Parsing Nursing Notes

Bill Long
CSAIL Clinical Decision Making Group
Motivation

„ Automate collection of data on ICU

patients
„ Nursing notes are only source of many
kinds of data
– Symptoms, diseases, med changes,
interventions, other observations
„ Nursing notes record impressions of
clinical staff during course
Use of NLP for ICU Cases

„ Highlighting pertinent information

from nursing notes for annotation
„ Automatically gathering and coding
history and diagnoses from discharge
summary
„ Extracting patient data from notes and
summary automatically
 The Problem
RESP--CONTS ON AC 12, 800, PEEP 10 FIO2 DROPPED TO 40%. SX'D X1 BY RT FOR
THICK BLOODY SPUTUM.
CARD--PA LINE PLACED INITAL # PA 70/35, CVP 22, PCWP 21. AWAITING CXR TO
CONFIRM PLACEMENT PRIOR TO SHOOTING CO #'S. WITH HISTORY OR TR, WILL
MOST LIKELY NEED CO BY FICK METHOD. IONIZED CA 0.86, CURRENTLY RECEIVING
8 AMPS CA GLUCONATE. RECEIVED 2U FFP, AND DDAVP FOR SWAN INSERTION.
NEURO--PT VERY RESTLESS AND AGITATED DURING SWAN INSERTION, PROPFOL
INC TO 75MC/KG/HR, MED WITH ATIVAN 4MG IVP, BOTH WITH GOOD EFFECT.
GU--U/O GOOD.
GI--ABD SOFT AND OBESE, CONTS WITH BLOOD IN MOUTHM ALTHOUGH ONLY SM
AMOUNTS.

PT continues to wean on trac collar 40% 12 lpm sats high 90's rr teens. goal: dc
vent today.
cv-at 90's bp^'ing 150/80 range titrated ntg back to 80 and ^captopril dose-
co/ci/svr after 18.75 mgs of captopril 5.7/2.9/1095- room to ^captopril further
Steps to Understanding

„ Tokenization
„ Recognizing special forms
„ Expanding abbreviations
„ Spelling correction
„ Separating sections
„ Classifying & coding words
„ Understanding phrases
 Processing
NAUSEA-NOVOMITING.ACTION-ZOFRAN 2MG IV GIVEN X1.
Tokenization

NAUSEA – NOVOMITING . ACTION - ZOFRAN 2MG IV GIVEN X1 .

Abbreviation expansion

NAUSEA – NOVOMITING . ACTION - ZOFRAN

(UNITVALUE (2 "milligram")) (ABBREV "intravenous") GIVEN (TIMES 1) .
Spelling correction

NAUSEA – ("no" "vomiting") . ACTION – (MEDICATION ZOFRAN)

(UNITVALUE (2 "milligram")) (ABBREV "intravenous") GIVEN (TIMES 1) .
Types of Tokens

„ word <word> „ 20cc/hr <unit value>

„ 133 <integer> „ c/o <abbrev>
„ 101.5 <float> „ x2 <abbrev>
„ 8/20 <date> „ CV: <section>
„ 8:30 <time> „ CV- <section>
„ 102-104 <range> „ dc’d <past tense>
„ >90 <range> „ po’s <plural>
„ 14% <percent> „ pt’s <possessive>
„ 200cc <unit value> „ 90’s <range>
Types of Abbreviations

„ s/p status post, c/o complains of

„ w/<word> with <word>
„ noc at night, po by mouth
„ med medication, chol cholesterol
„ + present, ^ elevated, ? possible
„ mg milligrams or magnesium
 Challenges

„ Some spelling is beyond the spell checker

– ungcahnged, ontpn, overnoc, imprvoedand
„ Many abbreviations local (first letters of
word usually)
– oint, liq, tol
„ Some abbreviations have multiple
meanings
– s=small, some; po=oxygen levels, by mouth;
bs=blood sugar, breath sounds, bowel sounds
Multiple Reps

AWAKE, ALERTAND ORIENTED,

FOLLOWS SIMPLE COMMANDS
ALERT AND ORIENTED
AWAKE ALERT AND ORIENTED
oriented x 3
ORIENTED X3
ALERT AND ORIENTED X3
ALERT AND ORIENTD
A+OX3
Example

