VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jval

Cost-Effectiveness of Enhancing Adherence to Therapy with Blood

Pressure–Lowering Drugs in the Setting of Primary
Cardiovascular Prevention
Lorenza Scotti, PhD1, Gianluca Baio, PhD1, Luca Merlino, MD2, Giancarlo Cesana, MD3, Giuseppe Mancia, MD3,4,
Giovanni Corrao, PhD1,
1
Department of Statistics and Quantitative Methods, Unit of Biostatistics, Epidemiology and Public Health, University of Milano-Bicocca, Milan, Italy; 2Operative
Unit of Territorial Health Services, Lombardia Regional Council, Milan, Italy; 3Department of Clinical and Preventive Medicine, University of Milano-Bicocca,
Milan, Italy; 4Clinica Medica, Ospedale S. Gerardo, Monza, Italy

AB ST RAC T

Objective: To estimate the cost-effectiveness of enhancing ad respectively) and increased the cost for drug therapy (from h1,325k to
herence to blood pressure (BP)-lowering drug therapy in a large h1,507k and h1,934k every 10,000 person-year, respectively). The
population without signs of preexisting cardiovascular (CV) disease. resulting incremental cost-effectiveness ratio decreased from h76k
Methods: A cohort of 209,650 patients aged 40 to 79 years resident in (95% confidence interval h74k–h77k) to h74k (95% confidence interval
the Italian Region of Lombardia and newly treated with BP-lowering h72k–h75k) for each CV event avoided by enhancing adherence from
drugs during 2000 to 2001 was followed from index prescription to baseline to 60% and 80%, respectively. Conclusions: Enhancing adher-
2007. During the follow-up, the 10,688 patients who experienced a ence to BP-lowering medications in the setting of primary CV
hospitalization for a coronary or cerebrovascular event were identified prevention might offer important benefits in reducing the risk of CV
(outcome). Adherence was measured by the proportion of days outcome, but at a substantial cost.
covered by the therapy with BP-lowering drugs. The cost- Keywords: adherence, administrative database, blood pressure–lowering
effectiveness of enhancing adherence was measured through the drugs, cardiovascular events, cost-effectiveness.
incremental cost-effectiveness ratio. Results: Enhancing adherence
from 52% (baseline) to 60% and 80% led to a reduced rate for CV Copyright & 2013, International Society for Pharmacoeconomics and
outcomes (from 85 to 83 and 77 events every 10,000 person-year, Outcomes Research (ISPOR). Published by Elsevier Inc.

this type necessarily lead to an increment in the overall costs

Introduction
because of increased drug use. This is particularly important
Randomized clinical trials have consistently shown that hyper- from the public health perspective because the economic
tension is a reversible risk factor; that is, a reduction in elevated resources available are limited and it is important to allocate
blood pressure (BP) induced by pharmacological treatments them in the best way to maximize the level of population health
reduces the risk of fatal and nonfatal cardiovascular (CV) events [17]. It is, therefore, suitable to jointly evaluate cost and effec-
[1]. It is also well known, however, that effective BP reduction is tiveness of the treatment to quantify the additional cost needed
rare in the hypertensive population [2–5], and in individuals with to increase the effectiveness of treatment-related enhanced
uncontrolled BP, the incidence of CV events is higher than among adherence. To the best of our knowledge, no such evaluation
patients who have achieved BP control [6]. Evidence shows 1) very has been carried out in the setting of primary prevention of CV
low adherence to BP-lowering medication in the setting of daily outcomes and using BP-lowering drug therapy.
clinical practice [7,8], 2) that poor adherence to antihypertensive We estimated the cost-effectiveness of enhancing adherence
medication is related to lack of BP control [9], and 3) that there is to BP-lowering drug therapy in a large population without signs
a relationship between adherence to BP-lowering drugs and the of preexisting CV disease in the setting of primary CV prevention.
risk of CV outcomes [10–15]. Costs and effectiveness were estimated from a population-based
These findings suggest that interventions aimed at increasing cohort study. Data were derived from an administrative database
adherence to BP-lowering agents would be effective in achieving monitoring the use of health services of the population residing
full benefits from drug treatment [16]. However, interventions of in the Lombardia region (Italy).

