Abstract

Objective

To examine the relationship of three alternative measures of adherence with seven negative outcomes associated with epilepsy for development of a quality measure in epilepsy.

Design

Retrospective cohort analysis.

Setting

PharMetrics national claims database.

Participants

Patients in the PharMetrics database for the years 2004–08 taking antiepileptic drugs.

Intervention

None.

Main Outcome Measures

For each definition of adherence, the odds ratios (ORs) comparing non-adherent with adherent groups were assessed for consistency and direction for the number of hospital admissions, emergency room (ER) visits, head injuries including traumatic brain injuries, falls, motor vehicle accidents (MVAs), fractures and a ‘seizure’ outcome defined as hospital admissions or ER visits with a primary diagnosis of epilepsy or convulsions.

Results

The inclusion criteria were met by 31 635 individuals. In the multivariate analysis, the adherent group had lower odds of hospital admissions with ORs for the eight specifications ranging from 0.729 to 0.872 and ER visits where ORs for the eight specifications ranged from 0.750 to 0.893. The eight ORs for head injuries ranged from 0.647 to 0.888. For fractures, the ORs ranged from 0.407 to 0.841. Our proxy for seizure was inconsistently associated with adherence status.

Conclusions

All the adherence measures defined non-adherent groups that were associated with negative outcomes in epilepsy.

Introduction

The Institute of Medicine has defined quality as ‘the degree to which health care services increase the likelihood of the desired health outcomes’ [1]. Health-care quality measures, consistent with the Institute of Medicine definition, can be defined to improve quality or performance. The purpose of this article is to define measures that could be used to improve the quality of care by measuring prescription drug utilization and possible undesirable outcomes in epilepsy.

Epilepsy is one of the most common neurologic disorders, affecting an estimated 2.1–2.7 million persons in the USA [2, 3], and costs approximately $15.5 billion in medical costs and lost or reduced earnings and productivity each year [3]. The majority of people with epilepsy are able to manage their condition by using one or more pharmacological therapies [4]. Previous studies have reported that ∼30–60% of patients with epilepsy are non-adherent to their prescribed antiepileptic drug (AED) therapy [5, 6]. In a patient survey, more than 70% of respondents reported AED dose omission [7]. In addition, some results have suggested that poor adherence is associated with higher risk of seizures [8], increases in medical resource utilization and costs [5] and severe clinical consequences, including increased risk of death [9].

There are no widely adopted health-care quality measures for the management of a population with epilepsy. Organizations that endorse health-care quality measures may rely on researchers to generate and publish evidence on the performance of candidate metrics as quality measures. In a population with epilepsy, the performance of an adherence-based quality measure could be assessed as its ability to define a non-adherent cohort with a greater likelihood of negative outcomes than the complementary adherent cohort. While adherence is only a proxy for actual patient outcomes, quality improvement efforts in epilepsy could be assessed by their effect in reducing the non-adherent population of patients with epilepsy.

A number of studies have documented an association between adherence and negative outcomes in epilepsy. Manjunath et al. [8] demonstrated that the hazard ratio for subsequent seizure with adherence measured using the medication possession ratio (MPR) was 0.83 (Manjunath's estimated hazard ratio has been inverted, since results are being reported relative to non-adherence throughout this article). A small study of 54 patients showed an association between seizure control and adherence measured on the Morisky scale [6]. In addition, a postal survey noted an 11% (P = 0.10) higher risk of seizure after a missed dose [7]. Davis et al. [10] demonstrated that non-adherence measured by the MPR was associated with approximately an 11% increased likelihood in hospitalizations, a 48% increased likelihood of admission to the emergency room (ER) and a 44% increased annual likelihood of injuries in motor vehicle accidents (MVAs). In an analysis of Medicaid data from three states, Faught et al. [9] demonstrated that poor adherence measured by the proportion of days with possession of at least one AED, typically referred to as proportion of days covered (PDC), resulted in a 50% increase in ER visits, an 86% increase in hospitalizations, a 108% increase in MVAs and a 21% increase in fractures.

