CHEST / 137 / 4 / APRIL, 2010 883
adherence to this therapy is poor. 1,2 Because of the
narrow therapeutic range of warfarin, even short
periods of subtherapeutic or supratherapeutic anti-
coagulation may expose patients to recurrent throm-
bosis or dangerous side effects such as intracranial
Periodic monitoring of patients’ international nor-
malized ratio (INR) level is a cornerstone of anti-
arfarin serves as the primary therapy to pre-
vent stroke and venous thromboembolism, but
coagulation therapy. 6-8 However, close monitoring,
even by specialized anticoagulation clinics, may fail to
identify periods of nonadherence until after a throm-
botic or hemorrhagic event has occurred. 9,10 Ideally,
provider assessment and patient self-reporting of
nonadherence would suffi ce to supplement INR
monitoring, but both have recently been shown to be
unreliable. 11 Therefore, our objective was to develop
a novel clinical prediction rule, 12,13 based on demo-
graphic, clinical, and psychosocial factors, that would
Background: Warfarin is the primary therapy to prevent stroke and venous thromboembolism.
Signifi cant periods of nonadherence frequently go unreported by patients and undetected by
providers. Currently, no comprehensive screening tool exists to help providers assess the risk of
nonadherence at the time of initiation of warfarin therapy.
Methods: This article reports on a prospective cohort study of adults initiating warfarin therapy
at two anticoagulation clinics (university- and Veterans Affairs-affi liated). Nonadherence, defi ned
by failure to record a correct daily pill bottle opening, was measured daily by electronic pill cap
monitoring. A multivariable logistic regression model was used to develop a point system to pre-
dict daily nonadherence to warfarin.
Results: We followed 114 subjects for a median of 141 days. Median nonadherence of the partici-
pants was 14.4% (interquartile range [IQR], 5.8-33.8). A point system, based on nine demographic,
clinical, and psychosocial factors, distinguished those demonstrating low vs high levels of nonad-
herence: four points or fewer, median nonadherence 5.8% (IQR, 2.3-14.1); fi ve points, 9.1% (IQR,
5.9-28.6); six points, 14.5% (IQR, 7.1-24.1); seven points, 14.7% (IQR, 7.0-34.7); and eight points
or more, 29.3% (IQR, 15.5-41.9). The model produces a c -statistic of 0.66 (95% CI, 0.61-0.71),
suggesting modest discriminating ability to predict day-level warfarin nonadherence.
Conclusions: Poor adherence to warfarin is common. A screening tool based on nine demographic,
clinical, and psychosocial factors, if further validated in other patient populations, may help to
identify groups of patients at lower risk for nonadherence so that intensifi ed efforts at increased
monitoring and intervention can be focused on higher-risk patients. CHEST 2010; 137(4):883–889
Abbreviations: INR 5 international normalized ratio; IN-RANGE study 5 International Normalized Ratio Adherence
and Genetics study; IQR 5 interquartile range; MEMS 5 medication event monitoring system; PRIME 5 Program for
the Reduction in Medical Errors; SF-36 5 Short Form-36
Can We Predict Daily Adherence to
Results From the International Normalized Ratio
Adherence and Genetics (IN-RANGE) Study
Alec B. Platt , MD , MSCE ; A. Russell Localio , PhD ; Colleen M. Brensinger , MS ;
Dean G. Cruess , PhD ; Jason D. Christie , MD , MSCE , FCCP ; Robert Gross , MD , MSCE ;
Catherine S. Parker , MD , MS ; Maureen Price , RN ; Joshua P. Metlay , MD , PhD ;
Abigail Cohen , PhD ; Craig W. Newcomb , MAR ; Brian L. Strom , MD , MPH ;
Mitchell S. Laskin , RPh ; and Stephen E. Kimmel , MD , MSCE
Data on patient demographic and clinical factors at the time of
study entry were obtained by trained research personnel through
the use of standardized questionnaires. Baseline demographic
and clinical factors were collected and consisted of age, gender,
race, education level, employment status, marital status, indica-
tion for therapy, warfarin use prior to the current indication, med-
ical comorbidities, tobacco smoking status, alcohol consumption,
a listing of all medications taken in the week before study enroll-
ment, and participants’ health-care utilization and access to health
care 17 in the preceding 12 months. At baseline, participants also
completed standardized questionnaires on psychosocial factors
hypothesized to predict poor adherence: the Short Form-36
(SF-36) mental component subscale, standard version, 18 to rate
their quality of life and the Cognitive Capacity Screening Exami-
nation to test for the presence of cognitive dysfunction. 19 We
included in the model only factors that would be available to the
clinician at the time of the initial evaluation.
