For Expert_Researcher only

43394330.pdf

Home >Psychology homework help >For Expert_Researcher only

Prospective Validation of Eight Different Adherence Measures for Use with Administrative Claims Data among Patients with Schizophreniavhe_543 989..995

Sudeep Karve, BPharm, MS,1 Mario A. Cleves, PhD,2 Mark Helm, MD,3 Teresa J. Hudson, PharmD,4

Donna S. West, RPh, PhD,5 Bradley C. Martin, PharmD, PhD6

1Department of Pharmacy Administration, College of Pharmacy, The Ohio State University, Columbus, OH, USA; 2Department of Pediatrics, College of Medicine, University of Arkansas for Medical Sciences, Little Rock, AR, USA; 3Department of Pharmacy Practice, College of Pharmacy, University of Arkansas for Medical Sciences, Little Rock, AR, USA; 4Department of Psychiatry, College of Medicine, University of Arkansas for Medical Sciences, Little Rock, AR, USA; 5Department of Pharmacy Administration, College of Pharmacy, The University of Mississippi, University, MS, USA; 6Division of Pharmaceutical Evaluation and Policy, Department of Pharmacy Practice, University of Arkansas for Medical Sciences, Little Rock, AR, USA

A B S T R AC T

Objective: The aim of this study was to compare the predictive validity of eight different adherence measures by studying the variability ex- plained between each measure and hospitalization episodes among Medicaid-eligible persons diagnosed with schizophrenia on antipsychotic monotherapy. Methods: This study was a retrospective analysis of the Arkansas Med- icaid administrative claims data. Continuously eligible adult schizophrenia (ICD-9-CM = 295.**) patients on antipsychotic monotherapy were iden- tified in the recruitment period from July 2000 through April 2004. Adherence rates to antipsychotic therapy in year 1 were calculated using eight different measures identified from the literature. Univariate and multivariable logistic regression models were used to prospectively predict all-cause and mental health-related hospitalizations in the follow-up year. Results: Adherence rates were computed for 3395 schizophrenic patients with a mean age of 42.9 years, of which 52.5% (n = 1782) were females,

and 52.8% (n = 1793) were white. The proportion of days covered (PDC) and continuous measure of medication gaps measures of adher- ence had equal C-statistics of 0.571 in predicting both all-cause and mental health-related hospitalizations. The medication possession ratio (MPR) continuous multiple interval measure of oversupply were the second best measures with equal C-statistics of 0.568 and 0.567 for any-cause and mental health-related hospitalizations. The multivariate adjusted models had higher C-statistics but provided the same rank order results. Conclusions: MPR and PDC were among the best predictors of any-cause and mental health-related hospitalization, and are recom- mended as the preferred adherence measures when a single measure is sought for use with administrative claims data for patients not on polypharmacy. Keywords: adherence, pharmacy claims, schizophrenia, validation.

Introduction

The International Society for Pharmacoeconomics and Outcomes Research Medication Compliance and Persistence Work Group defines medication adherence as “the extent to which a patient acts in accordance with the prescribed interval and dose of a dosing regimen” [1]. Medication nonadherence is commonly associated with adverse health conditions and increased eco- nomic burden to the health-care system [2–6], and is a critical issue especially in case of chronic therapies such as schizophre- nia. Adherence measures include direct and indirect techniques. Direct methods include biological assays, whereas indirect methods consist of pill counts, electronic monitors, and use of administrative database claims [7]. Indirect methods and, in particular, use of administrative data are becoming increasingly popular because they afford ease of use and are considered eco- nomical; however, there are no clear standards for measuring or calculating adherence with these indirect approaches.

Medication nonadherence rates in patients suffering from schizophrenia range from 20% to 89% based on the adherence

definition used [8]. Adherence measures using pharmacy claims data have been used to predict health-care cost and utilization [9–11]. Svarstad et al. found that persons with schizophrenia with poorer adherence had higher hospitalization rates than their more adherent counterparts. Additional studies have reported similar findings in persons with schizophrenia [5,6]. Non- adherence to the prescribed treatment may account for 40% of the rehospitalizations associated with schizophrenia [8]. Thieda et al. conducted a literature review for the years 1995 through 2002 evaluating the relationship between compliance and the economic costs of schizophrenia [4]. The authors concluded that lower compliance was associated with adverse outcomes such as increased relapse rates with associated costs ranging from $10,000 to $26,000. Bearing in mind the economic outcomes associated with medication nonadherence, schizophrenia serves as a favorable condition to determine the predictive validity of various adherence measures.