Measurement of adherence to medication therapy is not standardized. In a 2006 review, Hess et al. [11] identified 11 different measures of medication adherence. The measures of adherence used in four separate articles specific to AED therapy are summarized in Table 1.

Table 1

Adherence measures in published studies in epilepsy

Author Numerator Denominator 
Davis et al. (2008) [10Sum of days supply Days between the first AED prescription and expiration of the days' supply of the last observed AED refill 
Faught et al. (2008) [9Days in quarter with at least one AED in possession Days in quarter 
Ettinger et al. (2009) [13Sum of days supply Days between the first AED prescription and expiration of the days' supply of the last observed AED refill 
Manjunath et al. (2009) [8Sum of days supply in cumulative period assessed every 30 days Days in cumulative period assessed every 30 days 
Author Numerator Denominator 
Davis et al. (2008) [10Sum of days supply Days between the first AED prescription and expiration of the days' supply of the last observed AED refill 
Faught et al. (2008) [9Days in quarter with at least one AED in possession Days in quarter 
Ettinger et al. (2009) [13Sum of days supply Days between the first AED prescription and expiration of the days' supply of the last observed AED refill 
Manjunath et al. (2009) [8Sum of days supply in cumulative period assessed every 30 days Days in cumulative period assessed every 30 days 

AED, antiepileptic drug.

The purpose of this article is to determine how different definitions of adherence, evaluated at multiple thresholds, perform as quality measures in epilepsy. Specifically, the association of these measures with a range of negative outcomes will be described.

Methods

Data source

The data set for this retrospective analysis was the PharMetrics Patient-Centric Database (IMS Health, Norwalk, CT, USA) of claims from more than 90 different managed care organizations representing more than 55 million lives. The study data set encompassed claims with dates of service from 1 September 2003 through 30 September 2008.

Patient identification

Study patients were identified during the identification period, 1 March 2004 to 30 September 2007. The index date was defined as the date of the first AED pharmacy claim during the identification period with enrollment 6 months before and 12 months after the index date. The AED drugs were identified using the generic name field in the Pharmetrics data base. The definitions of study time periods are shown in Fig. 1. Patients were required to be at least 18 years old of age on the index date (no upper age limit) with at least one medical claim with any diagnosis of epilepsy (ICD-9 codes 345.xx) (in any of the four diagnosis fields) in the 6-month period prior to or on the index date. Patients had to have at least two AED pharmacy claims on separate dates and at least 60 days supply for products identified by the generic names for AED drugs (Table 2) during the study time frame. Only patients who were continuously enrolled for at least 6 months prior to the index date and continuously eligible for 12 months after the index date (follow-up period) were included in the study (Fig. 2).

Table 2

Patient characteristics (n = 31 635)