Our primary outcome, warfarin nonadherence, was measured
daily using an electronic medication event monitoring system
(MEMS) (MEMS caps; AARDEX; Zug, Switzerland). The MEMS
cap records the time and date when a participant opens the medi-
cation container. Each patient-day observed on warfarin therapy
was judged either “adherent” or “nonadherent”: an adherent day
occurred if exactly one electronic pill cap opening was recorded
during a given 24-h period; a nonadherent day occurred if a par-
ticipant either failed to actuate the MEMS cap container during a
24-h period or did so twice or more (as warfarin was prescribed
exclusively as once daily), or an opening occurred when the
participant had been specifi cally instructed to “hold” a dose for
We fi tted a MEMS cap directly on the pill bottle of those par-
ticipants who accessed their warfarin medication directly from
the pill bottle. Participants who used 7-day pillbox reminders to
administer all their medications were provided an empty bottle
with a MEMS cap attached and then instructed to open and close
the MEMS cap bottle each time they took their warfarin, thereby
creating an electronic diary of warfarin usage.
Using the patient-day as the unit of analysis, we developed
a prognostics model for the probability of nonadherence of a
patient on any given day using standard methods. We also imple-
mented, as a sensitivity analysis, a model that ignored the variation
in follow-up and considered only patient-level data.
We initially checked for correlation and for missing levels
among all potential predictors to arrive at complete, independent
factors. For example, we were forced to combine the variables of
employment status and age as they were overlapping (older sub-
jects tended to be unemployed). The self-reported household
income level was missing too often for analysis. Subjects with
missing questionnaire data (n 5 22) could not be used in the fi nal
analysis because we had too little information with which to
impute those questionnaire responses.
During our initial stages of model building, we used a multi-
variable logistic regression model with generalized estimating
equations (GENMOD; SAS, Inc.; Cary, NC) and an independence
working correlation structure to account for the repeated, dichot-
omous outcome of daily warfarin adherence in the 114 subjects in
the fi nal model. 20,21 This approach allowed us to assess accurately
the statistical signifi cance of each potential risk factor while taking
into consideration that we had many repeated observations of the
identify at the outset of therapy those subjects at high
and low risk for nonadherence to warfarin.
Clinical prediction rules have been developed in
other areas of medicine 14,15 to assist in medical deci-
sion making. However, we are not aware of previous
efforts to develop a clinical prediction rule in the
area of warfarin nonadherence or other high-risk
areas of medication nonadherence where even
short periods of nonadherence may have important
clinical consequences. Such a screening tool for
warfarin, if properly validated in other patient pop-
ulations and settings, could help direct programs of
intensifi ed counseling and monitoring for high-risk
Materials and Methods
Design and Study Population
Details of the study have been previously reported. 16 Briefl y,
the INR Adherence and Genetics (IN-RANGE) study, and its
component study, Program for the Reduction in Medical Errors
(PRIME), was designed as a prospective cohort study in which
patients aged 21 years and older with a target INR of 2.0 to 3.0
were recruited within 2 months of initiating anticoagulation ther-
apy. From April 2002 until April 2006, participants were enrolled
and followed at two specialized anticoagulation clinics: the Hospi-
tal of the University of Pennsylvania and the Philadelphia Veter-
ans Affairs Medical Center, both in Philadelphia, Pennsylvania.