Because of their usefulness in recent years, administrative claims data have been one of the most commonly used sources for calculating medication adherence. Medication adherence measured using pharmacy claims has been validated using other adherence measures such as patient reports, pill counts, question- naires, and interviews [12–17]. Despite these validation studies, there are no standards for the mathematical calculation of adher- ence using claims data. A systematic review by Andrade et al. [18] identified 136 studies that employed administrative claims

Address correspondence to: Bradley C. Martin, Division of Pharmaceuti- cal Evaluation and Policy, Department of Pharmacy Practice, University of Arkansas for Medical Sciences, 4301 W. Markham Street, Slot 522, Little Rock, AR 72205-7122, USA. E-mail: [email protected]

10.1111/j.1524-4733.2009.00543.x

Volume 12 • Number 6 • 2009 V A L U E I N H E A L T H

mailto:[email protected]

to calculate medication adherence and persistence. About 57% of the studies considered medication possession ratio (MPR) and related measures,10% used medication gaps, and 43% used switching and discontinuation in calculating medication adher- ence and persistence. Even among the 57% that considered MPR and related measures, the follow-up period definitions varied, ranging from a specified follow-up period (e.g., 1 year) to the period between the first and last refill. This study emphasized the lack of consensus among definitions and methods used to calcu- late adherence using administrative claims. Hess et al. published a study that identified 11 different adherence measures calculated using administrative claims data, and initiated the idea of stan- dardizing adherence measures [19]. This study, however, did not empirically validate these measures.

The primary aim of this study was to prospectively validate administrative claims-based adherence measures using hospital- ization as the end point for persons with schizophrenia. Medi- cation adherence was measured in terms of adherence to monotherapy within a class of drugs prescribed for treating schizophrenia. This study sought to identify adherence measures which had the best predictive validity, in an effort to offer prac- titioners and researchers an empirical basis for selecting an adherence measure among eight unique measures based on administrative claims data.

Methods

To assess the predictive validity of each adherence measure, the adherence rates among schizophrenia patients were compared with hospitalization rates. Our primary hypothesis is that an increase in adherence will be associated with lower hospitaliza- tion rates. This was a prospective study where adherence was assessed in year 1 following an index prescription, and hospital- ization rates were determined in the subsequent year.

The study used patient-linked administrative claims data for the Arkansas Medicaid population representing adjudicated paid claims for services rendered from January 1, 2000 through April 30, 2005. A schematic representation of the study periods is shown in Figure 1. Patients with schizophrenia were identified using the ICD-9-CM code of 295.xx (recorded in the medical and inpatient claims file) during the “enrollment period” July 1, 2000 through April 30, 2004. An “index date” defined as the date on which the first prescription for oral antipsychotic medication (conventional: fluphenazine, haloperidol, loxapine, pimozide, perphenazine, trifluoperazine, chlorpromazine, thioridazine,

molindone, thiothixene; atypical: risperidone, quetiapine, olan- zapine, aripiprazole, clozapine, ziprasidone) was filled by the patient in the enrollment period. Adherence rates, defined below, were computed using eight different adherence definitions for the 1-year period starting from the index date.

Subjects

The following inclusion and exclusion criteria were used to select study subjects:

1. Primary diagnosis for schizophrenia (ICD-9-CM = 295.**) [6,20] recorded in the medical and inpatient claims file during the period July 1, 2000 through April 30, 2004 (starting cohort size N = 11,587);

2. Patients with at least one prescription for oral antipsychotic medication in a 1-year period between July 1, 2000 and April 30, 2004 (N = 9025; patients excluded N = 2562);

3. Excluded patients less than 18 years of age at the index date and Qualified Medicare Beneficiaries (N = 7552; patients excluded N = 1473);

4. Continuous eligibility for the 6 months before and 24 months after the index date (N = 6344; patients excluded N = 1208);

5. At least two paid claims for an oral antipsychotic medication in a 1-year period during the period July 1, 2000 through April 30, 2004 (N = 5936; patients excluded N = 408);

6. Excluding patients taking two different oral antipsychotic medications simultaneously (N = 3971; patients excluded N = 1965);

7. Patients were required to have at least one inpatient, out- patient, pharmacy, or nursing home claim during the post- index period. This inclusion criterion was imposed to verify that patients were utilizing Medicaid benefits in the post- index period (N = 3957; patients excluded N = 14);

8. Excluding patients with a nursing home claim during the index period (final cohort size N = 3395; patients excluded = 562).

This study was approved by the Institutional Review Board at the University of Arkansas for Medical Sciences.

Variables

Measures of Medication Adherence Medication adherence was measured in terms of adherence to monotherapy within a class of drugs prescribed for treating

Enrollment Period: July 2000 - April 2004 Index Period* Post Index Period *

2000 2001 2002 2003 2004 365 days 365 days Pre-Index 180 Days

Index Rx

*Study Period Definitions: Pre-Index Period: 180 days before Index date Index Date: first antipsychotic Rx in the enrollment period Index Period: Index Date + 364 days Post Index Period: 365 days after Index Period

Example: Index antipsychotic Rx date: April 30, 2004 Pre-index period: November 1, 2003 - April 29, 2004 Index period: April 30, 2004 - April 29, 2005 and Post index period: April 30, 2005 - April 29, 2006

Figure 1 Description of study periods for a person diagnosed with schizophrenia.