Characteristic n Mean (SD) or % 
Demographic characteristics 
 Age (overall mean age as continuous variable) 31 635 47.44 (16.37) 
  18–<40 10 037 31.7 
  40–<65 17 323 54.8 
  ≥65 years old  4275 13.5 
 Gender 
  Male 13 884 43.9 
  Female 17 750 56.1 
  Unknown 
 Health plan type 
  HMO 9533 30.1 
  PPO 13 938 44.1 
  POS 4259 13.5 
  Missing/unknown 3905 12.4 
 AED type (if double agents for one patient, the first one is counted) 
  Carbamazepine 5457 17.3 
  Ethosuximide 75 0.2 
  Felbamate 59 0.2 
  Gabapentin 1763 5.6 
  Oxcarbazepine 1968 6.2 
  Lamotrigine 3397 10.7 
  Phenobarbital 1302 4.1 
  Phenytoin 9074 28.7 
  Primidone 523 1.7 
  Levetiracetam 3956 12.5 
  Tiagabine 114 0.4 
  Topiramate 2954 9.3 
  Pregabalin   
  Zonisamide 849 2.7 
  Valproic acid 144 0.5 
Patients with any inpatient admissions (%) 4870 15.4 
Patients with any ER visit (%) 7535 23.8 
Patients with TBI 209 0.7 
Patients with injury due to fall 32 0.1 
Patients with MVA 12 0.0 
Patients with fractures 246 0.8 
Patients with seizures 932 2.9 
Characteristic n Mean (SD) or % 
Demographic characteristics 
 Age (overall mean age as continuous variable) 31 635 47.44 (16.37) 
  18–<40 10 037 31.7 
  40–<65 17 323 54.8 
  ≥65 years old  4275 13.5 
 Gender 
  Male 13 884 43.9 
  Female 17 750 56.1 
  Unknown 
 Health plan type 
  HMO 9533 30.1 
  PPO 13 938 44.1 
  POS 4259 13.5 
  Missing/unknown 3905 12.4 
 AED type (if double agents for one patient, the first one is counted) 
  Carbamazepine 5457 17.3 
  Ethosuximide 75 0.2 
  Felbamate 59 0.2 
  Gabapentin 1763 5.6 
  Oxcarbazepine 1968 6.2 
  Lamotrigine 3397 10.7 
  Phenobarbital 1302 4.1 
  Phenytoin 9074 28.7 
  Primidone 523 1.7 
  Levetiracetam 3956 12.5 
  Tiagabine 114 0.4 
  Topiramate 2954 9.3 
  Pregabalin   
  Zonisamide 849 2.7 
  Valproic acid 144 0.5 
Patients with any inpatient admissions (%) 4870 15.4 
Patients with any ER visit (%) 7535 23.8 
Patients with TBI 209 0.7 
Patients with injury due to fall 32 0.1 
Patients with MVA 12 0.0 
Patients with fractures 246 0.8 
Patients with seizures 932 2.9 
Figure 1

Timeline of data collection.

Figure 1

Timeline of data collection.

Figure 2

Patient selection.

Figure 2

Patient selection.

Adherence measures

After reviewing the existing literature for epilepsy and adherence, three types of adherence measures were selected: PDC, MPR and gaps in therapy.

Proportion of days covered

The PDC was defined as the number of unique days with AED therapy in the 365 days after the index date divided by 365. Each day on which the patient had possession of an AED was counted as 1 day of possession, even for patients with multiple AED therapies. 

formula

Medication possession ratio

The MPR has widely divergent meanings in the literature. We defined the MPR as the sum of the days supplied from all AED prescriptions within 365 days after the index date divided by 365. When the days supplied overlapped the 365th day, the days supplied were only counted until the 365th day. When patients were on more than one product in the studied category, the MPR could have exceeded 1.0 and was truncated at 100% [12]. 

formula

Gap in therapy

Gap in therapy was defined as a period without drug supply, measured as the days between a prescription fill date and the number of days supplied and the date of the next prescription fill. For example, if an individual received a 30-day supply on 15 March, but no further drug supplies until 1 May, the gap in therapy would be 17 days.

These three types of adherence measures can be used to assign adherence status to a given patient, given an adherence threshold. In this study, PDC and MPR were examined at thresholds of 70, 80 and 90%, while the gaps in therapy measure were examined for threshold gaps of 15 and 30 days. The associated population level measure for each of these definitions was the percentage of patients with adherence rates above the threshold percentage (MPR and PDC) or without gaps in therapy at or exceeding the gap threshold.

Outcome measures

In this article, the following seven negative outcomes associated with epilepsy were studied: hospital admissions; ER visits; head injuries, including traumatic brain injuries (TBIs); fractures, falls, MVAs and a proxy measure for seizure. Medical claims were analyzed for the presence of these outcomes. Hospital and ER visit outcomes were assessed independent of diagnosis [i.e. they were not limited to claims with a diagnosis of epilepsy (ICD-9 codes 345.xx) or seizure (ICD-9 codes 780.3x)]. All claims were searched for the following codes. Head injuries (including TBI) were defined as ICD-9 codes 800–804, 850–854, 873 and 959.01 [9]. MVAs were defined as ICD-9 code E810.0–E825 and fractures were defined as ICD-9 codes 800–829 [9]. Falls were defined as ICD-9 codes E880.0–E886.9 [10]. A proxy for seizure was defined as a hospital or ER claim with a primary diagnosis of epilepsy or seizure, as the specific seizure event is not uniquely associated with a diagnosis code or medical claim.