The institutional review boards of the participating hospitals
approved the study, and all participants provided informed,
Manuscript received January 6, 2009; revision accepted October 9,
Affi liations: From the Center for Health Equity Research and
Promotion (Dr Platt) and (Drs Gross and Metlay), Philadelphia
Veterans Affairs Medical Center, Philadelphia, PA; the Reading
Hospital and Medical Center (Dr Platt), Reading, PA; the Center
for Clinical Epidemiology and Biostatistics, Department of
Biostatistics and Epidemiology (Drs Localio, Christie, Gross,
Metlay, Cohen, Strom, and Kimmel, Mss Brensinger and Price,
and Mr Newcomb) and Department of Medicine (Drs Christie,
Gross, Metlay, and Strom), University of Pennsylvania School of
Medicine, Philadelphia, PA; the Center for Education and
Research on Therapeutics (Drs Localio, Gross, Metlay, and
Strom), University of Pennsylvania, Philadelphia, PA; the Depart-
ment of Psychology (Dr Cruess), University of Connecticut,
Storrs, CT; the Department of Internal Medicine (Dr Parker),
Beth Israel Deaconess Medical Center, Boston, MA; and the
Depart ment of Pharmacy Service (Mr Laskin), Hospital of the
University of Pennsylvania, Philadelphia, PA.
Funding /Support: The IN-RANGE study was supported by grants
from the National Institutes of Health [R01-HL66176] and the
Agency for Health Research and Quality [P01-HS11530]. Dr Kimmel
was also supported by P20RR020741 and K24HL070936.
Correspondence to: Stephen E. Kimmel, MD, MSCE,
University of Pennsylvania School of Medicine, 707 Blockley
Hall, 423 Guardian Dr, Philadelphia, PA 19104-6021; e-mail:
© 2010 American College of Chest Physicians. Reproduction
of this article is prohibited without written permission from the
American College of Chest Physicians ( www.chestpubs.org/
CHEST / 137 / 4 / APRIL, 2010 885
The study consists of 114 persons with com-
plete psychosocial evaluation undergoing chronic
anticoagulation followed for a median of 141 days
( Fig 1 ). Warfarin nonadherence occurred in 4,793 of
Figure 1. Participant enrollment and protocol completion.
Table 1— Characteristics of 114 Study Participants
Age, y, mean (range)
Sex, male, No. (%)
African-American, No. (%) a
Education greater than high school, No. (%)
Employment/age, No. (%)
Disabled, , 55 y
Disabled, ? 55 y
Smoking status, No. (%)
Marital status, No. (%)
Insurance status, No. (% )
Indication, No. (%)
Atrial fi brillation
Dilated cardiomyopathy/LV thrombus
University Hospital site, No. (%)
History, No. (%)
Prior warfarin use
Cardiac arrhythmia, heart failure, ischemia, other
Peptic ulcer disease, gastritis
Abnormal bleeding, abnormal clotting, DVT, PE
Other comorbidities b
Number of medications taken at baseline,
SF-36 MCS score, mean (range)
General health status, No. (%)
Cognitive dysfunction, No. (%)
CCSE score , 20
CCSE score ? 20
CCSE 5 Cognitive Capacity Screening Examination; LV 5 left
ventricle; PE 5 pulmonary embolism; SF-36 MCS 5 short form-36
mental component subscale; TIA 5 transient ischemic attack;
VA 5 Veterans Affairs.
a Race was self-reported.
b Other comorbidities: hypothyroidism, hyperthyroidism, cancer, and
same patients. Nonadherence to warfarin for a given patient on a
given day was the dependent variable in our analysis. As use of the
patient-day as the unit of analysis accounted for the variation in
the number of days of follow-up by participants, we allowed the
number of patient-days on warfarin to vary by patient. 22 To pro-
vide estimates of variance and confi dence bounds, we performed
bootstrap resampling (at the patient level) and we reported bias-
corrected estimates. 23,24 We bootstrapped to avoid the risk of bias
in variance estimates from less than large samples of data. 25 This
modeling approach ensures that confi dence bounds are not overly
Beginning with factors considered for inclusion in the explana-
tory model, 27 we selected prognostic factors based on their ability
to improve the c -statistic, which was used as a measure of the abil-
ity to classify patients correctly. 28,29 We then assigned points to
each factor based on the relative sizes of the regression coeffi -
cients (scaled relative to the smallest coeffi cient and then rounded
to the nearest whole number), after which we reassessed model
discrimination and calibration (using c -statistics and comparisons
of observed vs expected percent adherence). We examined the
distribution of days of nonadherence across patients to determine
whether individual patients were infl uential to the fi nal estimates.