990 Karve et al.

schizophrenia. Patients on two different strengths of the same drug were included in the study. Second, we allowed for non- overlapping switching based on a rationale that the patient should be on some drug that controls schizophrenia. Thus, by excluding patients on multiple antipsychotic drugs, we reduced the complexity of adherence calculations.

The 11 adherence measures that were the initial bases of adherence measures are reported in Table 1 [19]. Three of these measures—MPR, medication refill adherence, and continuous measure of medication acquisition (CMA)—have mathematically equivalent formulas, and only MPR was considered for evalua- tion. Similarly, days-between-fills adherence rate and refill com- pliance rate (RCR) yield the same adherence values, and hence, only RCR was considered for evaluation. Thus, in our study, we compared eight unique measures that have been used to measure adherence using administrative prescription claims data.

The patients were assumed to be 100% adherent to their schizophrenia medications during hospitalization admissions in the 1-year index period. Hence, we subtracted the corresponding number of days a patient was hospitalized during the index period from the denominator for all the eight adherence mea- sures. To check if there was variation in the results, we conducted a sensitivity analysis where we calculated adherence rates without adjusting for the time patient was hospitalized in the 1-year, post-index period.

Dependent Variable Hospitalization in the “postindex period” was defined as any inpatient admission regardless of the primary diagnosis. Mental health-related inpatient admissions were defined as any inpatient admission with a primary diagnosis with the following ICD- 9-CM codes: 295.xx, 296.2x, 296.3x, 296.9x, 300.4x, 309.0x, 311.xx, 300.0x, 300.2x, 300.3x, 306.9x, 308.xx, 309.2x, 309.4x, 309.9x, 297.xx, 298.xx, 299.xx, 300.1x, 302.8x, 307.9x, 290.xx, 291.2x, 310.9x, 331.0. These ICD-9-CM diag- nosis codes were adapted from a previously published study by Weiden et al. [6].

Other Covariates

Demographics. Age (calculated based on the date of birth and the index date), gender, race (whites, blacks, and others) were based on the Medicaid recipient summary file.

Previous hospitalization. A marker variable “previous hospital- ization” was created and defined as any inpatient stay by the patient during the “index period.”

Comorbidity assessment. A measure of comorbidity was based on the Chronic Illness and Disability Payment System developed specifically for Medicaid programs [21]. Inpatient and outpatient claims during the index period were used in calculating comor- bidity scores for the patients.

Prior Cost Prior cost which included inpatient and outpatient nondrug costs was computed by summing the cost for each patient in the 1-year index period.

Admission to rehabilitation center during the index period: covariate indicating patients admitted to rehabilitation center during the index period.

Analysis To determine if the adherence measures are statistically different from each other, Tukey’s and Bonferroni’s multiple comparisons tests were used. To access the association between the adherence measures and hospitalization, we estimated Spearman’s correla- tion coefficients and logistic regression models for each of the eight adherence measures. Reduced models including only the adherence measures and multivariable models including all the covariates were estimated. Second, we inspected odds ratios (ORs) and confidence intervals (CIs) to check if the adherence measure was of the right direction (i.e., as adherence rates increased, hospitalization rates decreased). We used C-statistics to select adherence measure that explains the most variability in hospitalization rates. The C-statistic is defined as the area under receiver operating characteristic curve [22], and the model with the highest C-statistic was considered to be the most predictive of hospitalization. C-statistics for the eight adherence measures were estimated using the univariate and multivariable logistic regression models. Second, we also inspected the ORs to identify those adherence measures most strongly associated with hospi- talization rates.

A sensitivity analysis was conducted excluding patients dually eligible for Medicare benefits. This was done as a check to guard against possible hospital claims that might be missing from the Medicaid files for these dually eligible recipients. All the statisti- cal analyses were conducted using SAS 9.1 (SAS Institute Inc., Cary, NC, USA) hosted on the server or the Windows platform.

Results

Baseline Characteristics The baseline characteristics of the schizophrenia cohort are described in Table 2. The study cohort consisted of 3395 patients

Table 1 Mathematical formulas for the various adherence measures under evaluation

Adherence measure* Formula

Medication possession ratio† Number of days supply in index period/number of days in the study period (365 days) Medication refill adherence† [Number of days supply in index period/number of days in the study period (365)] ¥ 100 Continuous measure of medication acquisition† Number of days supply/total days to next fill or end of observation period (365 days) Proportion of days covered (PDC) [Number of days supply in index period/number of days in the study period (365)] ¥ 100 capped at 1 Refill compliance rate† (Number of days supply/last claim date - index date) ¥ 100 Days-between-fills adherence rate† 1 - [(Last claim date - index date) - total days supply/last claim date - index date] ¥ 100 Compliance ratio Number of days supply in the index period - last days supply/last claim date - index date Medication possession ratio, modified [Number of days supply/(last claim date - index date + last days supply)] ¥ 100 Continuous measure of medication gaps Total days of treatment gaps/total days to next fill or end of observation period (365 days) Continuous multiple interval measure of oversupply Total days of treatment gaps (+) or surplus (-)/total days to next fill or end of observation period

(365 days) Continuous, single interval measure of medication

acquisition Days supply obtained at the beginning of the interval/days in interval

*Hess et al. Ann Pharmacotherapy 2006. †Mathematically similar formulas for calculating adherence.