Statistical analysis

A difference of proportions test was used to compare the likelihood of each outcome based on adherence. This difference in proportions test was repeated for all eight definitions of adherence (MPR 70%, MPR 80%, MPR 90%, PDC 70%, PDC 80%, PDC 90%, gaps in therapy of 15 days and gaps in therapy of 30 days). Univariate odds ratios (ORs) and the 95% confidence intervals for each comparison of adherent to non-adherent patients were calculated.

To further examine how performance of the adherence measures was influenced by covariates, logistic regression models for four bivariate outcomes (hospital admissions, ER visits, proxy for seizure and accidents/injuries) were constructed, controlling for potentially influential covariates, such as age, gender, Deyo-adapted Charlson comorbidity index, comorbidities, including neurotic disorder, brain tumor, depression and dementia, type of insurance, defined as either commercial, Medicare, Medicaid or self-insured. Because the samples for head injuries, fractures, falls and MVAs were too small to construct separate models, they were collapsed into an outcome defined as accident/injury. ORs were once again generated for the eight definitions of adherence under consideration. No adjustments were made for multiple comparisons. We did not correct for multiple comparisons which, in this case, would simply expand the size of the confidence intervals of each point estimate.

Results

The application of study criteria to create the final study population is depicted in Fig. 2. After applying the criteria for eligibility, diagnosis and AED use, 31 635 patients were included in the study population.

The mean age of the population was 47.4 years, and 31.7% of the population was aged 18–39 years. Females comprised 56% of the population. The most commonly used AED was phenytoin, with 34.4% of the patients filling at least one prescription for that product, followed by carbamazepine (22.9%), and levetiracetam (20.5%). On average, patients filled prescriptions for 1.68 different AED products, with 63.7% exposed to only one product during the study period. The average number of prescriptions filled per patient in the 12-month follow-up period was 7.53. More than one-third (37.1%) of the population had a mental health diagnosis. Common mental health comorbidities included neurotic disorders (8.7%), depressive disorder (7.5%) and affective psychoses (7.1%). More than 17% of the population had scores >3 (range, 0–18) on the Deyo-adapted Charlson comorbidity index.

The mean adherence in terms of MPR was 84.3% across the study population. Using PDC, the mean adherence was 78.9%. Approximately half of the population (49.8%) had at least 1 gap of 15 days or more without AED therapy. Nearly, one-third (30.8%) had at least 1 gap of 30 days or more without AED therapy.

The univariate ORs for the seven outcome measures are shown in Table 3. ORs of <1.0 are consistent with the hypothesis that adherence is associated with a reduced likelihood of negative outcomes. The probability of admissions and ER visits was lower for all eight of the adherence measure definitions. The odds of head injuries and fractures were significantly lower for adherent patients defined by the two gaps in therapy measures and by all three PDC measures. For falls, MVAs and the proxy for seizure, no consistent pattern across the measures was observed.

Table 3

Unvariate ORs of the relationship between medication adherence and epilepsy-related outcomes (n = 31 635)