After obtaining a full model of predictive factors, we employed
bootstrap resampling to internally validate the model and check
whether the fi nal model was an overly optimistic assessment of
model predictive capabilities. 28
To choose an optimal cut-off for defi ning those at greatest risk
of nonadherence, a range of thresholds was reviewed using clini-
cal criteria and the tradeoff between the number of false positives
and false negatives. Owing to the greater cost of false negatives
(ie, failing to identify patients at risk for nonadherence), the
threshold chosen was prespecifi ed as being relatively low in the
point scale, even at a cost of a higher number of false positives (ie,
misclassifying some subjects as being at higher risk for nonadher-
ence when they are not at increased risk). After developing our
model on the level of patient-days on warfarin therapy, we applied
the IN-RANGE point score to our cohort at the patient level to
assess how the IN-RANGE model predicts nonadherence over
the entire course of a patients’ therapy on warfarin. This is the
level at which the point score model could be used clinically to
screen groups of patients for nonadherence. All statistical analyses
were performed using SAS v 9.1(SAS Institute) and Stata v 10.1
(StataCorp; College Station, TX).
with 1.0 representing no overoptimism), suggesting
that our model might be generalized to other samples
of patients. 28
Table 3 displays the range of sensitivity, specifi city,
and positive and negative predictive value cut-offs
available based on the IN-RANGE point system for
nonadherence performed on a patient-day level anal-
ysis. Adoption of a fi ve point threshold to dichoto-
mize between adherent subjects (fi ve points or fewer)
vs nonadherent subjects (six or more points) results
in a sensitivity of 81.3% (95% CI,70.1-89.5), specifi c-
ity of 38.3% (26.0-51.0), positive predictive value of
26.3% (21.5-31.1), and a negative predictive value of
88.3% (83.5-91.9). Selection of higher total point
thresholds results in a modest lowering of the nega-
tive predictive value but only a slight improvement in
the positive predictive value.
By contrast, when the IN-RANGE prediction
model is applied to the level of individual participants
(as opposed to the patient-days on warfarin), subjects
are more easily classifi ed into higher and lower risk
strata. For example, subjects with an IN-RANGE
point score of 4 points or fewer had a median level of
nonadherence of 5.8% (IQR, 2.3-14.1) compared
with median nonadherence for the entire cohort of
14.4% (IQR, 5.8-33.8). Subjects scoring higher on
the IN-RANGE point score prediction model dem-
onstrated progressively higher median nonadherence
rates ( Table 4 ).
In what we believe is the fi rst prospective cohort
study designed to develop a comprehensive predic-
tion model for warfarin nonadherence using objec-
tive electronic recording of daily pill taking, we
developed a prognostic model based on a total of nine
demographic, clinical, and psychosocial factors. The
purpose of this IN-RANGE model is to help provid-
ers identify groups of subjects with clinically mean-
ingful differences in nonadherence. In our cohort,
we found that subjects with an IN-RANGE point
score of 4 or fewer have a median level of nonadher-
ence of 5.8% (IQR, 2.3-14.1), as compared with the
overall level of nonadherence of 14.4% (IQR, 5.8-
33.8) in the entire cohort and nonadherence of 29.3%
(IQR, 15.5-41.9) for those with 8 points or greater.
High-scoring subjects, those who are at highest risk
for nonadherence, might form a target population for
increased monitoring and education, whereas those
with lower scores could require less intensive moni-
toring or counseling.
When the model is applied to predicting nonadher-
ence on the patient-day level of successive, discrete
days on warfarin (as compared with the patient level),
22,492 or 21.3% (95% CI, 17.6-25.6) of patient-days
observed. An additional 22 subjects who were not
included in the analysis because of failure to complete
all of the psychosocial questionnaires had a somewhat
higher rate of warfarin nonadherence (29% vs 21% of
patient-days observed, P 5 .08). When analyzed on
the level of the patient (as opposed to the patient-day
level), median nonadherence for the 114 subjects in
the cohort was 14.4% (interquartile range [IQR)],
5.8-33.8) and mean nonadherence was 21.2%
(SD 6 19.1%).
Baseline demographic data are presented in Table 1 .
The mean age of the cohort was 55 years old, and it
consisted of two-thirds men and 54% African-Americans.