Validation of Adherence Measures 991

with mean age of 42.8 years of which 52.5% (n = 1782) were female, 52.8% (n = 1793) were white, and 39% (n = 1327) were black. Approximately 47.7% (n = 1619) were also Medicare eli- gible; 23.2% (n = 788) of the patients had an inpatient admission during the index period. During the post-index period, 28.8% (n = 979) had an inpatient admission, and 18.1% (n = 616) had inpatient admission related to a mental health condition.

Adherence Rates The adherence rates for the schizophrenia cohort are described in Table 3. Adherence rates were observed to be slightly higher after excluding the number of days hospitalized from the denominator in the adherence calculation. The adherence rates varied between a low of 0.724 for proportion of days covered (PDC) and a high of 0.909 for RCR for the adherence measures where higher values correspond to better adherence. The adherence values were significantly different (P < 0.05) from each other for all the nongap-based possible pair-wise comparisons except RCR– continuous, single interval measure of medication acquisition (CSA), compliance ratio (CR)–medication possession ratio, modified (MPRm), and PDC–MPR (data not shown). The values for the gap measures, continuous measure of medication gaps (CMG) and continuous multiple interval measure of oversupply (CMOS), were 0.276 and 0.262, respectively, and values closer to zero correspond to better adherence. MPR, PDC, CMG, and CMOS, which use the entire index period in the denominator, tended to have lower adherence values than RCR, CR, and

MPRm which only consider the period between the first and last prescription in the index period.

All eight adherence measures were found to be significantly correlated with hospitalization rates. The six adherence measures MPR (-0.107, P � 0.001), PDC (-0.113, P � 0.001), RCR (-0.048, P � 0.006), CR (-0.07, P � 0.001), MPRm (-0.066, P � 0.001), and CSA (-0.040, P � 0.021) were negatively cor- related with any-cause hospitalization. Whereas, the two gap- based adherence measures CMG (0.113, P � 0.001) and CMOS (0.107, P � 0.001), are positively correlated with any-cause hospitalization.

The OR estimates and C-statistics for the full and reduced models for each of the eight adherence measures considering any-cause inpatient admission and mental health-related in- patient admission are presented in Tables 4 and 5.

All adherence measures other than CSA (OR = 0.941 95% CI: 0.797–1.113) and RCR (OR = 0.833, 95% CI: 0.678–1.024) were statistically significant predictors of any-cause hospitaliza- tion in the reduced models. The reduced model OR estimates varied between 0.417 (PDC) and 0.553 (CR) for the nongap measures indicating that as the adherence rates increase, the chances of hospitalization decrease. The two gap-based mea- sures, CMG and CMOS, had ORs of 2.398 and 2.253, respec- tively, which also demonstrate that improved adherence decreases the probability of hospitalization. The reduced models with PDC and CMG as adherence measures had the highest univariate C-statistics of 0.571. Mathematically, (PDC and CMG) and (MPR and CMOS) are the inverse of each other; hence, numerically similar C-statistics should be observed for these two pairs of measures. Similarly, CMG and PDC had the highest C-statistics of 0.687 in the full model; however, MPR and CMOS had slightly lower C-statistics of 0.686. Adherence measures PDC (OR = 0.498), MPR (OR = 0.519), CMG (OR = 2.007), and CMOS (OR = 1.927) had ORs farthest from 1. Similar relationships were observed for the models on mental health hospitalization where PDC and CMG had the highest C-statistics in both the reduced and full models.

Sensitivity Analysis

In predicting any-cause and mental health-related hospitalization, PDC and CMG were found to have the highest C-statistics of 0.584 and 0.579, respectively. Even after adjusting for other cova- riates, models containing these two measures (PDC and CMG) had the highest C-statistics of 0.686 and 0.671 in predicting any-cause and mental health-related hospitalization. Models with MPR and CMOS remained the second best predictors of any- cause and mental health-related hospitalization after adjusting for covariates. The same four measures had ORs farthest from 1.0.