 Address correspondence to: Giovanni Corrao, Department of Statistics, Unit of Biostatistics and Epidemiology, University of Milano-
Bicocca, Via Bicocca degli Arcimboldi, 8, 20126 Milan, Italy.
E-mail: [email protected].
1098-3015/$36.00 – see front matter Copyright & 2013, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).
Published by Elsevier Inc.
http://dx.doi.org/10.1016/j.jval.2012.11.008
 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4 319

number of days with drug available was calculated by dividing

Methods the total amount of active drug dispensed at each prescription by
the recommended defined daily dose. In this way, each day in the
follow-up period was labeled as ‘‘covered’’ or ‘‘not covered’’ by
Setting
drug availability, regardless of whether availability concerned
The data used for the present study were retrieved from the single or combined prescriptions. Adherence was measured by
health care service databases of Lombardia, a region of Italy that dividing the cumulative number of covered days by the number
accounts for about 16% (9 million) of its population. In Italy, of days of follow-up, a measure denoted as proportion of days
health care is provided by the National Health Service (NHS). In covered (PDC) [19,20]. PDC was categorized into the following four
Lombardia, this has been associated since 1997 with an auto- levels: very low (r25%), low (26%–50%), intermediate (51%–75%),
mated system of databases to collect a variety of data, including: and high adherence (475%).
1) an archive of residents who receive NHS assistance (effectively
the whole resident population), including demographic and
administrative data; 2) a hospital discharge database providing Outcome
data on diagnoses recorded for each admission in a public or Data on hospital discharge were used to identify cohort members
private hospital of the region; and 3) a database of outpatient who experienced a hospitalization for coronary or cerebrovascu-
drug prescriptions reimbursable by the NHS; this includes data lar (major CV) event during the follow-up. The WHO-MONICA
on Anatomical Therapeutic Chemical (ATC) code and the corre- criteria for the ascertainment of coronary and cerebrovascular
sponding amount of active drug for each prescription dispensed events were followed [21,22]. Coronary events included those
by a pharmacy of the region. related to acute myocardial infarction, acute or subacute types of
For each patient we linked the above databases by using a single ischemic heart disease, and interventions of coronary revascu-
identification code. To preserve privacy, each identification code larization. Cerebrovascular events included those related to sub-
was automatically converted to an anonymous code. The inverse arachnoid hemorrhage, intracerebral hemorrhage, unspecified
process was prevented by the deletion of the conversion table. Full intracranial hemorrhage, occlusion of cerebral arteries, acute
details of the procedure have been reported elsewhere [18]. cerebrovascular disease, and surgical interventions on intra- or
extracranial head or neck vessels. The occurrence of at least one
of these events was sufficient for a patient to be considered as
Study Cohort and Follow-Up
experiencing the outcome; the earliest date of hospital admission
The Lombardia residents who were beneficiaries of the NHS and recording one of these events was considered as the time of the
were aged 40 to 79 years represented the target population. outcome onset.
According to the 2001 Italian Census, this population comprised
4,341,438 individuals. Among them, we identified those who
received at least one BP-lowering drug prescription at any time Other Factors
from January 1, 2000, to December 31, 2001. The first drug The type of drug regimen at the index date (monotherapy or a
prescription was defined as the index prescription. The drugs combination of two or more drugs as first-line BP-lowering
considered belonged to all the available BP-lowering drug classes, therapy) and the number of BP-lowering classes used during
that is, agents acting on alpha-adrenergic receptors (ATC code the follow-up were recorded. Drugs used for heart failure or
C02), diuretics (ATC code C03), beta-blockers (ATC code C07), coronary heart disease (i.e., digitalis glycosides and organic
calcium channel blockers (ATC code C08), and agents acting on nitrates), lipid-lowering agents, other CV drugs, and antidiabetic
the renin-angiotensin system (ATC code C09), dispensed either as medications dispensed to each cohort member during the follow-
monotherapy or as a fixed-dose or extemporaneous combination up were also recorded. In addition, the Charlson comorbidity
of two or more drugs. index [23] was calculated by using diagnostic information from
To make the data as relevant as possible to the study aim, four inpatient visits in the 3 years prior to and 1 year after the index
categories of patients were excluded from data analysis: 1) date. The index was summarized by using two categories (i.e., 0
patients who had received at least one BP-lowering drug pre- or 1, respectively, suggesting the absence or presence of at least
scription within 3 years before the index prescription, to ensure one comorbidity factor).
the inclusion of newly treated individuals only; 2) patients who
had been hospitalized for CV disease or who had received a Estimating the Relationship between Adherence and Outcome
prescription of drugs used for coronary heart disease or heart
failure (e.g., digitalis and organic nitrates) in the 3 years preced- A time-to-event analysis was undertaken by using the Cox
ing the index prescription, to focus the data on primary CV proportional hazards model to estimate the hazard ratio (HR),
prevention; 3) patients who did not reach at least 1 year of follow- and the corresponding 95% confidence interval (CI), for the
up, to ensure at least 1 year of potential exposure to the drugs of association between adherence to BP-lowering drug therapy
interest; and 4) patients who had received only one BP-lowering and the time of outcome occurrence. The predictor variables of
drug prescription during the first year after the date of index interest were the dummy factors constructed according to the
prescription, based on the assumption that continuous drug categories of PDC, using very low adherence as the reference
treatment might not have been indicated for these patients. category. Estimates were adjusted for factors measured at base-
Each member of the cohort accumulated person-years (PY) of line (such as age, gender, antihypertensive drug regimen, and
follow-up from the date of index prescription until the first-ever Charlson comorbidity index) as well as during follow-up (i.e.,
hospital admission for CV disease or the end of the study period number of BP-lowering classes and cotreatments with CV and
(December 31, 2007). Individuals who transferred out of the antidiabetic drugs during follow-up). Because adherence, as well
region or died during follow-up were censored. as all the other factors measured during follow-up, can change
over time, the assessment of their effects requires properly
accounting for the time-varying nature of these variables. This
Assessment of Adherence to BP-Lowering Drug Therapy was done by fitting a Cox model that includes these factors as
All prescriptions dispensed to each cohort member during the time-dependent covariates [24]. For instance, by considering the
follow-up were used to measure the cumulative exposure to predictor variables of interest (i.e., the dummy factors of PDC),
BP-lowering drugs. Starting from the index prescription, the each subject’s cumulative adherence is recalculated from the
320 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4

