons. First, different PTAT definitions applies to
emergency teams and to the other autonomous profiles, as this time relates to their participation in
the procedure rather than only their self-evacuation. Secondly, the table is based on available
literature data obtained by different sources and, when specific data are lacking, expert judgement is
used. This issue outlines the importance of calibrating model input to establish whether the results
of the simulations correctly predict the relevant evacuation scenario [42].

Table 5: PTAT for autonomous occupant profiles adopted in the simulation of assisted evacuation
Autonomous occupant Pre-travel activity time (s) Remarks
Profile Distribution law
Type μ σ Min Max
Active Staff Log-normal 1 711 60 1 30 3 246 2 Familiar & Trained
Emergency response Log-normal 3 43 6.44 30 3 60 Familiar & Trained
Other autonomous profiles Log-normal 62.7 19.11 30
4 4 120 4 Uncertain familiarity and
(Autonomous occupant, including training & Not grouped
those mobility impaired) with an assisted occupant
1 Based on Alonso et al. data [27-28] for health care staff.
2 Valuesderived from Gwynne et al. [60-61].
3 Based on ISO/TR 16738 [54]; data range for awake & familiar profiles in level M1 occupancies.
4 Based on ISO/TR 16738 [54]; data range for awake & unfamiliar profiles in level M1 occupancies.
 For occupants who need assistance to evacuate, it is necessary to define the preparation time,
representing not only the time required to prepare the occupant with mobility impairment for
relocation [27-28] but also the time (and skills) needed to establish a communication link with a
person having cognitive or sensory impairments. The timings given in Table 6 are only for guidance
in medical care occupancies and depend on the assisted people involved, the staff training, the
equipment available and the degree of maintenance provided.

Table 6: Preparation times for assisted occupant profiles

Preparation time (s)
Assisted occupant profile Distribution law
Type μ σ Min Max
Assisted ambulant 1 Normal 60 20 μ -1.5σ μ +1.5σ
Assisted transported on a wheelchair 1 Normal 110 36 μ -0.3σ 3 μ +0.3σ 3
Assisted transported on a bed Assumed equal to assisted on a wheelchair
Assisted transported on a carry or evac chair 2 Normal 41.5 7.9 μ -1.2σ μ +1.3σ
Assisted transported with a hand-held rescue sheet 2 Normal 65.2 14.1 μ -1.4σ μ +1.5σ
Assisted transported with a hand-held stretcher 2 Normal 77.7 19.2 μ -0.9σ μ +2.2σ
1 Based on Alonso et al. data [27-28].
2 Based on Hunt et al. overall data [29-30] for carry chair for an assisting team of two health care operators.
3 Based on Hunt et al. overall data [29-30] for an assisting team of two health care operators.

2.4 Movement Groups

The nature of the social relationship between the occupant and the surrounding population is
among the factors that can influence evacuation performance. A group of occupants who share an
affiliation link, such as a family or a visitor to an in-patient or a guided group, will have a strong
tendency to stay together and move as a group [68-70], sharing their way-finding behaviour
following a leader while travelling toward a common destination. Movement groups schemes are
thus determined by the type of occupancy considered. In inclusive design, groups should also
consider occupants who need assistance, due to their physical or cognitive impairments [7, 28].
Movement schemes during the evacuation are differentiated in two classes, depending on the
characteristics of the occupants forming each group:
a) groups formed only by occupants having autonomous evacuation capabilities (category I and II
in Table 2, profiles listed in Table 3);
b) groups including both assisted occupants (category III, IV and V in Table 2, and profiles listed in
Table 4) and the autonomous occupants (different from the assisting staff).

2.5 Evacuation time probability distribution: number of iterations and accuracy of single loop Monte
Carlo methods
As our evacuation model includes the use of probabilistic variables to simulate the variability of
possible agent behaviours, a single scenario may produce a distribution of different occupant
evacuation time curves. Monte Carlo (MC) simulations, which rely on repeated random sampling of
a problem having multiple stochastic inputs in order to generate a distribution of possible results,
represent a method of analysis particularly suited to deal with the system of interest.
According to PD 7974-7, a probabilistic risk assessment (PRA) should reflect the variability in the
risk and take into account the uncertainty associated with the risk estimate. Sources of uncertainties
in evacuation modelling are classified into three main groups [71]:
a) parameters uncertainties (epistemic or aleatory), concerning specific estimates or values used in
setting up the evacuation model, mainly related to the human factors: pre-evacuation times,
walking speeds, preparation times, anthropometric data;
b) model uncertainty, related to the assumptions made in underlying scientific knowledge and
theoretical models or empirical relationships on which is based the selected evacuation model;
 c) completeness uncertainty, concerning issues that are excluded from the analysis but are known
to exist, such as human behaviours deviating from rationality or not following the optimal
strategy in wayfinding or propensity for fatigue.
As a first step, we assume that no epistemic uncertainty exists in the evacuation model; the
probability distributions listed in Tables 3 to 6 represent the intrinsic randomness of the relevant
parameters. The MC methods provide a mean to address the parameters uncertainty and can be
considered as the most basic form of random sampling of a problem having multiple stochastic inputs
to generate the probability distributions of possible outputs. Random realizations of all stochastic
variables are initially generated and are used within the deterministic simulation model in order to
return a single model realization. The process is repeated in order to evaluate the range and
probability distribution of possible outcomes: this scheme is also known as the single loop (1D) MC
solution. Alternative sampling schemes exist, such as Latin hypercube or orthogonal sampling, which
samples more accurately from the entire range of input distribution functions, or more complex MC
methods are available, such as the double-loop MC method [71] or the “boostrapping” technique [72].
As more sophisticated approaches are restricted to the research applications and are not
implemented in any available evacuation software for fire safety engineering applications and
considering the burden on computing resources required by performing hundreds or thousands of
repetitions as requested by bootstrap statistical technique, a single loop Monte Carlo technique has
been selected in this study. PRA based on MC trials should provide a clear statement regarding the
relationship between the number of iterations executed, n, and the confidence limits, associated with
a specified supplied confidence level(s), for the outputs of interest. The number of trials necessary is
dependent on both the degree of accuracy and the sensitivity of the results of interest (mean,
variance, proportion of population, etc.), especially if a tail of the distribution is investigated [71]. The
procedure is here described with reference to the evacuation time, the key variable in fire safety
engineering, but its application to relocation times or occupant exit times is straightforward.
It is anticipated that the convergence of the ET may not guarantee that the full range of evacuation
dynamics has been adequately represented [38], as will be shown in Section 4.2 with reference to the
worked case described in Section 3.

2.5.1 The number of Monte Carlo simulations required to obtain the desired precision
Before we obtain data there is uncertainty about the evacuation (and relocation) times
distribution. The number of iterations n required to achieve a specific bound on the half-width of the
Confidence Interval (CI) cannot easily be estimated a priori. Many inferential statistics are based on
the properties of the sampling distribution mean, whose importance derives from its use in drawing
conclusions about the population mean, thanks to the Central Limit Theorem (CLT), which states that
as the sample size increases the sample mean will be normally distributed for most underlying
distributions. Assuming the model is correct, one can obtain as small a statistical error as desired by
conducting a sufficiently large number of trials.
Let ET represent the evacuation time population for a given scenario with a priori unknown
distribution, having a mean of and a variance . The calculated simulations results { et1,…,etn }
are assumed to be the result of a random sample ET1, …, ETn of size n drawn from the ET population.
Let then consider the distribution of the sample mean, , consisting of all estimations of the ET

. If n is sufficiently large, the CLT states that

distribution’s mean possible by averaging random samples of size n drawn from ET, that has a mean
of and a variance asymptotically approach the
normal distribution, no matter the ET population distribution, so that:

= ; = / (1)

and the 100*(1 – α) % two-tails classical confidence interval for the mean of the ET distribution
is given approximately by:
 / < < / )≈1−
/√
P(− (2)

⁄√ (a quantity
If ET has a normal distribution, irrespective of the sample size n, the sample mean is normally

is called Δ, the half-width

distributed with expected mean equal to and standard deviation equal to
often called the standard error of the mean estimator). The quantity /
√
of the Confidence Interval (CI) of the mean, associated with a two-tails  confidence level.
Using the sample average ( "! = ∑$%& !$ / ) and standard deviation ('() = *∑$%&( !$ − "!) /( − 1)) as
unbiased estimators for the and , the classical CI for is obtained:

"! − /
,-.
< < ( "! + /
,-.
)= !012
√ √
(3a)

Δ =
,-.
/
√
(3b)

Formulas 3a) and 3b) should be strictly applied only when n is sufficiently large to justify their

The accuracy of the mean estimator (Δ ) should increase as the number of trials, n, increases; if the
use but have the strong advantage to be valid regardless of the shape of the population distribution.

being proportional to 1⁄√ .

model is properly formulated, this is statistically true with the potential error in the approximation

When the number of trials n is small, typically less than about 30 or 40 [73], the t distribution has
to be used, having a similar shape but requiring an additional parameter, the degrees of freedom, ν,
determined as the number of trials minus one. For a small sample, formula 3a) is replaced by:

"! − ! / ,4% &

,-.
< < ( "! + ! / ,4% &
,-.
)= !012
√ √
(3c)

Eq. 3c) is equal to Eq. 3) in Grandison work [38] (and in any study invoking the CLT for small
samples).
Inferences concerning the CIs for the population variance or standard deviation are more
difficult to obtain when the population distribution is not normal. Skewness or heavy tails can have a
drastic impact on the asymptotic coverage probability of the normal-based confidence intervals [74].
Bonett [75] proposed a confidence interval that performs well in small samples under moderate
departures from normality. His interval performs only slightly worse than the exact normal-based
confidence interval when sampling from a normal distribution. A larger sample size provides Bonett
confidence intervals with greater protection against nonnormality. Supposing that the ET population
distribution is normal, the 100*(1 – α) % two-tails classical CI for the variance is:
( &) ( &)
'() = < 2
< '() =
5678/9,:;<76
9 =>< 58/9,:;<76
9 =@A
(3d)

The confidence coefficients - z/2 or t/2,v) or B / ,4% & and B& / ,4% & - are obtained from tables

t and B distribution [73].

reporting their values as a function of confidence levels () and the degrees of freedom (ν) in case of

In evacuation modelling, the designer is generally interested in the upper tail of the evacuation
time distribution [76], being a measure of the system performance (the RSET criteria, discussed in
section 5). This key point is not discussed by Grandison [38] and is only touched with no specific
detail or guidance in Lovreglio e al. [37]. Thus, this paper includes the estimation of the accuracy in
evaluating a specified proportion of the simulations concluded with all occupants correctly relocated or
exited (e.g., the probability p of successfully completing the evacuation at a specified time). The
derivation of the CI for a population proportion, p, based upon a random sample of size n, can be
found in any text on introductory statistics [73]. The Wilson score method is here adopted, being best
suited for small values of n and when p is close to 0 or 1 (extreme values). The half-width of the
CI, ΔC , for the proportion D of ET population with confidence level approximately 100(1-)%, is
given in Eq. 4b), being centred on DE given in Eq. 4a), involving the unknown a priori D̂ , the sample
fraction of “successes”, that is advantageously obtained after conducting a first set of MC trials.