Atrial fi brillation and venous thromboembolism were
the main indications for warfarin therapy. Using
baseline variables previously shown to be associated
with poor adherence to warfarin, 27 we developed a
predictive model to identify those at highest risk for
daily nonadherence to warfarin ( Table 2 ). The model
produced a c -statistic of 0.66 (95% CI, 0.61-0.71). A
check for overoptimism using bootstrap resampling
reaffi rmed this c -statistic (shrinkage factor of 0.993,
Table 2— Predictors of Warfarin Nonadherence for 114
Predictor a b -Coeffi cientPoints
Education . high school
Disabled , 55 y
Disabled ? 55 y
No history of hypertension
Prior history of warfarin therapy, present
Cognitive functioning, impaired
(CCSE , 20)
SF-36 MCS score
Very low ( , 25)
Below average (35-44)
Above average (55-64)
Very high ( ? 75)
Baseline general health status
Ref 5 reference group. See Table 1 for expansion of abbreviations.
a Area under curve 5 0.66 (95% CI 0.61, 0.71); R 2 (regression
cor relation coefficient) for model 5 0.319; regression coefficient
constant 5 2 3.236.
CHEST / 137 / 4 / APRIL, 2010 887
aware of prior attempts at developing a clinical pre-
diction rule for nonadherence to warfarin or other
high-risk medications. Nonetheless, the IN-RANGE
prediction model, if adequately validated and refi ned
in other settings with different levels of medication
nonadherence, might help providers target important
interventions toward a group of patients who at base-
line are at increased risk of poor adherence.
The strengths of the IN-RANGE prediction model
are the prospective design; the thorough evaluation of
candidate demographics, clinical variables, and psy-
chosocial variables; and the electronic recording of
warfarin nonadherence. Limitations include, fi rst, the
lack of a validation cohort to test our prognostic model
in other settings. Our study, performed in an urban,
academic-based setting, will need to be replicated in
other patient populations and in groups with differing
prevalences of nonadherence to ensure generalizabil-
ity. However, as a check for internal validation, we did
perform bootstrap resampling and found there to be
no evidence of overoptimism in the estimates of vari-
ance. Second, there may be selection bias based on
the willingness of subjects to consent to electronic pill
cap monitoring, even though we found no signifi cant
differences in baseline demographic factors of study
subjects who agreed vs those who declined MEMS
caps. A third limitation is that the MEMS cap actua-
tion does not guarantee the patient actually consumed
the pill upon each occasion of bottle opening. Never-
theless, a close link exists between nonadherence to
warfarin and out-of-range INR values using MEMS
cap monitoring. 1 Fourth, administration of two ques-
tionnaires, the SF-36 mental component subscale and
the Cognitive Capacity Screening Exam, might make
implementation of the IN-RANGE prediction model
cumbersome in some settings, although these ques-
tionnaires are freely available and can be completed
in minutes. Further work examining other methods of
assessing these parameters (eg, shorter versions of the
SF-36) would be useful. Fifth, a priori , we limited
our model to the examination of baseline factors that
could be ascertained by clinicians at the time of treat-
ment initiation. Prior research has not identifi ed risk
factors for nonadherence that occur after initiating
therapy with warfarin. 27
the model’s ability to discriminate is more limited.
The modest discrimination on the patient-day level is
refl ected in the model’s c -statistic of 0.66 and the
relatively low positive predictive values over a range
of point scores (see Table 3 ). As a result, when view-
ing nonadherence on the patient-day level, there is
frequent misclassifi cation of patients with more adher-
ent days as being at higher risk for nonadherence.
The frequent misclassifi cation associated with the
model’s low positive predictive value would be an
important limitation if any planned intervention
involved signifi cant risk to the patient or a high cost
to the provider. However, if the potential interven-
tions posed minimal or no risk to the patient and were
relatively inexpensive (such as more intensifi ed coun-
seling or monitoring), then the low positive predic-
tive value might not be an important limitation.
Moreover, given the life-threatening nature and
high economic cost to the health system of a stroke,
hemorrhage, or recurrent venous thrombosis, the
overall cost-benefi t ratio might favor adoption of a
prediction screening tool, even at the cost of provid-
ing counseling or monitoring to a somewhat wider
group of patients. 30,31
Despite a comprehensive, prospective evaluation of
demographic, clinical, and psychosocial variables, we
were only able to arrive at a model with modest dis-
crimination ability. This result may point to a larger
issue of the diffi culty in predicting the behavior of
individuals in their daily medication usage, as has been
previously observed. 32-35 Furthermore, we are not
Table 3— Range of Sensitivity, Specifi city, and Predictive Values for Day-Level Analysis of Warfarin Nonadherence
(N 5 114 patients)
Cumulative Days, No. (%)
Specifi city, %
Value, % ( 95% CI)
Value, % (95% CI) NonadherentAdherent
? 5 points
? 7 points
? 8 points
CIs represent bias corrected estimates based on bootstrap resampling.