start of the follow-up to the time of each outcome occurrence. the baseline rates over the PDC categories at the corresponding
Thus, the HR associated with a category of PDC is derived by expected distribution of PY of follow-up. The annual number of
using information updated at the time of each observed outcome. hospitalizations for major CV events that would be avoided by
The same approach was used for the other time-varying factors increasing adherence at the level imposed by a given scenario
mentioned above. was estimated as the difference between the observed and
The methods described above have been reported in detail in expected number of CV events.
an article investigating the adherence-outcome relationship, Next, we computed the average annual cost for BP-lowering
authored by our group [15]. What was not investigated by our drug therapy observed at baseline and that expected for each
previous article was the cost-effectiveness profile of enhancing given scenario. Observed costs were measured from the perspec-
adherence. The next section describes in detail the methods used tive of the Italian NHS by using the amount that the Regional
for answering this question, which represents the main objective Health Authority reimbursed pharmacies for dispensing drugs.
of the current study. With reference to drug price in the year 2009, the average daily
cost for treating a patient in our setting was h0.62, ranging from
h0.07 to h0.90 for therapy with the generic diuretic chlortalidone
Estimating the Cost-Effectiveness Profile of Enhancing
to the combination of the angiotensin receptor antagonist can-
Adherence desartan with the diuretic hydrochlorothiazide, respectively. The
Starting from the observed average PDC (baseline), three scenar- expected drug costs were estimated for the expected distribution
ios were built by progressively enhancing the average level of of PY of follow-up over the categories of PDC, assuming that the
adherence to 60%, 70%, and 80%. For each scenario, we calculated prescription profile does not change when increasing adherence.
the decreased number of CV events with respect to the baseline The average annual incremental drug cost due to the increased
and the corresponding cost of drug therapy as follows. use of drugs was estimated as the difference between the
First, we considered the distribution of PY of follow-up over observed cost with baseline PDC and that expected for a given
the categories of PDC observed at baseline and that expected for a scenario.
given scenario. The expected distribution of PY of follow-up was Finally, by comparing each scenario with the baseline, we
computed from the expected PDC value for each patient obtained computed the incremental cost-effectiveness ratio (ICER) [26] as
by adding to his or her observed coverage a random value drawn the ratio between the incremental drug cost (in h) and the
from a two-sided truncated normal distribution [25], with mean number of CV hospitalizations avoided. According to the proce-
and variance equal to the difference between the average PDC at dure described above, the ICER must be interpreted as the
baseline and that imposed by the scenario. additional annual drug cost that would be borne by the NHS to
Then, we estimated the average annual incidence rate of avoid one hospitalization for a major CV event every year, as a
hospitalization for major CV events at baseline and that expected consequence of enhancing adherence at a certain level among
for a given scenario. The baseline rates over the PDC categories new users of BP-lowering drugs without established CV disease.
were computed from the incidence rate observed among patients Because the distribution of ICER is unknown, we used a
with very low adherence and the HRs obtained from the Cox nonparametric bootstrap method to account for the underlying
model. The annual number of hospitalizations for major CV uncertainty [27]. The 2.5th and the 97.5th percentiles of the boot-
events expected for a given scenario was computed by applying strap distribution were used to calculate the 95% CI for the ICER.

Fig. 1 – Flowchart of inclusion and exclusion criteria. BP, blood pressure; CV, cardiovascular.
 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4 321