H IJ8/9
G 9
/
DE =
&IJ8/9
9
/
(4a)

KCL(& CL)/ IJ8/9

9
/(M 9 )
ΔC = / &IJ8/9
9 / (4b)

These equations do not state directly how many iterations should be performed or, more
importantly, how accurate the resulting estimation is. These are important issues that the designer
should address including a justification for the criteria adopted. An appealing strategy is to specify a
priori the confidence level, , and the half-interval width, , and estimate the number of required
iterations n to the desired degree of precision, from the application of formulas 3b) or 4b):
9
J8/9 ,-.
9
=
Δ2
(5a)
N 'OP

9
J8/9 CL(& CL) CL(& CL)
C ≅ W∆ − 2∆C + X D̂ (1 − D̂ ) Y 9 − 4\ + 1]
∆S T-U>V< Z
(5b)
T-U>V< S T-U>V< S T-U>V<

As is proportional to the '() , its convergence should be monitored during the MC realizations.
The method proposed in [95] is focused on the accuracy achieved in the estimation of the key
stochastic variable of interest in performance-based design, the evacuation time, rather than the
vector representing the exit/relocation times of each individual occupant considered by other
Authors [36-39]. Moreover, it gives an explicit formulation of the estimate of the required number of
simulations to achieve a desired accuracy in the mean or in the proportion of the evacuation time
distribution (Eqs. 5a) and 5b)).

2.5.1.1. A predictor-corrector convergence scheme for Monte Carlo evacuation modelling

One common way of terminating an egress MC simulation scheme is to continue running until the
observed fluctuation (variation) of the estimated quantity of interest has "stabilized" [36-38, 71].
Although this methodology probably will meet its objectives, it does not allow the user to know in
advance how long the simulation will run. Also, such an approach often requires a larger number of
runs than are really necessary. The convergence scheme is similar to the one proposed by Lovreglio
et al. [37] or by Grandison [38], that adapt to evacuation modelling a common procedure in
conducting MC studies [77].
Suppose that one is prepared to run a minimum of nmin and a maximum of N Monte Carlo
realizations. nmin should be greater than 40 to invoke the use of CLT to estimate the CI regardless of
the nature of the population distribution concerned. The basic procedure is outlined with reference
to the evacuation time, the key variable in fire safety design, but its generalization to relocation times
or occupant exit times is straightforward.

of the mean ( "! 6 ) and of the standard deviation ('()6 ) of the evacuation time (ET). Calculate the
1. Conduct first a small number of egress simulations, in order of n1 ≌10, to obtain rough estimates

first estimate of 6 , using the t distribution critical values (being a small sample). This is the
“prediction” step.

3. Recalculate the mean ( "! 9 ) and standard deviation ('()9 ) of the ET sample and its descriptive
2. Perform an additional round of ≈ 30 simulations (to obtain a sample of about n2 ≌ 40 trials).

bound of the ET population variance ( ()9=>< , ()9=@A ). Using the z statistics, update the achieved
statistics. Check if this ET sample is normally distributed, and if so, calculate the lower and upper
 accuracies (Δ 2 , ΔC 2 ) and the number of trials required to achieve the design accuracy on the
mean ( C 2 9 ). This is the (first) “correction” step, which should refine the estimation on the
number of trials necessary to perform to achieve the required accuracy.

( ^ max ( 2 , 0$ ).
4. Check the convergence termination condition, on the accuracy on the mean of the ET sample:

5. If the termination condition is not verified and < _, perform an additional round of ≈ 30 runs
(or the quantity needed to reach maximum design number N) and repeat step 3 and 4 until
convergence is reached or the maximum number of simulations N is approached (meaning that
the design accuracy is not achieved).
This scheme establishes, within the CLT general framework, the convergence of a set of repeated
evacuation simulations, within predefined acceptance criteria, towards the mean of the ET
population distribution which could be obtained by performing an infinite number of trials. It is
recommended to monitor during the iterative process the convergence of the ET sample variance
and test if normality distribution applies, and simultaneously to check the convergence of the other
key parameters (i.e. RTs).

2.6. Introducing inclusivity in performance-based design: the RiSET criterion

Life safety is the key functional objective of performance-based fire safety design [4, 42, 64]. Life
safety goals consider a time-based comparison between two key factors:
1) ASET – depending upon parameters related to the fire dynamics and tenability conditions for
the occupants
2) RSET – depending on pre-travel and travel behaviours of the occupants
as a mean of demonstrating, conducting a quantitative analysis, that the design meets the
performance criteria.
ASET is calculated by fire models, based on (a cluster of) design fire scenarios while RSET is
determined using evacuation models. The calculations can be executed independently, using
different simulation tools and then reconciling the data in order to determine the time before
untenable conditions exist in occupied spaces, or performed in an integrated application where both
the fire model and the evacuation model are executed simultaneously. A simplified scheme of the
process related to escape is conventionally expressed using a timeline as shown in Figure 1. The
underlying strategy is the simultaneous evacuation of all autonomous occupants (or a group of
occupants immediately at risk in case of phased evacuation) on sounding of an alarm. This approach
is not fully adequate to bring inclusivity inside the fire safety design. In emergency, while the majority
of the occupants recognise the risk and self-evacuate finding autonomously the way-out, the people
who need assistance do not necessarily act in that way.
ASET
Margin of
RSET safety
Evacuation Time (ET)

PTAT Travel Time

Response
Recognition
Alarm
Detection

Ignition Evacuation Evacuation Tenability

start complete limit

Figure 1: Conventional approach: RSET vs ASET [64]

 Trained and skilled staff operators are required to contact those needing help that could
unwillingly otherwise remain inside, avoiding the risk of discriminatory response and failure due to
the lack of additional planning or special accommodation for people with disabilities, reduced staff
or staff not having the necessary training, unavailability of mobility devices [35].
Based on the principles outlined in this Section, this study proposes a generalization of the
conventional Required Safe Escape Time (RSET) introducing the Required Safe and inclusive Escape
Time (RiSET) criterion to include the occupants who need assistance and their service discipline. This
approach allows to establish a standard codification of occupant evacuation capabilities, being
autonomous or needing assistance. The basis of performance-based inclusive design is illustrated in
Figure 2. While the RSET calculation is traditionally based on deterministic methods, RiSET requires
a probabilistic risk analysis. It is assumed that each autonomous occupants, classified in categories 1
and 2, starts its evacuation on sounding of an alarm and begin to move when its PTAT is expired. The
destination is a place of safety for the occupants classified in categories 1 and 2 type a), which are
able to walk on even and uneven surfaces and negotiate stairs, while it is a safe refuge for the
occupants classified in category 2 type b).
Assisted evacuation is required for occupants classified in the categories 3 to 5. Assisting operators
are differentiated in Active staff and the Emergency response to prescribe specific set of rules
concerning the use of the means of escape (e.g., elevators) or specialised skills. The Active staff
identifies those employees having a procedural role in case of emergency, already assembled in the
corresponding compartment and are prepared for performing the assigned evacuation tasks. The
Emergency response are specialized operators who are trained in the building emergency
management systems and procedures, supervising the protected activity from the emergency control
center. Active staff and Emergency response start their PTAT activities on sounding of the alarm.
Again, both the PTAT and unrestricted walking speed, are stochastic variables specific for each
category. Each assisting operator starts to move when its PTAT is expired; the unrestricted walking
speed is used only along the path toward an assisted or after having completed all the assigned
service tasks. The service discipline defines how the occupants who need assistance are served by one
or more assisting operators, consisting of three components: the assisting staff skills and consistency,
the scheduling policy, and the mobility device eventually required to relocate the assisted occupant.
The dashed circle in Figure 2 exemplifies the scheduling policy adopted in serving the occupants who
need assistance, requiring a number of missions to move them from their initial positions to the
design destinations. A mission could be:
- assisting a person who is blind or low vision or with reduced mobility while walking along the
means of egress;
- notifying the emergency to a person who is deaf or hard of hearing and who rely on lip reading for
information;
- establishing a link with a person with cognitive disabilities (requires an operator with special
training or skills, known to the assisted person) and then providing guidance to and/or through the
means of egress;
- preparing a person with reduced mobility to be relocated using a mobility device.
Thus, movement group schemes are a necessary component of any inclusive design model, with
the constraint that a group can include only one assisted occupant with one or more autonomous
occupants sharing a social or assisting link.
The Autonomous occupants requiring assistance only to be notified the emergency, category 3 type
b), move autonomously to a place of safety only after they are contacted by an assisting operator.
The Autonomous occupants requiring assistance in way finding, category 3) type a), applies to those
occupants able to walk but requiring assistance in way finding or walking, due to their age or sensory
impairments or unhealthy conditions.
Categories 4 to 5 – Not autonomous – apply to the assisted evacuation of patients transferrable
only using a mobility device or a bed/incubator. In both cases, Type a) applies to patients transferrable
only an accessible route (for relocation on the same floor). Type b) is reserved to patients
transferrable on stairs. In the case of category 4, occupant transferred using a mobility device, it refers
to the use of an emergency travel device or a firefighter lift. In the case of a bedridden occupant, the
patient is transferrable on stairs only by means of a firefighter lift, with adequate accessibility.
Noting that the PTAT definition is clearly applicable only to the autonomous occupants capable of
self-evacuation, it is proposed to introduce a different term for occupants requiring assistance. PTAT
should be replaced by the PAL time, combining the Preparation time, representing the time required
to prepare the occupant with mobility impairment (or life supporting equipment) for relocation and
the time and skills required to the Assisting staff to establish a communication Link, verbal or a visual
(e.g., sign language), with an occupant having cognitive or sensory impairments.
RiSET requires the application of a computer evacuation agent-based model and reliable
information made available in scientific literature or in guidance documents, calibrated with specific
occupancy data. These principles have been applied in the worked example in Section 3. It must be
emphasized that the literature data are generally restricted to the assessment of fire scenarios
characterized by tenability conditions based upon zero smoke exposure and tolerable heat exposure
[42]; occupants are considered able to escape under tolerable downward heat radiation as they do
in ordinary conditions in a clear and cold air environment.
ASET is generally the result of the deterministic analysis of the design fire scenario (or a cluster of
fire scenarios), obtained by hand calculations or zone or field computer models of the fire dynamic,
incorporating an engineering judgement in case the detection and warning times are based only on
human senses and intervention. Where there is no automatic detection (e.g. level A3 alarm system in
PD 7974-6), the time to general warning is likely to be long and unpredictable, and might be any time
between a few minutes and several hours (in case of smouldering fires) [64].
The life safety performance criterion is associated with the achievement of an absolute target
(acceptance concept ‘AC3’ in PD 7974-7 [71]) translating the limit state for life safety: ASET>RSET
for each individual occupant, being autonomous or assisted. The 99th percentile evacuation time
prediction may be selected as a key parameter to calculate one RiSET value to be compared with the
ASET value corresponding to minimum time in which one occupant could be incapacitated.
The required number of simulations depends on the standard deviation of the ET population and
the accuracy required to obtain an estimate of the extreme values of the distribution. The suggested
number of simulations reported in literature [38] ranges from 10 to 2000. A dynamic assessment of
the behaviour of the ET statistic is recommended to optimize the number of simulations, provided
that a suitable convergence methodology is adopted, monitoring the evolution of the sample standard
deviation and testing for sample normality.
If the sample evacuation time distribution has a normal distribution, the 99th percentile inclusive
evacuation time (iETth) prediction is best obtained using the upper bounds of the mean (Eq. 3a or
Eq. 3c) and standard deviation (Eq. 3d), obtained after performing the MC trials necessary to reach
the convergence on the required accuracy on the mean of the ET distribution:
iETth = !012 + 2.33 ()=@A " +Δ
= (et ) + 2.33 ()=@A (13)
 ASET
Ignition