„ A: captopril 12.5mg po at 0600

„ P: ({DRUG-THERAPY
– {DOSE=(12.5 {MILLIGRAM})})
– (_ROUTE {ORALLY})
– {_DOSE-TIME=({_PTIME=(0600)})} {AT})})
– {CAPTOPRIL})})
Medications Drug therapy = red
Dosage = blue
 Symptom = red
Symptoms Category = blue
 Disease = red
Diseases Rhythm = blue
 Extracting & Coding PMH
PAST MEDICAL HISTORY: Included chronic obstructive pulmonary
disease, Wegener's vasculitis, steroid induced diabetes
mellitus, no history of coronary artery disease.
Lisp text extractor
case: (3450-De-ID Discharge Summary)
Included chronic obstructive pulmonary disease,
Wegener's vasculitis,
steroid induced diabetes mellitus,
no history of coronary artery disease.
Using UMLS plus NLP
{chronic obstructive pulmonary disease},
{Wegener's} {vasculitis},
{steroid} induced {diabetes mellitus},
{no} {history} of {coronary artery disease}.
Applying the Heart
Disease Program in ICU
Bill Long
CSAIL Clinical Decision Making Group
Methodology

„ Heart Disease Program

„ Uses physiologic model and Bayesian net
approach with temporal constraints
„ Generates complete hypotheses to
explain findings
„ Gives rank ordered list of hypotheses as
differential
 HDP on case 2069
History
age: 84 year old
sex: female
dyspnea: none
chest-pain: none
other-symptoms: none
known-diagnosis: of (1) coronary-heart-disease (2) hypertension (3)
acute-blood-loss for 1 day
procedures-done: (1) aortic-valve-replacement occurred 5 yrs ago (2)
coronary-artery-bypass-graft occurred 5 yrs ago
therapies: on (1) furosemide for 5 yrs (2) beta-blocker
non-cardioselective for 5 yrs (3) ace-inhibitor for 5 yrs (4)
coumadin for 5 yrs
Vital-Signs
blood-pressure: 108/56
pressure-abnormalities: none
heart-rate: 64
resp: 16
temp: 98.0
Case cont’d
Physical-Exam
appearance: appears diaphoretic
mental-status: conscious
chest: revealed no crackles
jugular-pulse: shows JVP normal
pulse: normal
auscultation: normal-exam
apical-impulse: normal
parasternal-impulse: none
abdomen: normal
extremities: (1) no pedal-edema (2) calf-tenderness/swelling (3)
cool/clammy extremities
Routine-Investigations
blood-chemistry (1 hr): Na: 136 K: 4.9 creat: 2.0
CBC (1 hr): HCT: 29 WBC: 13.0
HX-HYPERTENSION (possibly caused by RENAL-VASCULAR-DISEASE), treated with
BETA-BLOCKER and ACE-INHIBITOR
as indicated by known-diagnosis of hypertension; causing
LV-HYPERTROPHY causing
LOW CARDIAC-OUTPUT as indicated by cool/clammy extremities, (but
pressure-abnormalities no-pressure-drop); causing
TREATED-FLUID-RETENTION, treated with FUROSEMIDE
LOW BLOOD-PRESS as indicated by bp: 108 / 56
HIGH LA-PRESS, treated with FUROSEMIDE, (but no-orthopnea, no dyspnea);
causing
TACHYPNEA (R: 16)
PULMONARY-CONGESTION as indicated by WBC 1hr: 13.0, (but no-PND)
RENAL-INSUFFICIENCY (possibly caused by RENAL-VASCULAR-DISEASE)
as indicated by creat 1hr: 2.0, causing ANEMIA and HX-HYPERTENSION
ANTI-COAGULANT TOXIC causing
BLOOD-LOSS as indicated by known-diagnosis of acute-blood-loss
ANEMIA (caused by RENAL-INSUFFICIENCY is false) as indicated by HCT 1hr: 29,
(but no dyspnea, HIGH HEART-RATE ABSENT (hr: 64))
CORONARY-ARTERY-DISEASE, treated with CORONARY-ARTERY-BYPASS-GRAFT
as indicated by known-diagnosis of coronary-heart-disease, (but