A deterministic sensitivity analysis was performed by recal- the included patients. Overall, 44% of the patients had very low or
culating the ICER under the assumption that during follow-up all low adherence to BP-lowering drugs, and only 26% had high
patients are treated with the BP-lowering drug at the lowest cost adherence. Compared with patients who had high adherence,
among those available in the therapeutic class in which the those with a very low or low adherence were mainly women,
treatment begun. Accordingly, the average daily cost in this were initially treated with a combination of two or more BP-
situation was h0.075 (ranging from h0.069 to h0.41 for monother- lowering drugs, received a minor number of BP-lowering classes
apy with diuretics or a fixed combination of angiotensin receptor during follow-up, and had a better profile of cotreatments and
antagonist and diuretics, respectively). comorbidities.
For all the hypotheses tested, two-tailed P values less than The association between adherence to BP-lowering drug
0.05 were considered as significant. therapy and the risk of hospitalization for major CV event is
The analyses were performed by using SAS software version shown in Figure 2. Compared with patients with very low
9.1.3 (SAS Institute, Inc., Cary, NC). adherence level, those with low, intermediate, and high adher-
ence, respectively, showed HR of 0.95 (95% CI 0.90–1.01), 0.76 (95%
CI 0.72–0.81), and 0.72 (95% CI 0.68–0.76).
Figure 3 shows the trends in the number of hospitalizations
Results
for major CV events and cost for drug treatment (upper box), and
Inclusion and exclusion criteria are shown in Figure 1. The in the ICER (lower box), from baseline to any scenarios of
209,650 patients included in the study accumulated 1,244,870 enhanced adherence. The number of patients experiencing a
PY of follow-up (on average 6 years per patient) and 653,394 PY of hospitalization for major CV event decreases from 85.41 events
treatment with BP-lowering drugs (on average 52% of the time of every 10,000 PY observed at baseline to 83.03, 80.08, and 77.14
follow-up was spent with drug available). The cohort generated events every 10,000 PY, respectively, for 60%, 70%, and 80%
10,688 first hospitalizations for a major CV event; with an average level of adherence. Conversely, the cost for drug therapy
incidence rate of 86 cases every 10,000 PY. increases from the observed value of h1,324,727 at baseline to
Table 1 summarizes some selected characteristics of the h1,506,759, h1,724,770, and h1,933,440 every 10,000 PY, respec-
whole cohort and by adherence to statin therapy. The average tively, for 60%, 70%, and 80% average level of adherence. The
age at entry was 60 ⫾ 10 years, and 45% of the patients were men. estimates for the ICER and the corresponding 95% CI were h76,484
Most patients started BP-lowering therapy with one agent only, (h74,807–h78,152), h75,055 (h73,490–h76,623), and h73,605
but during the follow-up the majority of them received at least (h72,180–h75,157) for each hospitalization for a major CV event
three BP-lowering classes. About 44% of the patients received at avoided by enhancing the average level of adherence from base-
least once another drug for preventing or treating CV events line to 60%, 70%, and 80%, respectively. The corresponding values
during follow-up. Comorbidities, ascertained by a Charlson of the ICER were h25,400, h25,300, and h25,200 if all patients were
comorbidity index of more than 1, were present in about 13% of treated with the less expensive drug in each therapeutic class.

Table 1 – Selected characteristics of the included patients according to categories of adherence to therapy with
BP-lowering drugs.
Adherence level Py

Very low Low Intermediate High Total

(n ¼ (n ¼ (n ¼ 61,389) (n ¼ (n ¼
45,755) 47,374) 55,132) 209,650)

Length of follow-up (y), mean ⫾ SD 5.77 ⫾ 1.49 5.91 ⫾ 1.45 6.03 ⫾ 1.35 6.00 ⫾ 1.35 5.94 ⫾ 1.41 o.0001
Age (y), mean ⫾ SD 58.9 ⫾ 10.8 60.2 ⫾ 10.5 60.0 ⫾ 9.9 59.4 ⫾ 9.5 59.6 ⫾ 10.1 o.0001
Male gender, n (%) 17,248 (37.7) 19,733 (41.7) 27,654 (45.0) 28,663 (52.0) 93,298 (44.5) o.0001
Initial BP-lowering drug regimen, n (%)
Monotherapy 31,759 (69.4) 32,728 (69.1) 44,598 (72.6) 42,805 (77.6) 151,890 (72.5) o.0001
Combination of two or more drugs 13,996 (30.6) 14,646 (30.9) 16,791 (27.4) 12,327 (22.4) 57,760 (27.6)
No. of BP-lowering classes used during follow-up, n (%)
1 17,078 (37.3) 9,978 (21.1) 8,275 (13.5) 7,651 (13.9) 42,982 (20.5) o.0001
2 15,052 (32.9) 14,553 (30.7) 16,105 (26.2) 13,670 (24.8) 59,380 (28.3)
Z3 13,625 (29.8) 22,843 (48.2) 37,009 (60.3) 33,811 (61.3) 107,288 (51.2)
Users of other drugs during follow-up, n (%)
Drugs used in diabetes 5,133 (11.2) 6,261 (13.2) 9,253 (15.1) 9,899 (18.0) 30,546 (14.6) o.0001
Lipid-lowering drugs 10,233 (22.4) 12,727 (26.9) 18,764 (30.6) 17,839 (32.4) 59,563 (28.4) o.0001
Digitalis glycosides/organic o.0001
4,323 (9.4) 5,421 (11.4) 7,090 (11.6) 5,747 (10.4) 22,581 (10.8)
nitrates
Other cardiac drugs 5,112 (11.2) 5,917 (12.5) 7,552 (12.3) 6,085 (11.0) 24,666 (11.8) o.0001
Charlson comorbidity index, n (%)z
0 40,402 (88.3) 41,073 (86.7) 53,163 (86.6) 46,642 (84.6) 181,280 (86.5) o.0001
Z1 5,353 (11.7) 6,301 (13.3) 8,226 (13.4) 8,490 (15.4) 28,370 (13.5)

BP, blood pressure.