Detection

Alarm
Design fire scenario

inclusive Evacuation Time (iET)

Margin of
evacuation safety
Autonomous

Evacuation completed RiSET

Tenability limits
Category n.1
Autonomous
Place(s)
Category n.2 PTAT 1 Travel Time 1 of safety
Type b) (Recognition+Response) (Unrestricted walking
Autonomous speed)
with mobility
device
PTAT 2b Travel Time 2b
Category n.2
Type a)
Safe refuge(s)
Autonomous
with mobility PTAT 2a Travel Time 2a
device

Assisting staff
evacuation
Assisted

- Active staff PTAT Active staff

- Emergency
response
PTAT Emergency response
Category n.3
Type b)
Autonomous assisted in Place(s)
notification only
PAL 3b Travel Time 3a
(Unrestricted walking of safety
speed)
Category n.3
Type a)
Autonomous assisted in
way finding or walking PAL 3a Travel Time 3b
(Assisted travel
Category n.4 speed)
Type a) and Type b)
Not Autonomous
Service discipline

Mobility device required

PAL 4 Travel Time 4 Safe refuge(s)
Category n.5 Relocation area(s)
Type a) and Type b)
Not Autonomous
PAL 5 Travel Time 5
Transferrable only with
beds or incubators (Preparation And Link time)

Figure 2: Inclusive approach. The RiSET timeline

3. Results
3.1 Horizontal assisted evacuation of hospital ward combined with the vertical transfer of one In-patient
using a firefighter elevator: a case study
It is widely recognized that emergency evacuation in hospitals and care homes is a challenging
process that requires a strategy, well-trained staff, and careful execution, as it usually involves
vulnerable people with widely varying evacuation capabilities [28, 47, 78]. Researchers have
investigated emergency preparedness in health care facilities as a result of a wide variety of natural
disasters such as hurricanes [79-80], wildfires [81], earthquakes [82], and bomb threats [83], with a
focus on the resilience, e.g. the ability to function and accommodate a massive influx of patients in
the immediate aftermath of crisis situations [84-85]. Other studies deal with the issues that a hospital
faces when the occupants must be evacuated due to an internal emergency [9, 86-88].
Horizontal assisted evacuation of in-patients, where only the affected ward is cleared of its
occupants, is usually required in fire safety codes or regulations [1,13] in order to preserve the
hospital functionality. The evacuation process of a medical ward requires as a first step the relocation
of the patients to one or more areas of refuge located on the same floor, in accordance with an
established emergency actions plan. Table 7 reports the key design prescriptions for existing health
care occupancies concerning the horizontal portion of the escape route from Chapter 19 of NFPA 101
[1] and the Italian fire safety regulation [13]. Similar requirements are imposed and both codes are
in line with the applicable accessibility regulations.

Table 7: Key prescriptions concerning the horizontal portion of the escape route and relocation areas.
Existing health care occupancies NFPA 101 [1] Decree of 19th March 2015 [13]
Maximum horizontal travel distance to reach an 46 to 61 m 30 to 40 m
exit or an adjacent fire compartment
Minimum clear door width in the means of egress 81 cm 90 cm
Minimum clear and unobstructed width in the 112 cm 120 cm
means of egress from patients sleeping rooms
Minimum required space in In-patients 1.40 m2 or 2,8 m2 1.50 m2
the adjoining compartments Other occupants 0.56 m2 0.50 m2
for each occupant relocated
Minimum required space in In-patients 1.40 m2 or 2,8 m2 1.50 m2
the adjoining compartments Other occupants 0.56 m2 0.50 m2
for each occupant relocated
Occupant load factor in sleeping departments 11.1 m2/person 3 persons/in-patient bed
The horizontal assisted evacuation of a ward in a hospital floor combined with the vertical transfer
of one in-patient using a firefighters lift is simulated, implementing the occupant profiles with their
unrestricted walking speed and travel speed, the pre-travel times and preparation times, and the
movement groups described in the Section 2.
A twelve-stories building is here considered as an illustrative example of the proposed
methodology. The hospital building 3D model is shown in Figure 3. The application hospital floor
plan is the last (10th) floor having a rectangular shape of ~ 912 m2, with a central lift lobby and two
stairs block (S2 and S3). The floor is also served by two additional stairs (S1 and S5), remotely located
from each other. It accommodates one ward (W10) of ~ 310 m2 consisting of 12 patients sleeping
rooms (10 double occupancy, 2 single occupancy), 2 nurse stations and 3 service rooms. Each
habitable room has a unique exit access door, 85 to 105 cm wide, connected directly to the corridor
~ 60 m long and 330 cm wide (270 cm in the section leading to stairs S1). A meeting room, two
lobbies and a lounge are also located in the 10th floor.
Figure 3: Case study: hospital building 3D model

Figure 4 shows the 10th floor plan with the initial and final occupant positions in the safe refuges
(R1&R3 in stair S1, R2 in stair S5). The ward W10 is arranged as a compartment and the evacuation
is possible through two exit fire doors ~180 cm wide, leading respectively to the stairs S1 and S5.

Stair S2
Room #1 #12

#2 #11
Stair S3 E4
Refuge R2
Stair S5

#3 #4 #5 #6 #7 #8 #9 #10
R1 R3
Stair S1
Refuge
Refuge R2

Stair S5 10th Floor

Stair S1
Elevator E4

9th Floor
Refuge R1&R3

Emergency control
center @ 1st floor
(not shown)

Figure 4: Case study: model plan view with the initial and final positions of the occupants