Table 2 Baseline characteristics of schizophrenia cohort (N = 3395)

Mean (N) SD (%)

Age (years) 42.9 13.2 18–30 643 18.9 31–40 977 28.8 41–50 928 27.3 51–64 637 18.8 65 and above 210 6.2

Sex Female 1782 52.5 Male 1613 47.5

Race White 1793 52.8 Black 1327 39.1 Other/unknown 275 8.1

Medicare eligible 1619 47.7 Comorbidity score 2.1 1.2 Index period hospitalization 788 23.2 Post-index period hospitalization Any-cause hospitalization 979 28.8 Mental health–related hospitalization 616 18.1

Table 3 Adherence rates for the schizophrenia cohort (N = 3395)

Adherence measure

Adherence rates without accounting for hospitalization days

Adherence rates after excluding days hospitalized

Mean SD Mean SD

Medication possession ratio 0.733 0.309 0.738 0.310 Proportion of days covered 0.720 0.295 0.724 0.295 Refill compliance rate 0.902 0.384 0.909 0.386 Compliance ratio 0.766 0.287 0.772 0.288 Medication possession ratio, modified 0.789 0.246 0.794 0.247 Continuous measure of medication gaps (CMG)* 0.280 0.295 0.276 0.295 Continuous multiple interval measure of oversupply (CMOS)* 0.267 0.309 0.262 0.310 Continuous, single interval measure of medication acquisition 0.880 0.444 0.893 0.464

*CMG and CMOS are gap measure; lower values represent better compliance.

992 Karve et al.

Discussion

With the availability of different adherence measures that can be computed using claims data, researchers often face a dilemma in selecting an adherence measure. Our study compared the predic- tive validity of eight different adherence measures computed using administrative claims data for any-cause and disease- related inpatient episodes. To the knowledge of the researchers, the present study was the first of its kind to prospectively validate adherence calculated using pharmacy claims among schizophre- nia patients, and provides an empirical basis for selecting adher- ence measures based on administrative claims.

All but three of all the possible adherence pair-wise compari- sons yielded statistically different adherence values than each other, and there is a 25% difference between the most conserva- tive measure, PDC = 0.720, and the most optimistic, RCR = 0.902, so the selection of an adherence metric can yield fairly meaningful differences. All the adherence measures (except RCR and CSA) were found to be significant predictors of inpatient admission even after adjusting for covariates such as demo- graphic factors, prior hospitalization, comorbidity, and prior cost. Among the eight adherence measures, PDC and CMG had the highest C-statistics and OR farthest from 1, and were the best predictors of future hospitalization (any cause and mental health related). MPR and CMOS had very similar C-statistics and were the two second best predictors of any-cause and mental health- related hospitalization. Because the formula employed in calcu- lating PDC is similar to that of MPR, the only difference being that for PDC the adherence is capped at 1; these findings are expected. Our findings suggest that oversupply or drug stockpil- ing has a minor impact on hospitalization. These findings were

robust as PDC remained the best predictor after excluding patients eligible for Medicare benefits. Researchers can select between PDC, MPR, CMG, and CMOS in estimating adherence as the differences in discriminatory power among these four adherence measures remain relatively small. Nevertheless, we recommend using MPR or PDC for a couple of reasons. Both MPR and PDC have formulas in which better compliance corre- sponds to higher values, thus these two may be preferred because they provide more intuitive values. Moreover, in our previous research, we found MPR and PDC to be most predictive of hospitalizations among diabetes patients [23]. Researchers should, however, consider MPR as their primary choice in cal- culating adherence as it has been widely used in many pharmacy claims-based studies.

In our study, we found that adherence measures using the entire study period (365 days) as the denominators (MPR, PDC, CMG, and CMOS) were better predictors of hospitalization (univariate C-statistic range: 0.567–0.571) as compared to the adherence measures that only considered the period between first and last refill (RCR, CR, MPRm: univariate C-statistic range: 0.525–0.549). The primary reason for this might be because of the fact that if we only consider the period between first and last refills, then it does not account for early discontinuation and would overestimate adherence for persons that completely stop taking their medications. Consider a hypothetical scenario wherein a person fills three consecutive prescriptions of each 30-day supply and has a 90-day period between the first and last refills, then the RCR value will be 1.0 (RCR = 90/90), indicating perfect adherence. The MPR or PDC value for the same person would be 0.25 (MPR = 90/365). For patients with chronic con- ditions such as schizophrenia, it is important for these adherence

Table 4 Comparison of odds ratios (ORs), full model, and reduced model* (any-cause hospitalization)

Adherence measure

Full multivariate model Reduced univariate model

OR 95% Wald CI Pr >

chi-square C-

statistics OR 95% Wald CI Pr >

chi-square C-

statistics

Medication possession ratio 0.519 0.398 0.677 <0.001 0.686 0.444 0.350 0.562 <0.001 0.568 Proportion of days covered 0.498 0.377 0.659 <0.001 0.687 0.417 0.326 0.534 <0.001 0.571 Refill compliance rate 0.911 0.738 1.124 0.383 0.677 0.833 0.678 1.024 0.083 0.530 Compliance ratio 0.653 0.492 0.868 0.003 0.681 0.553 0.427 0.717 <0.001 0.544 Medication possession ratio, modified 0.617 0.445 0.856 0.004 0.681 0.508 0.378 0.683 <0.001 0.542 Continuous measure of medication gaps (CMG)† 2.007 1.518 2.653 <0.001 0.687 2.398 1.873 3.071 <0.001 0.571 Continuous multiple interval measure of oversupply (CMOS)† 1.927 1.478 2.513 <0.001 0.686 2.253 1.778 2.854 <0.001 0.568 Continuous, single interval measure of medication acquisition 0.897 0.756 1.064 0.212 0.677 0.941 0.797 1.113 0.479 0.525