 Adherence was measured according to the proportion of days of observation covered by BP-lowering medication. Adherence levels are the

following: very low coverage (r25%), low coverage (26%–50%), intermediate coverage (51%–75%), and high coverage (475%).
y
P values obtained from chi-square test or F test (length of follow-up and age).
z
Measuring the extension of comorbidity from inpatient encounters in the 3 y prior and 1 y after the index date.
322 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4

treatment with generic antihypertensive drugs does not lead to

lower adherence or more CV disease–related hospitalizations
compared with brand-name therapy [30]. Second, several
patients received only one antihypertensive drug prescription in
our setting. These patients were excluded from our analysis
because continuous drug treatment might not have been indi-
cated for most of them. In fact, we recently showed that these
patients have a better risk profile and lower CV disease incidence
than do those on multiple drug prescriptions [31]. Thus, it is
reasonable to speculate that with respect to patients receiving
multiple prescriptions, most of those on single dispensation have
relatively modest hypertension and nonpharmacological advice
on healthier lifestyles (e.g., cut back on salt, physical exercise,
Fig. 2 – Trend in hazard ratios, and corresponding 95% and smoking cessation) might be the more appropriate indication
confidence intervals, for dose-response relationship for these patients [31]. Whatever the reason, however, only one
between adherence to therapy with BP-lowering drugs and dispensation for treating a condition that once properly diag-
the risk of developing a cardiovascular outcome. Adherence nosed requires lifetime treatment undoubtedly represents an
to therapy with antihypertensives was assessed by the ratio inappropriate allocation of health resources. The burden of this
between the cumulative number of days during which the phenomenon is not negligible because in our setting it accounts
medication was available and the whole number of days of for 65,000 patients and h1,300,000 every year [31]. These resources
follow-up (proportion of days covered). This indicator was might be properly allocated for increasing adherence. Of course,
categorized into four groups: very low coverage (r25%), low nonpharmacological advice could have been more indicated than
coverage (26%–50%), intermediate coverage (51%–75%), and drug therapy also for a portion of the included patients. However,
high coverage (475%). Hazard ratios were obtained by we are not able to know the size of this subgroup. Combining
fitting the Cox proportional hazards model. Estimates are these arguments suggests that the primary prevention of CV
adjusted for gender, age at entry, concomitant use of other events by enhancing drug adherence might become competitive
drugs, number of antihypertensive classes dispensed with respect to costs of care for patients, provided that less
during follow-up, first-line therapy regimen, and Charlson expensive BP-lowering drugs are dispensed and only those
comorbidity index. BP, blood pressure. patients who really need lifetime treatment are subjected to drug
therapy.

Discussion
Our study confirms previous observations that adherence to
antihypertensive drug treatment is rather low in clinical practice
and that low adherence is associated with increased CV risk
[7,10–15]. This implies that interventions aimed at enhancing
adherence would avoid many hospitalizations for CV outcomes,
even if at a substantial cost. As a novel and original message, we
estimated that about h76k per year would be additionally spent
for every 10,000 subjects initiating antihypertensive therapy to
avoid one hospitalization for a major CV event every year by
means of enhancing the average adherence from 52% (baseline)
to 60%. Costs for avoiding the considered outcome only weakly
decreased by enhancing the average adherence from baseline to
70% (h75k) or 80% (h74k).
Such estimates should be compared with the cost borne by
the Italian NHS for attending to patients affected by major CV
disease. A recent Italian study estimated that the first episode of
acute myocardial infarction has an annual cost of h15k in the first
year and h3.7k in the following years [28]. It has also been
reported that in Germany the costs for hospitalization, drug
prescription, rehabilitation, and nursing are h19k and h43k,
respectively, during the first year and the lifetime after the first-
ever ischemic stroke [29]. At first glance, then, the cost of treating
a major CV event is lower than the cost that should be borne by
preventing one hospitalization for a major CV event by means of
enhancing drug adherence.
At least two considerations need, however, to be further
addressed before drawing conclusions on this subject. First, we Fig. 3 – Trends in the distribution in the number of nonfatal
showed that the cost of avoiding a CV hospitalization might be cardiovascular (CV) events and cost for drug therapy (upper
reduced by up to two third by using less expensive drugs. That is, box), and in incremental cost-effectiveness ratio, with
adherence-enhancing interventions might be much more cost- corresponding 95% confidence intervals (lower box), every
effective by encouraging the use of drugs at lowest cost among 10,000 person-year, from baseline to scenarios of
those available in a given therapeutic class, typically generic progressively enhancing average adherence to therapy with
drugs. Of course, this is true under the assumption that BP-lowering drugs. BP, blood pressure.
 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4 323