The two stairwells S1 and S5 allow the building autonomous occupants to reach the exits located
on the first floor (level of exit discharge) while their landings are safe refuges to relocate on the same
floor a designed number of In-patients. One In-patient in Room #3 is accommodated in the lift landing
lobby at the nearest floor level (9th floor) using a firefighters lift.
 The Active staff in charge of the assisted horizontal evacuation is formed by the three nurses in
staff to Ward 10. Two Emergency responders, coming from the emergency control center located at
the 1st floor, have the primary task to vertically evacuate the In-patient located in Room #3 and then
support the Active staff operators in relocating the other In-patients, if help is needed.
The number of nurses is determined according to the nurse to In-patients ratio, prescribing how
many In-patients each nurse is responsible for during a shift. This ratio is lower for critical care
facilities, as patients require more constant monitoring, and is higher long-term care facilities. In
Italy, suggested nurse staffing ratio are regulated by law [89]; it is 1 to 6 to 1 to 8 for long-term care
facilities and ordinary wards. In this study it is assumed to be ~ 1 to 7, so that the 22 In-Patients in
ward W10 are assisted by 3 nurses. In the scenario examined, it is assumed that 12 Visitors to in-
Patients are present in Ward 10 (range: 0 or 1 or 2 Visitors for one in-Patient).
In principle the adequacy of the health care occupancy emergency procedures and means of
escape should be demonstrated based on the time of day or night when the evacuation would be most
difficult, usually in the night shift when the in-patients are sleeping and fewer staff are present.
Given the objective of this study, it is more interesting to examine day-time visiting hours where the
maximum variety of the occupant profiles is observed. The scenario is based on an emergency that
occurs in ward W10 on the 10th floor and the safety planning rules the relocation of 21 In-patients in
the safe refuges provided in the stair landings S1 and S5. One bedded In-patient in Room #3 is
transferred to the nearest floor (9th floor) using the firefighters lift E4. Ten Visitors remain with the
in-Patients, while two Visitors are allowed to self-evacuate from the building, together with two
Workers present inside the ward W10. Full seating occupancy is assumed in the meeting room (18
occupants), two lobbies (15 occupants each) and a lounge (40 occupants).
The designer should also establish the number of groups initially present that break when the
alarm is given. In the scenario investigated, 2 Visitors abandoned the in-Patient they were visiting
and self-evacuate. This behaviour has been specified noting that:
- it is usually prescribed in emergency plans that the autonomous occupants not directly involved
in firefighting or in evacuation management move to the nearest exit;
- the emergency condition may break the social link with the In-Patient
To summarize, the assisted evacuation concerns 127 occupants in the 10th floor (22 in-Patients; 3
Active staff; 12 Visitors to In-patients; 2 Workers; 88 Autonomous occupants):
- 92 autonomous occupants (2 Workers, 2 Visitors not linked to In-Patients, 88 Autonomous
occupants) are instructed to reach an exit at the level of exit discharge (1st floor);
- 21 In-Patients and 10 linked Visitors are relocated in the safe refuges in stair landings S1 and S5,
using a variety of mobility devices;
- 1 In-Patient is relocated in the 9th floor;
- 3 Active staff operators are initially in the nurse stations.
The evacuation is conducted with the support of 2 Emergency responders, coming from the
emergency control center located in the 1st floor.
To summarize, the sequence simulated corresponds to the following evacuation scheme:
1. Start time is set to the order to relocate all the in-Patients in ward (W10) transferring 21 of them
to the safe refuges located in the same floor (stair S1 and S5 landings) and 1 in the 9th floor using
the firefighter lift, according to the relocation plan.
2. Active staff operators in charge of the evacuation process are initially inside the ward, in the nurse
stations, and collaborate jointly forming a first set of evacuation team.
3. Emergency response operators are initially located in the emergency control center at 1st floor
and move to the ward W10, using the firefighter lift, with the task to evacuate the In-patient in
Room #3 to 9th floor and then to support the Active staff in the relocation activities.
4. The other autonomous occupants react according to the assigned behavioural instructions,
directing to an exit or remaining with an in-Patient, starting an individual or a group movement.
5. After assisting all the in-Patients, the Active staff and the Emergency response operators finally
move to the nearest safe refuge in the 10th floor to remain there with the relocated occupants.
The design occupant profiles, initial positions and behaviours are reported in Table 8. Ten movement
group schemes are implemented. For each In-patient relocated, it is necessary to specify if assistance
is required and the details (number and skills of assisting operators, use of a mobility device).
Table 8: Design occupant initial positions, profiles and behaviours.
Location Groups Occupant profile Behaviour
10th floor Ward W10 (to be evacuated)
Room #1: Group 01 1 In-patient Autonomous with  Go To the specified Refuge area #02 in stair S5
In-patient manual wheelchair and 2 Visitor with initial delay: PTAT (random)
sleeping to in-patient (duplicate profile for
room assisted group movement)
1 In-patient Assisted Ambulant  Wait for assistance of Active staff team
[1 assistant]  Wait the preparation time (random)
 Go To the specified Refuge area #02 in stair S5
Room #2: 1 In-patient Autonomous  Go To the specified Refuge area #02 in stair S5
In-patient ambulant with crutches with initial delay: PTAT (random)
sleeping 1 In-patient Assisted Ambulant  Wait for assistance of Active staff team
room  Wait the preparation time (random)
 Go To the specified Refuge area #02 in stair S5
Nurse 2 Staff operators assigned to the  Assist Active Staff team with an initial delay equal
station #1 Evacuation Active Staff team to the PTAT (random)
 Go To any Refuge areas
Room #3: 2 Visitors to in-patient  Go To any Exit (at 1st floor or ground level) with
In-patient initial delay: PTAT (random)
sleeping 1 In-patient Assisted Vertical Evac  Wait for assistance of Emergency response team
room bed  Wait the preparation time (random)
 Go To Elevator E4 [firefighter lift] target: 9th floor
 Go To the specified Refuge in 9th floor
Room #4: 1 In-patient Assisted Evac chair  Wait for assistance of Active staff team
In-patient  Wait the preparation time (random)
sleeping  Go To specified Refuge area #02 in stair S5
room 1 Visitor to In-patient  Go To specified Refuge area #02 in stair S5 with
initial delay: PTAT (random)
Room #5: Group 02 1 In-patient Assisted Ambulant  Wait for assistance of Active staff team
In-patient [1 assistant] and 1 Visitor to In-  Wait the preparation time (random)
sleeping patient (duplicate profile for  Go To the specified Refuge area #02 in stair S5
room assisted group movement)
1 in-Patient Assisted Wheelchair  Wait for assistance of Active staff team
 Wait the preparation time (random)
 Go To specified Refuge area #02 in stair S5
Room #6: Group 03 1 In-patient Autonomous  Go To the specified Refuge area #02 in stair S5
In-patient Ambulant with rollator or walking with initial delay: PTAT (random)
sleeping frame and 1 Visitor to In-patient
room (duplicate profile for group
movement)
1 In-patient Autonomous  Initial delay: PTAT (random)
Ambulant with walking stick  Go To the specified Refuge area #02 in stair S5
Room #7: Group 04 1 In-patient Autonomous  Initial delay: PTAT (random)
In-patient Ambulant with rollator or walking  Go To the specified Refuge area #01 in stair S1
sleeping frame and 1 Visitor to In-patient
room (duplicate profile for group
movement)
1 In-patient Assisted Evac chair  Wait for assistance of Active staff team
 Wait the preparation time (random)
 Go To specified Refuge area #03 in stair S1
Table 8: Design occupant initial positions, profiles, and behaviours (continued).
Location Groups Occupant profile Behaviour
10th floor Ward W10 (to be evacuated)
Room #8: 1 In-patient Autonomous  Initial delay: PTAT (random)
in-patient  Go To the specified Refuge area #01 in stair S1
sleeping Group 05 1 In-patient Assisted Ambulant  Wait for assistance of the Active staff team
room [2 assistants] and 1 Visitor to In-  Wait the preparation time (random)
patient (duplicate profile for group  Go To the specified Refuge area #01 in stair S1
movement)
Room #9: Group 06 1 in-Patient Assisted Evac chair  Wait for assistance of the Active staff team
in-patient and 1 Visitor to In-patient  Wait the preparation time (random)
sleeping (duplicate profile for assisted  Go To the specified Refuge area #03 in stair S1
room group movement)
1 In-patient Autonomous  Initial delay: PTAT (random)
 Go To the specified Refuge area #01 in stair S1
Room #10: 1 In-patient Assisted Evac chair  Wait for assistance of Active staff team
in-patient  Wait the preparation time (random)
sleeping  Go To specified Refuge area #03 in stair S1
room 1 In-patient Autonomous  Initial delay: PTAT (random)
 Go To the specified Refuge area #01 in stair S1
Nurse 1 Staff operator assigned to the Same as Active staff in Nurse station #1
station #2 Evacuation Active Staff team
Room #11: Group 07 1 in-Patient Assisted Evac chair  Wait for assistance of Active staff team
in-patient and 2 Visitors to In-patient  Wait the preparation time (random)
sleeping (duplicate profile for assisted  Go To specified Refuge area #03 in stair S1
room group movement)
1 In-patient Assisted Ambulant  Wait for assistance of the Active staff team
[1 assistant]  Wait the preparation time (random)
 Go To the specified Refuge area #01 in stair S1
Room #12: 1 In-patient Autonomous with  Go To the specified Refuge area #01 in stair S1
in-patient electric wheelchair with initial delay: PTAT (random)
sleeping 1 In-patient Autonomous  Go To the specified Refuge area #01 in stair S1
room with initial delay: PTAT (random)
Service Group 08 2 Workers  Go To any Exit (at 1st floor or ground level) with
room #1 initial delay: PTAT (random)
Meeting 18 Autonomous occupants Same as Group 08
room
Lobby east Group 09 3 Autonomous occupants Same as Group 08
12 Autonomous occupants Same as Group 08
Lobby west Group 10 3 Autonomous occupants Same as Group 08
12 Autonomous occupants Same as Group 08
1st floor: 2 occupants (2 Emergency response operators)
Emergency 2 Emergency response operators  Go To Elevator E4 [firefighter lift] target: 10th
control assigned to the Evacuation floor with initial delay equal to the PTAT
center Emergency team (random)
 Assist Emergency team
 Change profile to Active staff
 Assist Active Staff team
 Go To Go To any Refuge areas

The total number of assisted In-patients is 10. In this study, a priority list is fixed in the scheduling
policy of the Active staff team: 1) In-patient in Room #11; 2) In-patient in Room #9; 3) In-patient in
Room #7. Pathfinder assigns the task to the required number of assisting operators, having the necessary
skills and being available, who are located at the minimum distance from the assisted occupant. If more
than one assisting operator is needed, the assisted occupant shall wait for the whole team to
assemble. Assisted ambulant In-patients or those transferred on wheelchair or evac chair or rescue
sheet can be serviced only by an Active staff team. The bedridden In-patient in Room #3 to be
transferred using the firefighters lift is assisted only by the Emergency response operators,
representing their first action to execute. All these specifications define the design service discipline.
Different assisting team formation and scheduling policies could be adopted, nevertheless the rules
here described allows a wide range of evacuation scenarios to be assessed.
The evacuation capabilities and pre-evacuation and preparation times distributions are selected
from Tables 3 to 6, while the mobility devices dimensions assumed in this study are given in Table 9,
based on manufacturer technical specifications and literature data.

Table 9: Design mobility device dimensions.

Type Length Width
Hospital bed 1 220 cm 100 cm
Wheelchair 2 95 cm 75 cm
Walking frame/rollator 3 50 cm 57 cm
Carry or evacuation chair 4 77 cm 52 cm
Hand-held stretcher 4 200 cm 45 cm
Hand-held rescue sheet 4 200 cm 75 m
1 Based on catalogue data (“Karismedica” hospital bed)
2 Based on catalogue data (“Althea” wheelchair)
3 Based on catalogue data (“Gibermedicali” DM733 walking frame)
4 Based on Hunt et al. data [29-30].