Table 4— Patient Level of Nonadherence to Warfarin,
by International Normalized Ratio Adherence and
Genetics Study Point Score (N 5 114 subjects)
% Incorrect Warfarin Doses
? 4 points, n 5 26
5 points, n 5 16
6 points, n 5 15
7 points, n 5 24
? 8 points, n 5 33
Summary for cohort, n 5 114
Affairs Medical Center. We wish to thank Mabel Chin, PharmD,
for her dedication to our fi eld work.
1 . Kimmel SE , Chen Z , Price M , et al . The infl uence of
patient adherence on anticoagulation control with warfarin:
results from the International Normalized Ratio Adherence
and Genetics (IN-RANGE) Study . Arch Intern Med .
2007 ; 167 ( 3 ): 229 - 235 .
2 . Waterman AD , Milligan PE , Bayer L , Banet GA , Gatchel SK ,
Gage BF . Effect of warfarin nonadherence on control of the
International Normalized Ratio . Am J Health Syst Pharm .
2004 ; 61 ( 12 ): 1258 - 1264 .
3 . Hylek EM , Skates SJ , Sheehan MA , Singer DE . An analysis
of the lowest effective intensity of prophylactic anticoagula-
tion for patients with nonrheumatic atrial fi brillation . N Engl
J Med . 1996 ; 335 ( 8 ): 540 - 546 .
4 . Hylek EM , Singer DE . Risk factors for intracranial hem-
orrhage in outpatients taking warfarin . Ann Intern Med .
1994 ; 120 ( 11 ): 897 - 902 .
5 . Laupacis A , Albers G , Dalen J , Dunn MI , Jacobson AK ,
Singer DE . Antithrombotic therapy in atrial fi brillation . Chest .
1998 ; 114 ( suppl 5 ): 579S - 589S .
6 . Singer DE , Albers GW , Dalen JE , et al ; American College
of Chest Physicians . Antithrombotic therapy in atrial fi brilla-
tion: American College of Chest Physicians evidence-based
clinical practice guidelines. 8th edition . Chest . 2008 ; 133 ( suppl 6) :
546S - 592S .
7 . Ansell J , Hirsh J , Poller L , Bussey H , Jacobson A , Hylek E .
The pharmacology and management of the vitamin K antago-
nists: the Seventh ACCP Conference on Antithrombotic and
Thrombolytic Therapy . Chest . 2004 ; 126 ( suppl 3 : 204S - 233S
[erratum: Chest. 2005;127(1)415-416] .
8 . Samsa GP , Matchar DB , Goldstein LB , et al . Quality of antico-
agulation management among patients with atrial fi brillation:
results of a review of medical records from 2 communities .
Arch Intern Med . 2000 ; 160 ( 7 ): 967 - 973 .
9 . Palareti G , Manotti C , DAngelo A , et al . Thrombotic events
during oral anticoagulant treatment: results of the inception-
cohort, prospective, collaborative ISCOAT study: ISCOAT
study group (Italian Study on Complications of Oral
Anticoagulant Therapy) . Thromb Haemost . 1997 ; 78 ( 6 ): 1438 -
10 . Palareti G , Leali N , Coccheri S , et al ; Italian Study on
Complications of Oral Anticoagulant Therapy . Bleeding
complications of oral anticoagulant treatment: an inception-
cohort, prospective collaborative study (ISCOAT) . Lancet .
1996 ; 348 ( 9025 ): 423 - 428 .
11 . Parker CS , Chen Z , Price M , et al . Adherence to warfarin
assessed by electronic pill caps, clinician assessment, and
patient reports: results from the IN-RANGE study . J Gen
Intern Med . 2007 ; 22 ( 9 ): 1254 - 1259 .
12 . Laupacis A , Sekar N , Stiell IG . Clinical prediction rules.