This study is unique in several respects. First, we used data included in our study may have had conditions other than hyper-
that were made available thanks to the fact that in Italy 1) a cost- tension that require BP-lowering drug administration. A recent
free uniformly organized health care system involves practically study, however, has reported that in Italy hypertension represents
all citizens, 2) the health care system is managed by a set of by far the most common diagnosis for the use of antihypertensive
databases including inpatients hospital discharge and outpati- drugs, with only about 20% being used for angina pectoris, myo-
ents drug prescription data, and 3) these databases guarantee cardial infarction, and heart failure (often associated with hyperten-
universal coverage of the population receiving health assistance sion anyway) and less than 1% for other acute indications, such as
from the NHS. Consequently, the complete history of health care edema [38]. Second, because the data on outcome onset have been
utilization of a very large unselected population is available and drawn from the hospital discharge archive, our estimates concern
real-world clinical practice, including health and economic bur- only nonfatal outcomes. Because there is no evidence of hetero-
den of medical interventions, may be investigated. To the best of geneous effects of antihypertensive drugs on the risks of fatal and
our knowledge, the current study offers unique insight into the nonfatal coronary events [39], however, the selection of nonfatal
potential of health care utilization database for estimating the outcomes only should not affect the validity of our estimates. Third,
cost-effectiveness profile of medical interventions on hyperten- our study does not account for several health benefits that might
sion. Second, we have previously reported that hospital discharge derive from enhancing adherence. For example, the treatment may
and drug prescription databases used for the current study reduce morbidity for angina, congestive heart failure, and chronic
showed a close concordance with either a population-based local heart failure, as well as all-cause morbidity and mortality [9–13].
registry of coronary and cerebrovascular events validated accord- Cost-effectiveness estimates are expected to have a much more
ing to MONICA criteria [32] or data provided by a network of favorable profile than that found in our study, whether these health
Italian general practitioners [33]. Finally, a number of sensitivity benefits of enhancing adherence were also taken into account.
analyses previously performed on the same setting as the current Fourth, adherence with treatment was derived from drug prescrip-
one allowed us to verify the robustness of our findings. For tions, which is the most feasible and widely used method to
example, we previously reported that there was no substantial estimate compliance in large populations [7]. With this method,
variation of the adherence-outcome association by using alter- however, the assumption has to be made that the PDC by a
native PDC categorizations, varying the length of follow-up, using prescription corresponds to the proportion of days of drug use,
alternative censoring criteria, and adjusting the estimates for which may not be invariably the case. Because misclassification of
unmeasured confounders [15,34]. exposure is expected to be independent from the disease status, the
Our study has a number of potential limitations. External reduction of CV risk associated with high levels of adherence might
validity (generalizability) of the findings is a first major concern. be larger than that quantified here [40]. If this was true, adherence-
The costs for BP-lowering drugs are extremely variable across enhancing interventions would become even more cost-effective.
countries and are likely to decrease with increasing competition Finally, because the allocation to levels of adherence to BP-lowering
from generics [35]. Similarly, incidence and mortality for major drugs was not randomized, the results of our study may be affected
CV events is also variable, showing lower values in Italy than in by confounding. Even if we attempted to limit confounding by
other Western countries [36]. Finally, with the aim of limiting adjustment for some demographic, therapeutic, and clinical factors,
sources of heterogeneity of the investigated population by the higher adherence could still be a surrogate for other unmeasured
inclusion of patients likely affected by comorbidities and cotreat- characteristics. The so-called healthy adherer effect might occur by
ments, individuals aged 80 years or older were excluded from our assuming that healthier patients are more likely to adherer to
analysis. As a consequence, our estimates cannot be universally therapy [41]. It should be taken into account, however, that patients
generalized. with worse clinical profile (i.e., those affected by chronic comorbid-
Both the components of the cost-effectiveness estimate are ities and on treatment for hyperlipidaemia and diabetes) are, on the
likely to be affected by sources of systematic uncertainty. As far one hand, at higher CV risk; however, on the other hand, these
as costs are concerned, our analysis did not include those of patients more frequently adhered to BP-lowering therapy than did
interventions’ implementation. A recent review of studies on those with a better clinical profile. This finding is consistent with the
interventions improving adherence to antihypertensive and lipid- existing literature [10] and suggests that the protective effect of
lowering medications identified a variety of different means of adherence on the CV risk, and consequently the benefits in enhan-
improving adherence [37]. These may be broadly classified as cing adherence, might be larger than that estimated by our study if
involving professional input by a doctor, nurse, and/or pharma- the difference in clinical profiles over the strata of adherence was
cist or involving patients’ education regarding their therapies. taken into account.
The most effective approaches for improving adherence were All these potential limitations taken together suggest that out
those based on personalized, intensive interventions closely findings are indicative of only the direction, rather than the exact
monitoring patients’ adherence and involving more than one magnitude, of the trends in costs and effectiveness among
visit during follow-up. A general positive relationship between scenarios of enhancing adherence.
the estimated cost for intervention implementing and its impact In conclusion, our results confirm that the management of
on improving adherence was also reported. This suggests that hypertension is unsatisfactory in the general practice because of
costs of enhancing adherence are likely to be not negligible with quite low level of treatment adherence. Our approach based on
respect to those directly charged for the increased use of drugs. both real-world drug utilization and hospital discharge informa-
We are not able, however, to include the costs of implementation, tion supplies evidence that interventions for enhancing adher-
as well as of the management of hypertension in the population ence might offer important benefits in reducing the risk of
and of increasing the use of general practitioner time. This is hospitalization for major CV outcomes, although with a substan-
because to our best knowledge, no studies have been performed tial cost. Important resources saving might be obtained by
in Italy on this subject. A general worsening of the reported cost- encouraging the use of less expansive medications, such as
effectiveness estimates is expected if other cost sources were generic drugs, and by discouraging the waste of resources for
taken into account. the treatment of patients who do not need drug therapy. In turn,
As far as measuring effectiveness is concerned, the sources of this would free resources to finance interventions of enhancing
uncertainty in measuring the relationship between adherence and adherence. While the final decision on the sustainability of the
outcome should be carefully considered. First, no information was cost-related improvement of therapeutic adherence rests on the
available on the diagnosis of hypertension. Thus, the patients Regional Health Authority, we believe that the interventions
324 VA L U E I N H E A LT H 1 6 ( 2 0 1 3 ) 3 1 8 – 3 2 4