The model has been implemented in Pathfinder [91], version 2024.1.0813. Since the 2018 release
[92], Pathfinder provides a support for assisted evacuation with mobility devices, group movements
and refuges (in the “steering mode”). In Pathfinder assistance can be called only by agents with
mobility impairments (“clients”). This constraint does not allow a direct codification of the assisted
ambulant profile, overcome simulating a “virtual” mobility device, with a polygonal shape resembling
a person, and prescribing the number of assistants and their relative position during the service. A
“client” instance is activated by the “Wait for Assistance” behaviour action, eventually with the
request of a particular set of emergency team, while the availability to act as a member of a particular
set of emergency team is activated by the “Assist” behaviour action. Mobility impaired occupants that
do not have autonomous movement capability remain in the position where they are left by the
assisting team. If the relocation is in a safe refuge, in Pathfinder the impaired person can unduly
impede the entry of other occupants or limit the space availability especially if a mobility device, such
as a bed or rescue sheet, is required for the transfer. To tackle this issue, specific areas of refuge for
assisted non ambulant profiles should be defined, providing at least two virtual doors so that the
assisted non ambulant occupant can be allocated properly and the assisting operators can move out
without remaining unduly entrapped.
The assisting tasks are the first to be executed. This means that at the start of the simulation all
the occupants who need assistance (“clients”) are “rescued” by all the available assisting occupants,
having the necessary skills, with an initial time delay modelling their PTAT. If an occupant needs
assistance, collective movement will not start until the assistance team has assembled, and after the
delay caused by the preparation phase. The preparation phase duration depends on the assisted
occupant’s needs and capabilities. This kind of group will not break until the service is completed
leaving the assisted occupant in the safe destination area; hence the requirement in Pathfinder that
all the group members (excluding the assistants) must share the same behaviour.
To implement a movement group scheme which includes an assisted occupant with other
autonomous occupants, it is necessary in Pathfinder to “duplicate” the autonomous profiles in their
corresponding assisted ambulant profiles, changing the shape attribute, selecting a polygonal form,
and defining a mobility vehicle resembling the human body (i.e. an octagon), with no attached
assistant.
 When a group is formed, its movement is mainly controlled in Pathfinder by two concepts:
connected state and the option to choose a group leader to be selected from a specific profile. If a
group is in a "disconnected" state (e.g., the mutual distance among group members exceeds a
prescribed maximum value), occupants with autonomous profiles will walk toward the leader. A
group in a "connected" state will move toward the goal dictated by its behaviour and eventually
slowdown along the path if they accidentally get “disconnected”. In Pathfinder there is no facility to
modify the group constitution during the execution of a task.
The basic groups schemes implemented in the simulation model are described in Table 10. Each
assisted In-patient transported on a mobility device or assisted ambulant In-patient, by definition
requires individual assistance. Thus, a group can include only one assisted occupant with one or more
autonomous occupants – the Visitors to in-patients - sharing a social link.

Table 10: Basic movement groups schemes adopted in the simulation of hospital ward evacuation.
a) Movement groups for occupants having autonomous evacuation capabilities
2 or more Visitors to In-patients (or generic autonomous occupants)
2 or more Workers (not in charge of egress assistance)
1 Autonomous In-patient and 1 or more Visitors to In-patient
1 Autonomous but mobility impaired (5 categories) and 1 or more generic autonomous occupants
b) Movement groups for assisted occupants 1
1 Assisted ambulant and 1 or more Visitors to in-patients (or generic autonomous occupants)
1 Assisted transported on a wheelchair or evac chair and 1 or more Visitors to in-patients
1 Assisted transported with hand-held rescue sheet and 1 or more Visitors to in-patients
1 Assisted transported with hand-held stretcher and 1 or more Visitors to in-patients
1 Assisted transported on a bed and 1 or more Visitors to in-patients
1 Each group will include in addition the prescribed number and skilled assisting operators.

During the simulation Pathfinder adapt the speeds reported in Tables 3 and 4 depending on the
density around the occupant, and, when group movement is considered, also taking into account the
constraints imposed on the mutual distance among the group members. When a group is constituted,
it moves mainly at the speed of its slowest member. The unrestricted walking speed for the Active
staff/Emergency response personnel is therefore used only when the agent is travelling toward an
assisted or has completed all the assigned service tasks. In order to allow the Emergency responders
to collaborate with the Active staff, in Pathfinder their Behavior shall include the action “Change
Profile”, after completing the vertical evacuation of the In-patient in Room #3.
The scenario here considered, horizontal assisted evacuation with the aid of mobility devices,
including group movements for both assisted and autonomous occupant profiles, combined with the
vertical evacuation using the firefighters lift, is very challenging for the designer whatever is the
simulation model adopted.

3.2 Monte Carlo simulation results

Using the probability distribution functions list in Tables 3 to 6, initial random realizations of all
input stochastic variables (pre-evacuation time, preparation time, unrestricted speed for both
autonomous and assisted profiles) are generated. A deterministic calculation is then performed,
applying the design behavioural rules specified in Table 8 and the design service discipline, to return
a single model output. The outcome of interest is here represented by the time required to relocate
1 In-patient to the 9th floor and 21 In-patients in ward W10 in the safe refuge areas in landing stairs
S1 and S5 on the same 10th floor, whose maximum value being the evacuation time (ET).
In order to evaluate the range and the probability distribution functions of ET and of the relocation
times, the process is repeated. 142 MC trials were run using Pathfinder to create a dataset, which is
analysed in Section 4 to discuss the impact of the number of trials on model predictions. Given the
complexity of the scenario investigated, each single run used has been checked to verify that all the
agents and the mobility devices act as expected during their travelling paths and do not remain
unduly idle or blocked. For the scenario considered, the ET frequency function and its cumulative
distribution function obtained are reported in Figure 5. The ET values are in the range 499-805 s,
classified as “slow” according to the Life Safety Code © classification [1].
at or below a specified evacuation time

Fraction of simulation results

Fraction of simulation results

within a specified range

Evacuation Time (ET) (s) Evacuation Time (ET) (s)

Figure 5: Evacuation time (ET). Statistics based on 142 Monte Carlo simulations
Calculated data have been statistically treated to obtain the histogram plots of the time-series;
equally spaced bins (e.g., time interval) are selected to group data, except where a gap clearly
separates the first (autonomous) and last (assisted) occupant arrival times. Appropriated bin widths
are in the range 5 to 40 s, based on the Rice and the Freedman-Diaconis’ rules [93]. The distributions
of the time of arrival of the first and of the last occupant (being the relocation time) are shown in
Figure 6 for the safe refuge areas #01 &#03 in stair landing S1 (RT1&RT3) and #02 in S5 (RT2), and
the relocation of the bedridden In-patient in Room #3 moved using the firefighters lift in the 9th floor.
The distributions of the time of arrival of the first and of the last autonomous occupant (initially
in the two lobbies, in the lounge and the meeting room) to reach the exit at the discharge level (1st
and ground floor) is illustrated in Figure 7.
In an assisted evacuation scenario, the key parameters are the number and skills of the assisting
staff and the PTAT and preparation times [25, 27-28, 35]. The calculated ET and RT probability
distribution functions are determined by the prescriptive rules concerning the service discipline and
by the stochastic variables determining the pre-travel activity times and the preparation times, and
the unrestricted walking speeds and assisted travel speeds. When examining the simulation results,
there are differences both in the order the assisted occupants are served and in the composition of
the assisting teams serving each assisted occupant. Statistics are given in the following Table 11 (ET)
and Table 12 (RT1, RT2).
When both assisted and autonomous occupants share a common destination, the autonomous
occupant profiles, which are not involved in movement group scheme involving an assisted occupant,
are the first to arrive and therefore their design characteristics (PTAT and unrestricted walking
speed) determine the first part of the usage of the target destination. The flow of the assisted
occupants then follows, or is sometimes partially overlapped, depending mainly on the service
discipline and the mobility device eventually required for the transferral. Besides the unrestricted
speed and the travel path, the queuing for assistance and preparation times of assisted occupants
governs the usage of the safe refuge areas.
Fraction of simulation results

First occupant First occupant

within a specified range

Last occupant Last occupant

Time of arrival of the first/last occupant (RT2) (s) Time of arrival of the first/last occupant (RT1) (s)
a) safe refuge area #02 in stair S5 (4 autonomous In- b) safe refuge area #01 in stair S1 (6 autonomous In-
patients, 3 Assisted ambulant In-patients, 2 Assisted In- patients, 2 Assisted ambulant In-patients and 2 Visitors
patients and 5 Visitors to In-patients) In-patients)
Fraction of simulation results

First occupant
within a specified range

Last occupant

Time of arrival of the first/last occupant (RT3) (s) Time to relocate the In-patient in Room #3 (RT4) (s)
c) safe refuge area #03 in stair S1 (4 assisted In-patients d) safe refuge area in 9th floor (1 assisted In-patient)
and 3 Visitors to In-patients)
Figure 6: Time of arrival of first/last occupant: a) safe refuge area #02 in stair S5 (4 autonomous In-
patients, 3 Assisted ambulant In-patients, 2 Assisted In-patients and 5 visitors to In-patients);
b) safe refuge area #01 in stair S1 (6 autonomous In-patients, 3 Assisted ambulant In-patients,
3 Assisted ambulant In-patients and 2 visitors to In-patients); c) safe refuge area #03 in stair
S1 (4 Assisted ambulant In-patients and 3 visitors to In-patients); d) 1 assisted In-patients
vertically evacuated. Statistics based on 142 Monte Carlo simulations.
Fraction of simulation results

First occupant
within a specified range

Last occupant

Figure 7: Time of arrival of first/last autonomous occupant to reach the exit (92 occupants).
Statistics based on 142 Monte Carlo simulations
 4. Discussion
The results of the 142 MC simulations of the assisted evacuation scenario presented in Section 3.2
(Figures 5 to 7) are here analysed applying the principles introduced in Section 2. It is remarked that
the stochastics variables are described by probability distributions derived from the statistics given
in Tables 3 to 6 (not fictitious ones), including both autonomous and assisted profiles.