*Full multivariate model includes covariates—adherence measure, comorbidity score, age, gender, race, prior hospitalization, and prior cost. Reduced univariate model includes only the adherence measure. †CMG and CMOS are gap measure; lower values represent better compliance.

Table 5 Comparison of odds ratios (ORs), and full multivariate and reduced univariate model* (mental health hospitalization)

Adherence measure

Full multivariate model* Reduced univariate model

OR 95% Wald CI Pr >

chi-square C-

statistics OR 95% Wald CI Pr >

chi-square C-

statistics

Medication possession ratio 0.544 0.402 0.736 <0.001 0.665 0.456 0.347 0.600 <0.001 0.567 Proportion of days covered 0.522 0.380 0.717 <0.001 0.666 0.430 0.323 0.572 <0.001 0.571 Refill compliance rate 0.789 0.611 1.017 0.068 0.657 0.711 0.548 0.923 0.011 0.536 Compliance ratio 0.622 0.451 0.858 0.004 0.660 0.522 0.387 0.706 <0.001 0.549 Medication possession ratio, modified 0.594 0.411 0.859 0.006 0.660 0.484 0.344 0.681 <0.001 0.547 Continuous measure of medication gaps (CMG)† 1.916 1.396 2.632 <0.001 0.666 2.328 1.749 3.099 <0.001 0.571 Continuous multiple interval measure of oversupply (CMOS)† 1.838 1.359 2.486 <0.001 0.665 2.191 1.665 2.883 <0.001 0.567 Continuous, single interval measure of medication acquisition 0.919 0.758 1.115 0.394 0.655 0.957 0.787 1.165 0.663 0.528

Validation of Adherence Measures 993

measures to account for early discontinuations, and our data support this fact. This study also affirms the importance of medication adherence in preventing future hospital utilization [3,5,8,24]. It should also be noted that MPR and PDC were not statistically different from each other and yielded similar results which reflect the same manner in which these are calculated for all values less than 1.0.

Certain limitations of this study need to be acknowledged. We calculated medication adherence using claims data, and thus the actual consumption of the medication was not estimated although medication possession is imperative for its consump- tion. We measured medication adherence in terms of adherence to monotherapy within a class of drugs prescribed for treating schizophrenia, but we allowed for patients taking different strengths of the same drug simultaneously and for switches between antipsychotics. Focusing on persons taking one antipsy- chotic at a time limits the generalizability, and the findings may not be applicable to patients on more than one concomitant therapy, which may be presumed to have greater disease severity and/or lack perceived efficacy. Unfortunately, the adherence mea- sures we investigated do not have standardized methods for incorporating multiple concurrent medication use in the calcula- tions, and we believe that this is an important next step in advancing adherence measures to handle these complex situa- tions. Depending on the scope of studies, some analyses consider medication switches as new therapy starts. Our analysis adopted the perspective of measuring adherence to antipsychotics in general, but not to any one particular antipsychotic and ignored medication switches within the antipsychotic class. This perspec- tive will yield different adherence values than those that focus on adherence to a particular antipsychotic, although the relationship between the adherence values and subsequent hospitalizations may not be greatly different. Additionally, our study design allowed for patients taking different strengths of the same drug simultaneously, thus, adherence rates may be overestimated in such patients. Several other factors, such as severity of the illness, side effects, occupational status, education level, patient support system, and side effects may affect medication adherence as well as hospitalization, were not available in the data set and were not included in the multivariable models, and if included, these addi- tional variables may alter the adjusted relationships we report. Lastly, the C-statistics reported in the univariate models were low; none of the adherence measures alone had a C-statistic >0.60, which indicates that adherence by itself, although statis- tically significant, is a relatively modest factor influencing subse- quent hospitalization. The full models which included prior hospitalization, comorbidity burden, and patient demographics in addition to the adherence measures had higher C-statistics greater than 0.60, but less than 0.70 which indicate that subse- quent hospitalization can only be modestly predicted given the confines of an administrative data source.