discussed here have potential and should be considered by public [20] Newby LK, LaPointe NM, Chen AY, et al. Long-term adherence to
health decision makers. evidence-based secondary prevention therapies in coronary artery
disease. Circulation 2006;113:203–12.
Source of financial support: This study was funded by grants
[21] Tunstall-Pedoe H, Kuulasmaa K, Amouyel P, et al. Myocardial infarction
from the Italian Minister for University and Research (‘‘Fondo and coronary deaths in the World Health Organization MONICA
d’Ateneo per la Ricerca’’ portion, year 2009). Project: registration procedures, event rates, and case-fatality rates in
38 populations from 21 countries in four continents. Circulation
1994;90:583–612.
R EF E R EN C ES [22] Asplund K, Bonita R, Kuulasmaa K, et al. Multinational comparisons of
stroke epidemiology: evaluation of case ascertainment in the WHO
MONICA stroke study. Stroke 1995;26:355–60.
[23] Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index
[1] Mancia G, De Backer G, Dominiczak A, et al. 2007 Guidelines for the for use with ICD-9-CM administrative databases. J Clin Epidemiol
management of arterial hypertension. J Hypertens 2007;25:1105–87. 1992;45:613–9.
[2] Wolf-Maier K, Cooper RS, Kramer H, et al. Hypertension treatment and [24] Marubini E, Valsecchi MG. Analysing Survival Data from Clinical Trials
control in five European countries, Canada, and the United States. and Observational Studies. New York: John Wiley & Sons, 1995.
Hypertension 2004;43:10–7. [25] Robert CP. Simulation of truncated normal variables. Stat Comput
[3] Kearney PM, Whelton M, Reynolds K, et al. Worldwide prevalence of 1995;5:121–5.
hypertension: a systematic review. J Hypertens 2004;22:11–9. [26] Detsky AS, Laupacis A. Relevance for cost-effectiveness analysis to
[4] Mancia G, Grassi G. Systolic and diastolic blood pressure control in clinicians and policy makers. JAMA 2007;298:221–4.
antihypertensive drug trials. J Hypertens 2002;20:1461–4. [27] Wang H, Zhao H. A study on confidence intervals for incremental cost-
[5] Chobanian AV, Bakris GL, Black HR, et al. National High Blood Pressure effectiveness ratios. Biometr J 2008;50:505–14.
Education Program Coordinating Committee. Seventh report of the [28] Mantovani LG, Fornari C, Madotto F, et al. Burden of acute myocardial
Joint National Committee on Prevention, Detection, Evaluation, and infarction. Int J Cardiol 2011;150:111–2.
Treatment of High Blood Pressure. Hypertension 2003;42:1206–52. [29] Kolominsky-Rabas PL, Heuschmann PU, Marschall D, et al. Lifetime
[6] Weber MA, Julius S, Kjeldsen SE, et al. Blood pressure dependent and cost of ischemic stroke in Germany: results and national projections
independent effects of antihypertensive treatment on clinical events in from a population-based stroke registry: the Erlangen Stroke Project.
the VALUE Trial. Lancet 2004;363:2049–51. Stroke 2006;37:1179–83.
[7] Fitz-Simon N, Bennett K, Feely J. A review of studies of adherence with [30] Van Wijk BLG, Klungel OH, Heerdink ER, de Boer A. Generic substitution
antihypertensive drugs using prescription databases. Ther Clin Risk of antihypertensive drugs: does it affect adherence? Ann Pharmacother
Manag 2005;1:93–106. 2006;40:15–20.
[8] Vrijens B, Vincze G, Kristanto P, et al. Adherence to prescribed [31] Corrao G, Zambon A, Parodi A, et al. Incidence of cardiovascular events
antihypertensive drug treatments: longitudinal study of electronically in Italian patients with early discontinuations of antihypertensive,
compiled dosing histories. BMJ 2008;336:1114–7. lipid-lowering, and antidiabetic treatments. Am J Hypertens
[9] DiMatteo MR, Giordani PJ, Lepper HS, et al. Patient adherence and 2012;25:549–55.