Odds ratio (95% CI) 15-day gap 30-day gap MPR 70% MPR 80% MPR 90% PDC 70% PDC 80% PDC 90% 
Admission 0.707 (0.665, 0.752) 0.746 (0.700, 0.795) 0.498 (0.466, 0.532) 0.747 (0.700, 0.797) 0.801 (0.753, 0.853) 0.650 (0.609, 0.694) 0.641 (0.603, 0.682) 0.648 (0.609, 0.690) 
ER visits 0.683 (0.648, 0.719) 0.700 (0.662, 0.739) 0.465 (0.439, 0.493) 0.761 (0.720, 0.805) 0.811 (0.770, 0.855) 0.674 (0.638, 0.713) 0.684 (0.649, 0.721) 0.690 (0.655, 0.727) 
Head injuries, including TBIs 0.719 (0.546, 0.946) 0.647 (0.490, 0.854) 0.744 (0.549, 1.008) 0.804 (0.603, 1.073) 0.888 (0.673, 1.171) 0.679 (0.510, 0.903) 0.688 (0.524, 0.904) 0.649 (0.489, 0.862) 
Falls 0.594 (0.290, 1.215) 0.979 (0.463, 2.068) 1.022 (0.442, 2.365) 1.217 (0.547, 2.711) 1.015 (0.496, 2.077) 0.697 (0.336, 1.446) 0.575 (0.287, 1.149) 0.387 (0.174, 0.863) 
MVAs 0.707 (0.224, 2.229) 1.335 (0.361, 4.932) 0.572 (0.172, 1.901) 0.568 (0.180, 1.790) 0.853 (0.271, 2.687) 0.511 (0.162, 1.611) 0.575 (0.185, 1.783) 0.388 (0.105, 1.432) 
Fractures 0.466 (0.356, 0.609) 0.529 (0.411, 0.681) 0.796 (0.598, 1.058) 0.841 (0.643, 1.099) 0.777 (0.603, 1.000) 0.653 (0.503, 0.849) 0.591 (0.460, 0.760) 0.407 (0.305, 0.541) 
Seizure 0.844 (0.740, 0.962) 0.931 (0.810, 1.070) 1.283 (1.084, 1.517) 1.348 (1.156, 1.572) 1.502 (1.302, 1.734) 1.013 (0.874, 1.175) 0.924 (0.808, 1.056) 1.005 (0.882, 1.146) 
Odds ratio (95% CI) 15-day gap 30-day gap MPR 70% MPR 80% MPR 90% PDC 70% PDC 80% PDC 90% 
Admission 0.707 (0.665, 0.752) 0.746 (0.700, 0.795) 0.498 (0.466, 0.532) 0.747 (0.700, 0.797) 0.801 (0.753, 0.853) 0.650 (0.609, 0.694) 0.641 (0.603, 0.682) 0.648 (0.609, 0.690) 
ER visits 0.683 (0.648, 0.719) 0.700 (0.662, 0.739) 0.465 (0.439, 0.493) 0.761 (0.720, 0.805) 0.811 (0.770, 0.855) 0.674 (0.638, 0.713) 0.684 (0.649, 0.721) 0.690 (0.655, 0.727) 
Head injuries, including TBIs 0.719 (0.546, 0.946) 0.647 (0.490, 0.854) 0.744 (0.549, 1.008) 0.804 (0.603, 1.073) 0.888 (0.673, 1.171) 0.679 (0.510, 0.903) 0.688 (0.524, 0.904) 0.649 (0.489, 0.862) 
Falls 0.594 (0.290, 1.215) 0.979 (0.463, 2.068) 1.022 (0.442, 2.365) 1.217 (0.547, 2.711) 1.015 (0.496, 2.077) 0.697 (0.336, 1.446) 0.575 (0.287, 1.149) 0.387 (0.174, 0.863) 
MVAs 0.707 (0.224, 2.229) 1.335 (0.361, 4.932) 0.572 (0.172, 1.901) 0.568 (0.180, 1.790) 0.853 (0.271, 2.687) 0.511 (0.162, 1.611) 0.575 (0.185, 1.783) 0.388 (0.105, 1.432) 
Fractures 0.466 (0.356, 0.609) 0.529 (0.411, 0.681) 0.796 (0.598, 1.058) 0.841 (0.643, 1.099) 0.777 (0.603, 1.000) 0.653 (0.503, 0.849) 0.591 (0.460, 0.760) 0.407 (0.305, 0.541) 
Seizure 0.844 (0.740, 0.962) 0.931 (0.810, 1.070) 1.283 (1.084, 1.517) 1.348 (1.156, 1.572) 1.502 (1.302, 1.734) 1.013 (0.874, 1.175) 0.924 (0.808, 1.056) 1.005 (0.882, 1.146) 

ER, emergency room; MPR, medication possession ratio; MVA, motor vehicle accident; PDC, proportion of days covered; TBI, traumatic brain injury.