In our study, we initially considered two outcome measures, hospitalization and total nonpharmacy costs for discriminating between adherence measures based on past empirical evidence which has consistently demonstrated a relationship between adherence and these two outcome measures. After our initial analysis, we found a positive correlation between the adherence measures and nonpharmacy costs, indicating that an increase in adherence was associated with increased costs which is in con- trast to several studies that have found a negative correlation indicating that lower adherence was associated with increased health-care expenditures and utilization [3–6,8,25]. The positive correlation observed in our data may be because of the fact that approximately 83% of our study sample was in a rehabilitation center during the index or the postindex period. These rehabili-

tation centers are expensive and monitor adherence which may drive up adherence resulting in the positive association between nonpharmacy costs and adherence we observed. Given the likely endogeneity between cost and adherence in these rehabilitation centers and the fact that a very high percentage of our sample accessed these centers, we chose to not use nonpharmacy cost to discriminate between these different adherence measures. Lastly, there are other possible dependent measures such as medication augmentation or switching that may also have the potential to discriminate between adherence measures, and we encourage future research in this area.

In our literature review, we came across nonconsistent nomenclature used for adherence measures where the same term might have two different mathematical expressions or definitions [18]. For example, adherence measures such as continuous mul- tiple refill interval measure of medication availability (CMA), medication-total (MED_TOT), are used synonymously with MPR [18,19,26,27]. Conversely, we also found that different terms have the same mathematical expression or definition [18,28–33]. Because the objective of our study was to validate the measures, we adopted the nomenclature and formulas used by Hess et al. in their study [19]. Secondly, our adherence defi- nitions were based on the most recent published literature [19] at the time this study was undertaken, and we acknowledge that newer adherence definitions may have evolved. One of the adher- ence measures, PDC, has been defined differently than defined in this report, and the alternative definition for PDC should be considered [34]. The PDC definition in our paper may alterna- tively be described as a “capped” MPR or MPR modified measure. Because our study provides the precise formulas for the adherence measures, we suggest that future researchers and prac- titioners consider using these definitions, except for PDC, and names when describing adherence measures in the future.

Conclusion

Significant associations were observed between six different mea- sures of medication adherence any-cause and disease-specific hospitalization rates among patients with schizophrenia. Among schizophrenia patients not prescribed polypharmacy, we found that the PDC, MPR, CMOS, and CMG were the best predictors of any-cause and mental health-related hospitalization as com- pared to other adherence measures. Overall adherence measures considering an index period of 365 days as the denominator were better predictors of subsequent hospitalization as compared to adherence measures that only considered the period between the first and last refills. The MPR and PDC provide similar intuitively appealing adherence values, where better adherence corresponds to higher values, and had the highest predictive validity for subsequent hospitalizations. Ideally, persons should explore a range of possible metrics to measure adherence, but in instances when only a single adherence calculation is sought, persons should first consider these measures when calculating adherence from administrative claims data for persons with schizophrenia. Our recommendations are based on the ability of the adherence measures to predict one outcome—hospitalization. Additional studies are warranted to measure the predictive ability of adher- ence measures on other potential outcomes.

We would like to thank Gary Moore and Shiming Dong for their invalu- able assistance in extracting the raw claims files and creating research ready data sets, and providing technical assistance when needed. This article was improved with the helpful comments of three anonymous reviewers.

994 Karve et al.

Source of financial support: This study was completed without any direct financial support.

Supporting Information for this article can be found at: http:// www.ispor.org/publications/value/ViHsupplementary.asp

References

1 Cramer JA, Roy A, Burrell A, et al. Medication compliance and persistence: terminology and definitions. Value Health 2008;11: 44–7.

2 Dunbar-Jacob J, Mortimer-Stephens MK. Treatment adherence in chronic disease. J Clin Epidemiol 2001;54(Suppl. 1):S57–60.

3 Svarstad BL, Shireman TI, Sweeney JK. Using drug claims data to assess the relationship of medication adherence with hospitaliza- tion and costs. Psychiatr Serv 2001;52:805–11.

4 Thieda P, Beard S, Richter A, Kane J. An economic review of compliance with medication therapy in the treatment of schizo- phrenia. Psychiatr Serv 2003;54:508–16.

5 Valenstein M, Copeland LA, Blow FC, et al. Pharmacy data iden- tify poorly adherent patients with schizophrenia at increased risk for admission. Med Care 2002;40:630–9.

6 Weiden PJ, Kozma C, Grogg A, Locklear J. Partial compliance and risk of rehospitalization among California Medicaid patients with schizophrenia. Psychiatr Serv 2004;55:886–91.

7 Osterberg L, Blaschke T. Adherence to medication. N Engl J Med 2005;353:487–97.

8 Knapp M, King D, Pugner K, Lapuerta P. Non-adherence to antipsychotic medication regimens: associations with resource use and costs. Br J Psychiatry 2004;184:509–16.

9 Kalsekar I, Iyer S, Mody R, et al. Utilization and costs for com- pliant patients initiating therapy with pioglitazone or rosiglita- zone versus insulin in a Medicaid fee-for-service population. J Manag Care Pharm 2006;12:121–9.

10 Powers CA, Meyer CM, Roebuck MC, Vaziri B. Predictive mod- eling of total healthcare costs using pharmacy claims data: a comparison of alternative econometric cost modeling techniques. Med Care 2005;43:1065–72.