medical treatment outcomes: a meta-analysis. Med Care [32] Corrao G, Zambon A, Nicotra F, et al. Persistence with oral and
2002;40:794–811. transdermal hormone replacement therapy and hospitalisation for
[10] Mazzaglia G, Ambrosioni E, Alacqua M, et al. Adherence to cardiovascular outcomes. Maturitas 2007;57:315–24.
antihypertensive medications and cardiovascular morbidity among [33] Corrao G, Parodi A, Zambon A, et al. Reduced discontinuation of
newly diagnosed hypertensive patients. Circulation 2009;120:1598–605. antihypertensive treatment by two-drug combination as first step:
[11] Kettani FZ, Dragomir A, Côté R, et al. Impact of a better adherence to evidence from daily life practice. J Hypertens 2010;28:1584–90.
antihypertensive agents on cerebrovascular disease for primary [34] Corrao G, Conti V, Merlino L, et al. Results of a retrospective database
prevention. Stroke 2009;40:213–20. analysis of adherence to statin therapy and risk of nonfatal ischemic
[12] Perreault S, Dragomir A, White M, et al. Better adherence to heart disease in daily clinical practice in Italy. Clin Ther 2010;32:300–10.
antihypertensive agents and risk reduction of chronic heart failure. [35] Danzon PM, Furukawa MF. International prices and availability of
J Intern Med 2009;266:207–18. pharmaceuticals in 2005: drugs spending differences reflect availability
[13] Perreault S, Dragomir A, Roy L, et al. Adherence level of and use, not just prices. Health Affairs 2011;27:221–33.
antihypertensive agents in coronary artery disease. Br J Clin Pharmacol [36] Tunstall-Pedoe H, Kuulasmaa K, Tolonen H, et al., for the WHO MONICA
2010;69:74–84. Project. MONICA. In: Tunstall-Pedoe H, ed., Monograph and Multimedia
[14] Dragomir A, Côté R, Roy L, et al. Impact of adherence to Source Book. Genevea: World Health Organization, 2003.
antihypertensive agents on clinical outcomes and hospitalization [37] Chapman RH, Ferrufino CP, Kowal SL, et al. The cost and effectiveness
costs. Med Care 2010;48:418–25. of adherence-improving interventions for antihypertensive and lipid-
[15] Corrao G, Parodi A, Nicotra F, et al. Better compliance to lowering drugs. Int J Clin Pract 2010;64:169–81.
antihypertensive medications reduces cardiovascular risk. J Hypertens [38] Poluzzi E, Strahinja P, Vargiu A, et al. Initial treatment of hypertension
2011;29:610–8. and adherence to therapy in general practice in Italy. Eur J Clin
[16] McDonald HP, Garg A, Haynes RB. Interventions to enhance patient Pharmacol 2005;61:603–9.
adherence to medication prescriptions: scientific review. JAMA [39] Turnbull F. Blood Pressure Lowering Treatment Trialists’ Collaboration.
2002;288:2868–79. Effect of different blood-pressure-lowering regimens on major
[17] Drummond MF, O’Brien B, Stoddart GL, et al. Methods for the Economic cardiovascular events: results of prospectively-designed overviews of
Evaluation of Health Care Programmes (2nd ed.). Oxfordrk: Oxford randomised trials. Lancet 2003;362:1527–35.
University Press, 1997. [40] Copeland KT, Checkoway H, McMichael AJ, Holbrook RH. Bias due to
[18] Corrao G, Cesana G, Merlino L. Pharmacoepidemiological research and misclassification in the estimation of relative risk. Am J Epidemiol
the linking of electronic healthcare databases available in the Italian 1977;105:488–95.
region of Lombardy. BioMed Stat Clin Epidemiol 2008;2:117–25. [41] Brookhart MA, Patrick AR, Dormuth C, et al. Adherence to lipid-
[19] DiMatteo MR, Giordani PJ, Lepper HS, et al. Patient adherence and lowering therapy and the use of preventive health services: an
medical treatment outcomes: a meta-analysis. Med Care investigation of the healthy user effect. Am J Epidemiol
2002;40:794–811. 2007;166:348–54.