4.1 Descriptive statistics

Aggregating the output stochastic variables of interest (the evacuation time: ET; relocation times:
RT1, RT2, RT3, RT4) into groups by size and displaying the values as a histogram provides the
approximate shape of the probability density function. The output values can themselves be used as
an empirical distribution, thereby calculating the percentiles and other statistics. These statistics can
then be used for developing confidence bands, as discussed in section 2.5.1; the precision of the
expected value of the variable of interest and the distribution shape approximations improve as the
number of simulation trials increases as is clearly shown in Figure 8 and Table 11, for ET, Figure 9
and Figure 10 and Table 12, for RT1 and RT2.

50 trials 100 trials 142 trials

Frequency

ET (s)
Figure 8: Evacuation time (ET) histograms evolution as a function of the number of Monte Carlo trials

Most results presented in Section 2.5.1 are valid regardless of the shape of the underlying distribution
provided that the sample is large enough; but the normality (or even the symmetry in the distribution)
make the inferences more robust even in case of small samples [73]. The method can be generalised
adopting Bonett [75] confidence intervals with a larger sample size.

Table 11: ET descriptive statistics evolution as a function of the number of Monte Carlo trials
ET statistics 25 50 75 100 125 142

"! (s)
trials trials trials trials trials trials

'() (s)
Mean 648 659 656 651 649 646
,-.
Standard deviation 58 55 58 61 58 57
√
Standard error (s) 11,5 7,7 6,6 6,1 5,2 4,8
95% CI for the mean Δ (s) 23,7 15,1 13,0 11,9 10,2 9,4

!012 (s)
Median (s) 637 651 649 641 638 637

!0$ (s)
Minimum 575 575 515 499 499 499

( !012 - !0$ ) (s)

Maximum 805 805 805 805 805 805
Range 230 230 290 306 306 306
IQR [Q3-Q1] (s) 50 70 60 68 66 63
Kurtosis 2,2 0,2 0,3 0,2 0,2 0,3
Skewness 1,4 0,8 0,7 0,5 0,6 0,6
 50 trials 100 trials 142 trials
Frequency

RT1 (s)
Figure 9: Relocation time in safe refuge area #01 in stair S1 (RT1) histograms evolution as a function of
the number of Monte Carlo trials

50 trials 100 trials 142 trials

Frequency

RT2 (s)
Figure 10: Relocation time in safe refuge area #02 in stair S5 (RT2) histograms evolution as a function
of the number of Monte Carlo trials

Table 12: Relocation times RT1 and RT2: descriptive statistics evolution as a function of the number of
Monte Carlo trials
RTs statistics Relocation time in refuge area Relocation time in refuge area
#01 in stair landing S1 (RT1) #02 in stair landing S5 (RT2)
25 50 100 142 25 50 100 142

" (s)
b!
trials trials trials trials trials trials trials trials

'c) (s)
Mean 631 637 632 630 584 604 588 583
,d.
Standard deviation 63 54 55 52 57 69 76 73
√
Standard error (s) 12,6 7,7 5,5 4,4 11,5 9,8 7,6 6,1
95% CI for the mean Δ e (s) 26,1 15,0 10,8 8,6 23,7 19,2 14,8 12,0

b!0$ (s)
Median (s) 622 628 624 624 563 600 575 569

b!012 (s)
Minimum 535 535 499 499 494 489 417 417

(b!012 - b!0$ ) (s)

Maximum 805 805 805 805 711 751 788 788
Range 270 270 306 306 217 262 371 371
IQR [Q3-Q1] (s) 122 66 66 64 140 97 95 89
Kurtosis 2,4 1,8 1,1 1,1 -0,6 -0,6 0,2 0,3
Skewness 1,5 1,1 0,8 0,8 0,6 0,5 0,7 0,7
A normal probability plot (NPP) or a quantile-quantile plot (QQP) should be constructed for each
random output variable of interest. Deviation from a straight line indicates that the population
distribution is not normal. If the NPP or QQP indicates normality, one of the statistical tests for
normality can then be performed to quantify the confidence level of a normality assumption. The
Shapiro-Wilk test and the D’Agostino-Pearson test have been selected due to their best global
performance compared to other normality tests [94]. Both tests confirmed that the normal
distribution model does not fit the ET observations. However, the magnitude of the difference
between the sample distribution and the normal distribution is small, as shown in Figure 13.

Figure 13: Evacuation time (ET) QQ-Plot for a sample of 142 Monte Carlo trials

4.2. Convergence analysis and accuracy

The application of the methodology presented in 2.5.1 to estimate and update the number of trials

are: Δfgh ijklmn = 10 s, α = 0.05, 0$ = 40, N=nmax=142. The evolution of the minimum, maximum, average
that is needed to achieve a certain level of precision is applied to the worked case. The design basis

and standard deviation of the ET distribution as a function of the MC number of trials is illustrated in
Figure 14, while Table 13 reports the evolution of the half-width of the confidence interval and the
inference on the total number of simulations required to obtain the design accuracy.

Max
Evacuation time (ET) (s)

Min
Standard deviation (s)

Figure 14: Evacuation time (ET) average ( "!) and standard deviation ('() ) evolution
Number of simulations
 (design basis: o T-U>V< =10 s, α = 0.05)
Table 13 Evacuation time (ET) convergence as a function of the Monte Carlo number of trials

" (s)
ET n1 =10 n2 =40 n3 =70 n4=100 n5=130 nmax=142
et
'() (s)
640 649 656 652 649 647
42 55 55 59 57 56

Δ
9,7 9.2
(< Δf
(s) 29.9 20.8 13.0 11,6
T-U>V<
)

89 114 117 137 125 123

Normal distribution No

After conducting 10 runs (n1 = 10), the first estimates of the mean ( "!&p = 640 s) and of the standard
&p
deviation ('()
mean ( Δ =29.9 s) and the first estimate of the total required number of trials ( 6 =89) are
10
=42 s) of the ET distribution are obtained. Using the t distribution values, the CI for the

simulations (n2 = 40 trials), the sample mean ( "! Mp =649 s) and standard deviation ('() Mp
calculated, thus completing the “prediction” step. After performing an additional round of 30
=55 s) of ET

statistics, the updated accuracy is Δ = 20.8 s and the corrected estimate of the number of trials is
40
are recalculated and its descriptive statistics is obtained and tested for normality. Using the z

now Mp =114. This is the first “correction” step. As the convergence termination condition is not

trials, the termination criteria is verified as expected: Δ130= 9.7 s <Δfgh ijklmn and n4 > &sp =123. The
verified, additional rounds of 30 runs are executed repeating the process. When a sample of n5 = 130

analysis is extended to 142 runs to confirm the convergence. The ET distribution is not normally
distributed, even if the magnitude of the difference is small, reflecting its dependency on the service
discipline and the preparation time distributions. In this case the statistical inference provides the
designer a rough estimate of the accuracy of the ET achieved when the proposed termination
criterium is reached.
The same convergence and accuracy analysis has been conducted for the relocation times. Details
are omitted in this paper.
If the design basis specifies the same accuracy (e.g., and ) for all the stochastic variables of
interest (in our case ET, RT1&RT3, RT2 and RT4), the one having the higher standard deviation,
whatever is the mean value, is the controlling parameter in determining the number of MC simulations
required. Stochastic convergence of the evacuation time, which is based on the last service
completion or the last autonomous occupant exit, does not guarantee the convergence of the
relocation times stochastic distributions within the same design accuracy (e.g., and ), even if these
tasks are completed at an earlier time.
Noting that the sample evacuation time distribution has only a slight deviation from the normal
distribution, using the results for 130 runs reported in Table 13 and the calculated ()=@A being equal
to 64.45, we obtain the inclusive evacuation time applying Eq. 13:
" +Δ
iETth = (et ) + 2.33 ()=@A = (649+9.7) + 2.33*64.45 = 809.1 s (14)

in line with the maximum sample value of 805 s, observed after having performed 142 trials.