11 Sokol MC, McGuigan KA, Verbrugge RR, Epstein RS. Impact of medication adherence on hospitalization risk and healthcare cost. Med Care 2005;43:521–30.

12 Choo PW, Rand CS, Inui TS, et al. Validation of patient reports, automated pharmacy records, and pill counts with electronic monitoring of adherence to antihypertensive therapy. Med Care 1999;37:846–57.

13 Cook CL, Wade WE, Martin BC, Perri M. Concordance among three self-reported measures of medication adherence and phar- macy refill records. J Am Pharm Assoc (Wash) 2005;45:151– 9.

14 Garber MC, Nau DP, Erickson SR, et al. The concordance of self-report with other measures of medication adherence: a summary of the literature. Med Care 2004;42:649–52.

15 Grymonpre R, Cheang M, Fraser M, et al. Validity of a prescrip- tion claims database to estimate medication adherence in older persons. Med Care 2006;44:471–7.

16 Kwon A, Bungay KM, Pei Y, et al. Antidepressant use: concor- dance between self-report and claims records. Med Care 2003; 41:368–74.

17 McKenzie DA, Semradek J, McFarland BH, et al. The validity of Medicaid pharmacy claims for estimating drug use among elderly

nursing home residents: the Oregon experience. J Clin Epidemiol 2000;53:1248–57.

18 Andrade SE, Kahler KH, Frech F, Chan KA. Methods for evalu- ation of medication adherence and persistence using automated databases. Pharmacoepidemiol Drug Saf 2006;15:565–74.

19 Hess LM, Raebel MA, Conner DA, Malone DC. Measurement of adherence in pharmacy administrative databases: a proposal for standard definitions and preferred measures. Ann Pharmacother 2006;40:1280–8.

20 Gianfrancesco F, Rajagopalan K, Wang RH. Hospitalization risks in the treatment of schizophrenia: comparison of antipsychotic medications. J Clin Psychopharmacol 2006;26:401–4.

21 Kronick R, Gilmer T, Dreyfus T, Lee L. Improving health-based payment for Medicaid beneficiaries: CDPS. Health Care Financ Rev 2000;21:29–64.

22 Ash A, Shwartz M. R2: a useful measure of model performance when predicting a dichotomous outcome. Stat Med 1999;18:375– 84.

23 Karve SJ, Cleves M, Helm M, et al. An empirical basis for stan- dardizing adherence measures derived from administrative claims data among diabetic patients. Med Care 2008;46:1125–33.

24 Gilmer TP, Dolder CR, Lacro JP, et al. Adherence to treatment with antipsychotic medication and health care costs among Med- icaid beneficiaries with schizophrenia. Am J Psychiatry 2004;16: 692–9.

25 Eaddy M, Grogg A, Locklear J. Assessment of compliance with antipsychotic treatment and resource utilization in a Medicaid population. Clin Ther 2005;27:263–72.

26 Steiner JF, Prochazka AV. The assessment of refill compliance using pharmacy records: methods, validity, and applications. J Clin Epidemiol 1997;50:105–16.

27 Cantrell CR, Eaddy MT, Shah MB, et al. Methods for evaluating patient adherence to antidepressant therapy: a real-world com- parison of adherence and economic outcomes. Med Care 2006; 44:300–3.

28 Darkow T, Henk HJ, Thomas SK, et al. Treatment interruptions and non-adherence with imatinib and associated healthcare costs: a retrospective analysis among managed care patients with chronic myelogenous leukaemia. Pharmacoeconomics 2007;25: 481–96.

29 Kalsekar ID, Madhavan SS, Amonkar MM, et al. Depression in patients with type 2 diabetes: impact on adherence to oral hypoglycemic agents. Ann Pharmacother 2006;40:605–11.

30 Lau DT, Nau DP. Oral antihyperglycemic medication nonadher- ence and subsequent hospitalization among individuals with type 2 diabetes. Diabetes Care 2004;27:2149–53.

31 Recker RR, Gallagher R, MacCosbe PE. Effect of dosing fre- quency on bisphosphonate medication adherence in a large lon- gitudinal cohort of women. Mayo Clin Proc 2005;80:856–61.

32 Stern L, Berman J, Lumry W, et al. Medication compliance and disease exacerbation in patients with asthma: a retrospective study of managed care data. Ann Allergy Asthma Immunol 2006;97:402–8.

33 Stroupe KT, Teal EY, Tu W, et al. Association of refill adherence and health care use among adults with hypertension in an urban health care system. Pharmacotherapy 2006;26:779–89.

34 Martin BC, Wiley-Exley EK, Richards S. Contrasting measures of adherence in the presence of simple drug use, medication switch- ing, and therapeutic duplication. Ann Pharmacother 2009;43:36– 44.

Validation of Adherence Measures 995

http://www.ispor.org/publications/value/ViHsupplementary.asp