Table 4 reports the adjusted ORs and 95% confidence intervals for negative outcomes between adherent and non-adherent patients using multivariate logistic regression.

Table 4

Multivariate associations between medication adherence and epilepsy-related outcomes (n = 31 634a)

Odds ratio (95% CI) 15-day gap 30-day gap MPR 70% MPR 80% MPR 90% PDC 70% PDC 80% PDC 90% 
Admissions 0.775 (0.725, 0.827) 0.808 (0.755, 0.866) 0.796 (0.739, 0.857) 0.822 (0.767, 0.882) 0.872 (0.816, 0.932) 0.739 (0.689, 0.793) 0.729 (0.682, 0.779) 0.741 (0.693, 0.792) 
ER visits 0.750 (0.710, 0.793) 0.764 (0.721, 0.809) 0.831 (0.780, 0.886) 0.852 (0.803, 0.903) 0.893 (0.844, 0.944) 0.780 (0.735, 0.828) 0.789 (0.746, 0.835) 0.793 (0.750, 0.838) 
Accidents/injuries 0.639 (0.521, 0.785) 0.676 (0.553, 0.826) 0.897 (0.716, 1.123) 0.934 (0.756, 1.154) 0.928 (0.759, 1.134) 0.770 (0.625, 0.948) 0.757 (0.620, 0.924) 0.593 (0.478, 0.736) 
Seizures 0.917 (0.803, 1.046) 1.003 (0.871, 1.155) 1.431 (1.207, 1.697) 1.493 (1.278, 1.746) 1.640 (1.418, 1.895) 1.158 (0.997, 1.347) 1.051 (0.916, 1.205) 1.138 (0.996, 1.300) 
Odds ratio (95% CI) 15-day gap 30-day gap MPR 70% MPR 80% MPR 90% PDC 70% PDC 80% PDC 90% 
Admissions 0.775 (0.725, 0.827) 0.808 (0.755, 0.866) 0.796 (0.739, 0.857) 0.822 (0.767, 0.882) 0.872 (0.816, 0.932) 0.739 (0.689, 0.793) 0.729 (0.682, 0.779) 0.741 (0.693, 0.792) 
ER visits 0.750 (0.710, 0.793) 0.764 (0.721, 0.809) 0.831 (0.780, 0.886) 0.852 (0.803, 0.903) 0.893 (0.844, 0.944) 0.780 (0.735, 0.828) 0.789 (0.746, 0.835) 0.793 (0.750, 0.838) 
Accidents/injuries 0.639 (0.521, 0.785) 0.676 (0.553, 0.826) 0.897 (0.716, 1.123) 0.934 (0.756, 1.154) 0.928 (0.759, 1.134) 0.770 (0.625, 0.948) 0.757 (0.620, 0.924) 0.593 (0.478, 0.736) 
Seizures 0.917 (0.803, 1.046) 1.003 (0.871, 1.155) 1.431 (1.207, 1.697) 1.493 (1.278, 1.746) 1.640 (1.418, 1.895) 1.158 (0.997, 1.347) 1.051 (0.916, 1.205) 1.138 (0.996, 1.300) 

ER, emergency room; MPR, medication possession ratio; PDC, proportion of days covered. Covariates: age, gender, Charlson comorbidity index, region (Northeast, South, Midwest, West), payment type (commercial, Medicare, Medicaid, self-insured, other), plan type (HMO, PPO, POS, other), number of concomitant drugs excluding anti-seizure medications (classes include antianxiety, antidepressants, antimania, antipsychotics, and mood stabilizers), number of comorbidities, provider speciality or primary care, time from diagnosis to index date.

aOne patient with missing gender was excluded from the multivariate analysis.