5. Conclusions
The inclusion of mobility, sensory or cognitive impairments in evacuation modelling in a general
inclusive framework is proposed to establish a standard codification of the occupant profiles.
Mobility is combined with way-finding ability to obtain a basic set of five categories that still retains
the potential to describe the performance characteristics of building users and is suitable to be
implemented in agent-based computer evacuation models developed over recent years. Apart from
occupant characteristics, it is necessary to define the service discipline, consisting of three
components: the staff skills and consistency, the scheduling policy, and the mobility device eventually
required to relocate the assisted occupant. The principles of performance-based inclusive design are
discussed: a generalization of the conventional RSET is proposed introducing the Required Safe and
inclusive Escape Time (RiSET) criterion. Noting that the conventional PTAT definition is clearly
applicable only to the autonomous occupants capable of self-evacuation, it is proposed to introduce
a different term for occupants requiring assistance. PTAT should be replaced by the PAL time,
combining the Preparation time and the time and skills required to the Assisting staff to establish a
communication Link, verbal or a visual (e.g., sign language), with an occupant having cognitive or
sensory impairments. While the RSET calculation is traditionally based on deterministic methods,
RiSET requires a probabilistic risk analysis.
A general procedure based on Monte Carlo methods and the Central Limit Theorem is presented
and the convergence scheme and criteria discussed. The number of simulations required for a
specified accuracy on the ET distribution is established using a predictor-corrector scheme. A
dynamic assessment of the behaviour of the ET statistic is recommended to optimize the number of
simulations, provided that a suitable convergence methodology is adopted, monitoring the evolution
of the sample standard deviation and testing for sample normality. The 99th percentile evacuation
time prediction may be selected as a key parameter to calculate one RiSET value to be compared with
the ASET value corresponding to minimum time in which one occupant could be incapacitated.
The predictive capabilities of the model are applied to the scenario of assisted horizontal
evacuation from a hospital ward combined with the vertical transfer of one in-patient using a
firefighters lift. Many of the limitations noted in previous studies (Ursetta et al. [34]], Alonso and
Ronchi [28]) concerning the use of mobility device and the preparation time for in-patients are
overcome. Some difficulties still remain in Pathfinder and require ad hoc adaption and checks [92]:
1) assistance can be called only by agents with mobility impairments; 2) mobility impaired occupants
remain in the position where they are left by the assisting team and can unduly impede the entry of
other occupants or limit the space availability especially if a bed or rescue sheet is required for the
transfer; 3) movement group schemes which include an assisted occupant with other autonomous
occupants require to “duplicate” the autonomous profiles involved, defining assisted ambulant
profiles with a vehicle shape resembling the human body and no attached assistant; 4) mobility
devices like beds or rescue sheets remain sometimes unduly idle or blocked along their travelling
path. The proposed model includes all the stochastics variables considered in performance-based
design: occupant profiles with their unrestricted walking speed and travel speed, the pre-travel times
and preparation times, and the movement groups. Thus, it has sufficient flexibility to be calibrated
with site specific data and has the potentiality to be used in emergency planning of assisted
evacuation verifying the design service discipline. The results of the model need to be compared to
data from actual evacuations as part of a validation exercise before concluding that it gives results
that can be used for evacuation planning.
Conflicts of interest
The authors declare no conflict of interest associated with the manuscript.
Acknowledgements
The author expresses his gratitude to Thunderhead Engineering for the opportunity to use the
software PATHFINDER and acknowledges Fabio Alaimo Ponziani and Valter Ricci, who co-authored
the study presented in FEMTC 2018, and the Italian Department of firefighters, of the public rescue
and civil defence, Ministry of the Interior, for supporting the research.
References
[1] National Fire Protection Association, Life Safety Code ©, NFPA 101, Quincy, MA: National Fire
Protection Association, 2021 Edition.
[2] British Standards Institution, Ed., Fire safety in the design, management and use of buildings. Code
of practice, BS 9999. London: BSI, 2017.
[3] German Institute for Standardization, Ed., Fire safety engineering - Part 1: Basic principles and
codes of practice, DIN 18009-1. Berlin: DIN, 2016.
[4] Italy, Decree of the Minister of the Interior of 18th October 2019, Amendments to Annex 1 to the
Decree of the Minister of the Interior of 3rd August 2015 “Approval of technical specifications for
fire prevention, pursuant to article 15 of legislative decree of 8th March 2006 n. 139”.
[5] Great Britain, The Building Regulations. Approved document B, Fire Safety, volume 2: Buildings
other than dwellings. Newcastle-Upon-Tyne: NBS, 2019.
[6] British Standards Institution, Ed., Application of fire safety engineering principles to the design of
buildings – Code of practice, BS 7974. London: BSI, 2019.
[7] A. Rahouti, S. Datoussaid, R. Lovreglio, A sensitivity analysis of a hospital evacuation in case of
fire. The impact of the percentage of people with reduced mobility and the staff to occupant’s
ratio, presented at the Fire and Evacuation Modeling Technical Conference, Malaga, Spain, 2016.
[8] T. Yamamura, H. Murai, H. Kurioka, K. Kobayashi, H. Satoh, K. Ohnishi, A study on the evaluation
method of fire drill on nursing homes for the elderly, J. Environ. Eng. (Transactions of AIJ), 2020,
85 (77) (2020) 299-309.
[9] A.K. Childers, K.M. Taaffe, Healthcare facility evacuations: lessons learned, research activity, and
the need for engineering contributions, J. Healthc. Eng. 1 (2010) 125-140.
[10] K.E. Boyce, T.J. Shields, G.W.H. Silcock, Toward the characterization of building occupancies for
fire safety engineering: capability of people with disabilities to read and locate exit signs, Fire
Technol. 35 (1) (1999) 79-86.
[11] International Code Council/American National Standard Institute, Accessible and Usable
Buildings and Facilities, ICC/ANSI A117.1 International Code Council, Inc., 2017.
[12] Italy, Decree of the Minister of the Public Works of 14th June 1989 n. 236, Technical specifications
necessary to guarantee the accessibility, adaptability and usability of private and public
buildings in order to overcame and eliminate architectonical barriers.
[13] Italy, Decree of the Minister of the Interior of 19th March 2015, Update of the fire prevention
technical specifications dated 18 September 2002 for the design, construction and operation of
public and private health care occupancies.
[14] Italy, Decree of the Minister of the Interior of 29th March 2021, Approval of the fire prevention
technical specifications for health care occupancies.
[15] E.D. Kuligowski, R.D. Peacock, B.L. Hoskins, A Review of Building Evacuation Models”, National
Institute of Standards and Technology, 2010. Technical Note 1680.
[16] E.D. Kuligowski, Computer evacuation models for buildings, in: M.J. Hurley, D.T. Gottuk, J.R. Hall,
K. Harada, E.D. Kuligowski, M. Puchovsky, J.L. Torero, J.M. Watts, C.J. Wieczorek (Eds.), SFPE
Handbook of Fire Protection Engineering, Springer New York, New York, NY, 2016, pp. 2152–
2180.
[17] H. Vermuyten, J. Beliën, L. De Boeck, G. Reniers, T. Wauters, (2016), A review of optimisation
models for pedestrian evacuation and design problems, Saf. Sci. 87 (2016) 167-178.
[17] E. Ronchi, E.D. Kuligowski, P.A. Reneke, R.D. Peacock, D. Nilsson, The Process of Verification and
Validation of Building Fire Evacuation Models, National Institute of Standards and Technology,
2013. Technical Note 1822.
[18] E. Ronchi, E.D. Kuligowski, D. Nilsson, R.D. Peacock, P.A. Reneke, Assessing the verification and
validation of building fire evacuation models, Fire Technol. 52 (2014) 197–219.
[19] E. Ronchi, A. Corbetta, E.R. Galea, M. Kinateder, E. Kuligowski, D. McGrathf et al., New approaches
to evacuation modelling for fire safety engineering applications, Fire Saf. J. 106 (2019) 197-209.
[20] E. Ronchi, Developing and validating evacuation models for fire safety engineering, Fire Saf. J.
120 (2021) 103020, Proceedings of the 13th International Symposium on Fire Safety Science.
[21] S. Tsuchiya, Y. Hasemi, Y. Furukawa, Evacuation characteristics of group with wheelchair users,
in: Proceedings of the 7th Asia-Oceania Fire Science & Technology Symposium, 2007.
[22] S. Tsuchiya, Y. Hasemi, Characterization of individual aided-evacuation in hospital building, J.
Architect. Plan. Architect. Ins. Japan 613 (2007) 1-6.
[23] S. Tsuchiya, H. Yuji, Characterization of group aided-evacuation in hospital ward, J. Architect.
Plan. Architect. Ins. Japan 640 (2009) 1271-1276.
[24] H. De-Ching, C. Shen-Wen, L. Chien-Hung, H. Po-Ta, S. Yi-Ting, S. Huei-Ru, A study for the
evacuation of hospital on fire during construction, Procedia Eng. 11 (2011) 139-146.
[25] D. Golmohammadi, D. Shimshak, Estimation of the evacuation time in an emergency situation in
hospitals, Comp. Ind. Eng. 61 (4) (2011) 1256–1267.
[26] J. Ze-min, Z. Pei-hong, S. Rong-xue, T. Xiang-liang, Investigation and simulation on human
evacuation behaviour in large hospital building in Shenyang, Procedia Eng. 71 (2014) 101-106.
[27] V. Alonso, Egress modelling in health care occupancies, The Fire Protection Research
Foundation, Quincy, MA (2014).
[28] V. Alonso-Gutierrez, E. Ronchi, The Simulation of Assisted Evacuation in Hospitals, presented at
the Fire and Evacuation Modelling Technical Conference, Malaga, Spain, 2016.
[29] A. Hunt, E.R. Galea, P.J. Lawrence, An analysis of the performance of trained staff using movement
assist devices to evacuate the non-ambulant, in: Proceedings of the fifith International
Symposium on Human Behaviour in Fire, 2012, pp. 328-339.
[30] A. Hunt, E.R. Galea, P.J. Lawrence, An analysis and numerical simulation of the performance of
trained hospital staff using movement assist devices to evacuate people with reduced mobility,
Fire Mater. 39 (4) (2015) 407-429.
[31] A. Hunt, Simulating hospital evacuation, presented at the Fire and Evacuation Modeling
Technical Conference, Malaga, Spain, 2016.
[32] A. Hunt, E.R. Galea, P.J. Lawrence, I.R. Frost, S.M.V. Gwynne, Simulating movement devices used
in hospital evacuation, Fire Technol. 56 (2020) 2209-2240.