no-chest-pain-history); causing
RENAL-VASCULAR-DISEASE as indicated by
causing RENAL-INSUFFICIENCY and HX-HYPERTENSION
Findings not accounted for: calf-tenderness/swelling, AORTIC-VALVE-REPLACEMENT.
Predicting Effects of
Therapy
„ Assume cardiovascular system goes
from one steady state before therapy
to a second steady state after therapy
„ Infer initial steady state from clinical
measurements
„ Use therapy and one measurement to
infer “quantity” of therapy
„ Use therapy and quantity to estimate
other parameters
 Hemodynamic Model
((BLOOD-VOL blood-vol-initial) ; constant for short times (renin-angio ignored)
(VENOUS-CONSTR (- 1.0 (* .12 nitro)))
(VENOUS-VOL (- BLOOD-VOL PULM-VOL
(/ 3.2 (+ (* .11 exercise)
(* (+ .7 (* .3 SYMP-STIM)) VENOUS-CONSTR)))))
; right output
(RAP (- (* 5.7 VENOUS-VOL)(* RESIST-VENOUS-RET CO)))
(RVEDP RAP)
(RV-OUTPUT (* RV-COMPL RV-EMPTYING (* 1.375 (+ RVEDP 4.0))))
(PA-PRESS (+ LAP (* PULM-VASCUL-RESIST RV-OUTPUT)))
; left output
(LAP (if (< PULM-VOL 0.7) (- (* 36.0 PULM-VOL) 10.2)
(- (* 125.0 PULM-VOL) 72.5)))
(LVEDP (- LAP (* 620.0 MITRAL-STENOSIS (expt (/ CO DIAS-TIME) 2))))
(LV-OUTPUT (* (IF (< LVEDP 8.0)(+ (* 0.75 LVEDP))(+ (* .5 LVEDP) 2.0))
LV-COMPL LV-EMPTYING))
(AR-FLOW (* .00137 (- BLOOD-PRESS 30.0) DIAS-TIME AORTIC-REGURG))
(CO (- LV-OUTPUT AR-FLOW))
 Hemodynamic Model
(cont’d)
; blood pressure and sympathetic response
(BLOOD-PRESS (+ (* CO SVR) RAP))
(SYSTOL-PRESS (+ BLOOD-PRESS (* 228.0 AORTIC-STENOSIS
(expt (/ CO SYST-TIME) 2))))
(SYMP-STIM (+ (* -.03 (- BLOOD-PRESS blood-press-base
(* 47.5 exercise))) 1.0))
(VAGAL-STIM (+ 1.0 (* .033 (- BLOOD-PRESS blood-press-base
(* 60.6 exercise)))))
; heart rate
(HEART-RATE (+ base-heart-rate
(- (* 37.5 (+ (* SYMP-STIM (/ (+ 1.0 beta-block)))
(* .376 dobut-amine)))
(if (< VAGAL-STIM 1.0)(* 23.0 VAGAL-STIM)
(if (< VAGAL-STIM 2.0)(+ (* 17.0 VAGAL-STIM) 6.0)
(+ (* 10.0 VAGAL-STIM) 20.0))))))
(SYST-TIME (+ 17.3 (* .075 HEART-RATE)))
(DIAS-TIME (- 42.7 (* .075 HEART-RATE)))
; left systolic function
(INOTROP (* SYMP-STIM (/ (+ 1.0 beta-block))(+ 1.0 (* .3 hydral-azine))
(+ 1.0 (* 1.33 dobut-amine))))
(LV-SYSTOLIC-FUNCT (* INOTROP lv-systolic-funct-base))
(LV-EMPTYING (* LV-SYSTOLIC-FUNCT
(IF (< SYSTOL-PRESS 100) 1.0
(+ 1.4 (* -0.004 SYSTOL-PRESS)))))
; right systolic function
(RV-SYSTOLIC-FUNCT (* INOTROP rv-systolic-funct-base))
(RV-EMPTYING (* RV-SYSTOLIC-FUNCT
(if (< PA-PRESS 20) 1.0 (- 1.1 (* .005 PA-PRESS)))))
; vascular resistance
(SVR (/ (+ svr-base (* SYMP-STIM (- 1.0 (* -.4 beta-block)) 5.0)
(* exercise -4.65))
(+ 1.0 (* .64 hydral-azine)(* .66 dobut-amine))))
(RESIST-VENOUS-RET (+ (* SVR 0.025) .9))
(PULM-VASCUL-RESIST (/ (* (+ pvr-k1 (* RV-OUTPUT -.041))
(+ 1.0 (* .5 beta-block)))
(+ 1.0 (* .625 hydral-azine)(* .62 dobut-amine))))
; pulmonary volume
(PULM-VOL (integral (- RV-OUTPUT CO)))))
Results
Sample case
„ Dobut „ 0 „ 1
„ Blood „ 0 „ 1
„ BP „ 100 „ 125 (120)
„ HR „ 120 „ 94 (70)
„ CO „ 4 „ 7
„ LAP „ 8 „ 7.5
„ PAP „ 40 „ 63
„ RAP „ 4 „ 3.5
„ MVO2 „ 13 „ 26
Challenge

„ For ICU data the problem is explaining

changes
„ Start with one state, determine
implied parameters
„ Interpret new state as an unknown
perturbation of old state