A review and suggested modifi cations of methodological stan-
dards . JAMA . 1997 ; 277 ( 6 ): 488 - 494 .
13 . McGinn TG , Guyatt GH , Wyer PC , Naylor CD , Stiell IG ,
Richardson WS ; Evidence-Based Medicine Working Group .
Users’ guides to the medical literature: XXII: how to use
articles about clinical decision rules . JAMA . 2000 ; 284 ( 1 ):
79 - 84 .
14 . Auleley GR , Ravaud P , Giraudeau B , et al . Implementation of
the Ottawa ankle rules in France. A multicenter randomized
controlled trial . JAMA . 1997 ; 277 ( 24 ): 1935 - 1939 .
15 . van Belle A , Büller HR , Huisman MV , et al ; Christopher
Study Investigators . Effectiveness of managing suspected
In summary, patient adherence to warfarin is poor,
with one in fi ve doses taken incorrectly, even in the
setting of an anticoagulation clinic. Periodic INR
monitoring is crucial but may fail to capture periods
of nonadherence until after an adverse event has
occurred. A screening tool that provides some incre-
mental benefi t in identifying poor adherence over
simple INR monitoring or provider and patient
assessments could be useful clinically. If further vali-
dated in other populations, the IN-RANGE prediction
model could be used to help exclude a subgroup of
subjects at lowest risk for poor adherence, so that
efforts at increased monitoring and intervention can
be focused on those at higher risk for nonadherence.
Author contributions: Dr Kimmel had full access to all of the
data in the study and takes responsibility for the integrity of the
data and the accuracy of the data analysis.
Dr Platt: contributed to data analysis and manuscript preparation
Dr Localio: contributed to study design, data analysis, and manu-
script preparation and revision.
Ms Brensinger: contributed to data analysis and manuscript prep-
aration and revision.
Dr Cruess: contributed to study design, data analysis, and manu-
script preparation and revision.
Dr Christie: contributed to study design, data analysis, and manu-
script preparation and revision.
Dr Gross: contributed to study design, data analysis, and manu-
script preparation and revision.
Dr Parker: contributed to data analysis.
Ms Price: contributed to data collection and data analysis.
Dr Metlay: contributed to study design and funding.
Dr Cohen: contributed to data collection and manuscript prepara-
tion and revision.
Mr Newcomb: contributed to data analysis.
Dr Strom: contributed to study design, funding, data analysis, and
manuscript preparation and revision.
Mr Laskin: contributed to study design, data collection, data anal-
ysis, and manuscript preparation and revision.
Dr Kimmel: contributed to study design, funding, data collection,
data analysis, and manuscript preparation and revision.
Financial/ nonfi nancial disclosures: The authors have reported
to CHEST the following confl icts of interest: Dr Kimmel has
served as a consultant and/or received research funding from
several pharmaceutical companies, including Pfi zer, Merck,
Novartis, GlaxoSmithKline, Centocor, and Bayer, all unrelated to
this paper. Dr Kimmel has received investigator-initiated research
funding from the National Institutes of Health, the Agency for
Health Research and Quality, and the Aetna Foundation for
warfarin research. Dr Gross has served as a consultant and/or
received research funding from GlaxoSmithKline and Bristol-
Myers Squibb, and Dr Strom has served as a consultant to Bristol-
Myers Squibb and other pharmaceutical companies, all unrelated
to this paper. Drs Platt, Localio, Cruess, Christie, Parker, Metlay,
and Cohen; Mss Brensinger and Price; and Messrs Newcomb and
Laskin have reported that no potential confl icts of interest exist
with any companies/organizations whose products or services may
be discussed in this article.
Role of sponsors: The National Institutes of Health and the
Agency for Health Research and Quality, which provided grants,
had no role in the design and conduct of the study; the collection,
management, analysis, and interpretation of the data; or the
preparation, review, and approval of the manuscript.
Other contributions: This study was conducted at the Hospital
of the University of Pennsylvania and the Philadelphia Veterans
www.chestpubs.org Download full-text
CHEST / 137 / 4 / APRIL, 2010 889
pulmonary embolism using an algorithm combining clinical
probability, D-dimer testing, and computed tomography .
JAMA . 2006 ; 295 ( 2 ): 172 - 179 .