[33] M. Yokouchi, Y. Hasegawa, R. Sasaki, R. Gaku, Y. Murata, N. Mizuno, et al., Operations analysis of
hospital ward evacuation using crowd density model with occupancy area and velocity by
patient type, in: Proceeding of the 2017 Winter Simulation Conference, 2017, pp. 2984-2993.
[34] D. Ursetta, A. D’Orazio, L. Grossi, G. Carbotti, S. Casentini, L. Poggi, Egress from a Hospital Ward:
A Case Study, presented at the Fire and Evacuation Modeling Technical Conference,
Gaithersburg, USA, 2014.
[35] National Fire Protection Association, Emergency Planning Guide for People with Disabilities,
NFPA DARAC, Quincy: National Fire Protection Association, 2016 Edition.
[36] E. Ronchi, P.A. Reneke, R.D. Peacock, A method for the analysis of behavioural uncertainty in
evacuation modelling, Fire Techn., 50 (2014) 1545–1571.
[37] R. Lovreglio, M. Spearpoint, M. Girault, The impact of sampling methods on evacuation model
convergence and egress time, Reliab. Eng. Syst. Saf., 185 (2019) 24-34.
[38] A. Grandison, Determining Confidence Intervals, and Convergence, for Parameters in Stochastic
Evacuation Models, Fire Techn., 56 (2020) 2137–2177.
[39] E. Smedberg, M. Kinsey, E. Ronchi, Multifactor Variance Assessment for Determining the Number
of Repeat Simulation Runs in Evacuation Modelling, Fire Technol. 57 (2021) 2615-2641.
[40] M.K. Lindell, R.W. Perry, The Protective Action Decision Model: theoretical modifications and
additional evidence, Risk Anal. 32(4) (2012) 616–632.
[41] E. D. Kuligowski, Human Behavior in Fire, in: M.J. Hurley, D.T. Gottuk, J.R. Hall, K. Harada, E.D.
Kuligowski, M. Puchovsky, J.L. Torero, J.M. Watts, C.J. Wieczorek (Eds.), in SFPE Handbook of
Fire Protection Engineering, Springer New York, New York, NY, 2016, pp. 2090–2114.
[42] SFPE Society of Fire Protection Engineers, SFPE Guide to Human Behavior in Fire, 2nd Edition,
Springer (2019).
[43] J.D. Sime, Affiliative Behaviour During Escape to Building Exits, J. of Environ. Psychol. 3 (1)
(1983) 21–41.
[44] G. Proulx, Evacuation Times and Movement Times in Apartment Buildings, Fire Saf. J. 24 (1995)
229-246.
[45] National Fire Protection Association, Guide on Alternative Approaches to Life Safety, NFPA
101A, Quincy: National Fire Protection Association, 2022 Edition.
[46] J.I. Kailes, Tips for Interacting with People with Disabilities, Harries Family Center for Disability
and Health Policy (2011).
[47] J.I. Kailes, Checklist for Integrating People with Disabilities and Others with Access and
Functional Needs into Emergency Planning, Response & Recovery, Harries Family Center for
Disability and Health Policy (2014).
[48] A. Serra, Rescue and people with special needs profiling, International workshop on rescue and
people with special needs – comparing experiences, Rome, Italy, 2014.
[49] World Health Organization, ICF - International classification of functioning, disability and health,
WHO (2017).
[50] European Committee for Standardization, Safety rules for the construction and installation of
lifts. Particular applications for passenger and good passenger lift. Part 72: firefighters lifts, EN
81-72. Brussels: CEN, 2020.
[51] European Committee for Standardization, Safety rules for the construction and installation of
lifts. Particular applications for passenger and good passenger lift. Part 70: accessibility to lifts
for persons including persons with disability, EN 81-70. Brussels: CEN, 2018.
[52] F.A. Ponziani, A., Tinaburri, V. Ricci, Modelling meta-communication: augmented communication
in rescue operation and agent based modelling, presented at Complex Systems 2018
international conference, Opatija, Croatia, 2018.
[53] National Association for Home & Care Hospice, Emergency preparedness packet for home health
agencies (2008).
[54] International Organization for Standardization, Fire-safety engineering - Technical information
on methods for evaluating behaviour and movement of people, ISO/TR 16738. Geneva: ISO,
2009.
[55] P. Geoerg, F. Berchtold, S. Gwynne, K. Boyce, S. Holl, A. Hofmann, Engineering egress data
considering pedestrians with reduced mobility, Fire Mater. 43 (7) (2019) 759-781.
[56] International Maritime Organization, Revised Guidelines on Evacuation Analysis for New and
Existing Passenger Ships, MSC.1/Circ.1533, International Maritime Organization, 2016.
[57] J.J. Fruin, Pedestrian Planning and Design, Elevator World, Inc., Mobile, AL (1987).
[58] K.E. Boyce, T.J. Shields, G.W.H. Silcock, Toward the characterization of building occupancies for
fire safety engineering: capabilities of disabled people moving horizontally and on an incline, Fire
Technol. 35 (1) (1999) 51-67.
[59] K. Miyazaki, M. Katsuhara, H. Matsukura, N. Kiriya, Evacuation experiments for wheelchair user
and pedestrian group, The Journal of Japan Institute of Navigation, 108 (2003) 129-138.
[60] S. Gwynne, E.R. Galea, J. Parke, J. Hickson, The collection of pre-evacuation times from evacuation
trials involving a hospital outpatient area and a university library facility, Proceedings of the
seventh international symposium fire safety science, 2003, pp. 877-888.
[61] S. Gwynne, E.R. Galea, J. Parke, J. Hickson, The collection and analysis of pre-evacuation times
from evacuation trials and their application to evacuation modelling, Fire Technol. 39 (2003)
173-195.
[62] British Standards Institution, Ed., Application of fire safety engineering principles to the design
of buildings. Part-5: Fire and rescue service intervention (Sub-system 5), PD 7974-5. London:
BSI, 2020.
[63] K. Christensen, Y. Sasaki, Agent-Based Emergency Evacuation Simulation with Individuals with
Disabilities in the Population, J. J. Artif. Soc. Soc. Simul. 11 (3) (2008): 9.
[64] British Standards Institution, Ed., Application of fire safety engineering principles to the design
of buildings. Part-6: human factors: life safety strategies - Occupant evacuation, behaviour and
condition (Sub-system 6), PD 7974-6. London: BSI, 2019.
[65] D.A. Samochine, K. Boyce K, T.J. Shields, An Investigation into staff behaviour in unannounced
evacuations of retail stores – implications for training and fire safety engineering, in: D. Gottuk,
B. Lattimer (eds.), in: Proceedings of the eight international symposium, 2005, pp.519–530.
[66] T.J. Shields, K.E. Boyce, A study of evacuation from large retail stores. Fire Saf. J. 35 (2000) 25–
49.
[67] S.M.V. Gwynne, D.A. Purser, D.L. Boswell, A. Sekizawa, Understanding and representing staff pre-
warning delay, J. Fire Prot. Eng. 22 (2) (2012) 77-99.
[68] J.D. Sime, The outcome of escape behaviour in the Summerland fire: Panic or Affiliation?, in:
Proceedings of the international conference on Building Use and Safety Technology, 1985, pp.
81-90.
[69] J.D. Sime, Movement toward the familiar: person and place affiliation in a fire entrapment setting,
Env. Behav., 17 (6) (1985) 697-724.
[70] W. Feinberg, Primary group size and fatality risk in a fire disaster, in: Proceedings of the second
International Symposium on Human Behaviour in Fire, 2001.
[71] British Standards Institution, Ed., Application of fire safety engineering principles to the design
of buildings. Part-7: probabilistic risk assessment, PD 7974-7. London: BSI, 2019.
[72] B. Efron, R. Tibshirani, An Introduction to the Bootstrap, Chapman and Hall, New York (1993).
[73] J.L. Devore, Probability and statistics for engineering and the sciences, 8th ed., Brooks/Cole,
Boston, MA (2010).
[74] H. Scheffé, The Analysis of Variance, Wiley, New York (1959).
[75] D. G. Bonett, Approximate confidence interval for standard deviation of nonnormal distributions,
Comp. Stat. & Data An. 50 (2006) 775–782.
[76] A. Grandison, S. Deere, P. Lawrence, E. R. Galea, The use of confidence intervals to determine
convergence of the total evacuation time for stochastic evacuation models, Ocean Eng. 146
(2017) 234-245.
[77] W. Oberle, Monte Carlo Simulations: Number of Iterations and Accuracy, US Army Research
Laboratory Report ARL-TN-0684 (2015).
[78] N.C. Wabo, P. Ortenwall, A. Khorram-Manesh, Hospital evacuation; planning, assessment,
performance and evaluation, J. Acute Dis., 1 (2012) 58-64.
[79] C. Castro, D. Persson, N. Bergstrom, S. Cron, Surviving the storms: emergency preparedness in
texas nursing facilities and assisted living facilities, J. Ger. Nurs. 34 (2008) 9-16.
[80] C. Cocanour, S.Allen, J. Mazabob, J. Sparks, C. Fischer, J. Romas, K. Lally, Lessons learned from the
evacuation of an urban teaching hospital, Arch. Surg. 137 (2002) 1141-1145.
[81] J. Barnett, M. Dennis-Rouse, V. Martinez, Wildfire disaster leads to facilities evacuation, Orthop.
Nurs. 28 (2009) 17-20.
[82] T. Nagata, S. Himeno, A. Himeno, M. Hasegawa, A.K. Lefor, M. Hashizume, Y. Maehara, M. Ishii,
Successful hospital evacuation after the Kumamoto earthquakes, Japan, 2016, Disaster Medicine
and Public Health Preparedness, 11 (2017) 517-521.
[83] J. Augustine, J.T. Schoettmer, Evacuation of a rural community hospital: lessons learned from an
unplanned event, Disaster Management & Response 3 (2005) 68-72.
[84] World Health Organization, Hospital safety index – Guide for evaluators, WHO (2015).
[85] S. Keret, M. Nahari, O. Merin, L. Aharonson-Daniel, S., Goldberg, B. Adini, Facilitating emergency
hospital evacuation through uniform discharge criteria, Am. J. Em. Med., 35 (2017) 681-684.
[86] K.M. Taaffe, R. Kohl, D. Kimbler, Hospital evacuation: issues and complexities, Proceeding of the
2005 Winter Simulation Conference, 2005, pp. 943–950.
[87] K.M. Taaffe, M. Johnson, D. Steinmann, Improving hospital evacuation planning using simulation,
Proceeding of the 2006 Winter Simulation Conference, 2006, pp. 509-515.
[88] E.Tayfur, K.M. Taaffe, Allocation of resources for hospital evacuations via simulation,
Proceedings of the 2007 Winter Simulation Conference, 2007, pp. 1148-1154.
[89] Italy, Decree of the Minister of Health of 2nd April 2015 n. 70, Technical specifications
establishing the qualitative, structural, technological and quantitative standards for hospitals.
[90] K.E. Boyce, T.J. Shields, G.W.H. Silcock, Toward the characterization of building occupancies for
fire safety engineering: capability of disabled people to negotiate doors, Fire Technol. 35 (1)
(1999) 68-78.
[91] Thunderhead Engineering, Pathfinder – Technical reference, 2020.
[92] A. Tinaburri, F.A. Ponziani, V. Ricci, Agent based modeling of meta-communication with assisted
people during emergency egress, presented at the Fire and Evacuation Modelling Technical
Conference, Gaithersburg, MD, USA, 2018.
[93] D.W. Scott, Multivariate Density Estimation: Theory, Practice, and Visualization, John Wiley &
Sons, Inc., (1992).
[94] M. Patricio, F. Ferreira, B. Oliveiros, F. Caramelo, Comparing the performance tests with ROC
analysis and confidence intervals, Comm. Stat. Simulat. Comput. 46 (10) (2017) 7535-7551.
[95] A. Tinaburri, Principles for Monte Carlo agent-based evacuation simulations including occupants
who need assistance. From RSET to RiSET. Fire Saf. J. 127 (2022) 1–21.