16 . Platt AB , Localio AR , Brensinger CM , et al . Risk factors for
nonadherence to warfarin: results from the IN-RANGE study .
Pharmacoepidemiol Drug Saf . 2008 ; 17 ( 9 ): 853 - 860 .
17 . Medical Expenditure Panel Survey (MEPS). FAMES Panel
1 Round 2 Access to Care (AC) Section February 15, 1998 .
questionnaires.jsp . Accessed November 26, 2007.
18 . Ware JE Jr , Sherbourne CD . The MOS 36-item short-form
health survey (SF-36). I. Conceptual framework and item
selection . Med Care . 1992 ; 30 ( 6 ): 473 - 483 .
19 . Jacobs JW , Bernhard MR , Delgado A , Strain JJ . Screening
for organic mental syndromes in the medically ill . Ann Intern
Med . 1977 ; 86 ( 1 ): 40 - 46 .
20 . Pepe MS , Anderson GL . A cautionary note on inference
for marginal regression models with longitudinal data and
general correlated response data Commun Statist Simula.
1994 ; 23 ( 4 ): 939 - 951 .
21 . Diggle PJ , Liang K-Y , Zeger SL . Analysis of Longitudinal
Data . Oxford, England : Clarendon Press ; 1994 .
22 . Pepe MS . The Statistical Evaluation of Medical Tests for
Classifi cation and Prediction . Oxford, England : Oxford
University Press ; 2003 .
23 . Altman DG , Andersen PK . Bootstrap investigation of the
stability of a Cox regression model . Stat Med . 1989 ; 8 ( 7 ):
771 - 783 .
24 . Mick R , Ratain MJ . Bootstrap validation of pharmacodynamic
models defi ned via stepwise linear regression . Clin Pharmacol
Ther . 1994 ; 56 ( 2 ): 217 - 222 .
25 . Davison AC , Hinkley DV . Bootstrap Methods and Their
Application. Cambridge, England : Cambridge University Press ;
26 . Efron B , Tibshirani RJ . An Introduction to the Bootstrap .
New York, NY : Chapman and Hall ; 1993 .
27 . Platt AB , Localio AR , Brensinger CM , et al . Risk factors
for nonadherence to warfarin: results from the IN-RANGE
study. Pharmacoepidemiol Drug Saf . 2008 ; 17 ( 9 ): 853 - 860 .
28 . Harrell FE . Regression Modeling Strategies: With
Applications to Linear Models, Logistic Regression, and
Survival Analysis . New York, NY : Springer-Verlag New York,
Inc. ; 2001 .
29 . Pepe MS , Janes H , Longton G , Leisenring W , Newcomb P .
Limitations of the odds ratio in gauging the performance of a
diagnostic, prognostic, or screening marker . Am J Epidemiol .
2004 ; 159 ( 9 ): 882 - 890 .
30 . You JH , Chan FW , Wong RS , Cheng G . The potential clinical
and economic outcomes of pharmacogenetics-oriented man-
agement of warfarin therapy—a decision analysis . Thromb
Haemost . 2004 ; 92 ( 3 ): 590 - 597 .
31 . Sullivan PW , Arant TW , Ellis SL , Ulrich H . The cost effec-
tiveness of anticoagulation management services for patients
with atrial fi brillation and at high risk of stroke in the US .
Pharmacoeconomics . 2006 ; 24 ( 10 ): 1021 - 1033 .
32 . Osterberg L , Blaschke T . Adherence to medication . N Engl J
Med . 2005 ; 353 ( 5 ): 487 - 497 .
33 . Coronary Drug Project Research Group. Infl uence of adher-
ence to treatment and response of cholesterol on mortality in the
coronary drug project . N Engl J Med . 1980 ; 303 ( 18 ): 1038 - 1041 .
34 . Horwitz RI , Viscoli CM , Berkman L , et al . Treatment adher-
ence and risk of death after a myocardial infarction . Lancet .
1990 ; 336 ( 8714 ): 542 - 545 .
35 . Granger BB , Swedberg K , Ekman I , et al ; CHARM inves-
tigators . Adherence to candesartan and placebo and out-
comes in chronic heart failure in the CHARM programme:
double-blind, randomised, controlled clinical trial . Lancet .
2005 ; 366 ( 9502 ): 2005 - 2011 .