For the outcomes of hospital admissions and ER visits, all ORs are <1.0, indicating reduced likelihood of the event for adherent patients. For accidents/injuries, adherence is again directionally associated with lower rates of accidents and injuries across all adherence measures. Some of the confidence intervals include one making it less certain that adherence is associated with a better outcome. For the seizure proxy outcome, directionally unexpected ORs were seen and the confidence intervals crossed 1.0 for all of the measures except the MPR measures.

Discussion

Previous work has found that poor adherence with AED therapy is associated with adverse clinical events [8–10, 13]. This article illustrates that the choice of adherence measures has implications for quality improvement purposes. These measures are consistent with those developed by the Pharmacy Quality Alliance (PQA) for β-blockers, angiotensin-converting enzyme inhibitors, calcium channel blockers, biguanide, sulfonylurea, thiazolidinedione and statins [14]. The PQA measures use the PDC and a single gap of ≥30 days.

Hundreds of articles in Pubmed have assessed adherence measures and many have examined single measures of adherence as predictors of adverse events. Only a few have compared measures within a single disease as predictors of adverse outcomes. Martin et al. [15] found that for evaluating medication class effects in schizophrenic patients, the PDC and the MPR were similar, with the biggest differences in patients who switched drugs within a class. Weiden et al. [16] in another study of schizophrenia found that all measures of non-adherence, including gaps in therapy of as little as 10 days and MPR were correlated with increased risk of hospitalization. In diabetics, Karve et al. [17] compared eight measures, including the MPR, gaps in therapy and PDC as predictors of hospitalization among diabetics and found that the MPR and PDC out-performed other measures in predicting the likelihood of hospitalization. Vink et al. [18] found little difference between gaps and therapy >30 days and MPR above 80% for diabetics.

This article compared the association of several adherence measures, using different thresholds, to examine possible outcomes. We find some differences in the associations across measures. Previous work has mainly focused on a single adherence measure, without consideration of different thresholds. The use of a standard 80% threshold, as is common in the literature, is often unsupported by evidence in a given therapeutic area. In this study, multiple thresholds were examined for each type of adherence measure. In the univariate analysis of the two most tangible outcomes, hospital admissions and ER visits, varying the adherence threshold level for MPR resulted in significantly different likelihood of adverse clinical outcomes. As the MPR threshold defining the adherent population rose from 70 to 80% to 90%, the likelihood of adverse outcome of the adherent group, while still lower than its associated non-adherent group, increased relative to that non-adherent group. In other words, the difference in the likelihood of these negative events was more apparent when comparing populations with MPR < 70% with those with MPR ≥ 70% than it was when comparing populations with MPR < 90% with those with MPR ≥ 90%. The PDC and gaps in therapy measures did not have any pattern of statistically significant ORs in response to varying thresholds.

These results can be used by organizations seeking to track the quality of care in their populations with epilepsy or by agencies responsible for endorsing quality measures. Other considerations for optimal quality measures, such as complexity of calculation and ease of interpretation, were not addressed here.

All of the adherence measures at all levels are in the expected direction, but the measures vary in the strength of their association with less common events. Narrow confidence intervals, indicating stronger measures, were found for admissions, ER visits, head injuries and factures. For the latter two outcomes, the confidence intervals were wider for the MPR, indicating that these are not as strongly associated with these outcomes. This is probably because of the small number of events in these outcomes. For seizures and falls, the MPR measures are not in the expected direction and the confidence intervals for all measures are large. In the case of falls, this is because of the small number of observed events. In the case of seizures, it is likely because diagnostic coding is insufficient to differentiate between new onset seizures and a history of seizures, making the outcome measure itself unreliable.

Limitations

There are a number of limitations that must be kept in mind when interpreting the results of this study. First, the associations between adherence and outcomes reported from this retrospective analysis cannot be assumed to be causal. Despite the incorporation of multivariate analysis, there may be other explanatory factors unobservable through claims data. Secondly, while claims data are a reliable measure of prescription refill behavior, they may not accurately reflect adherence in terms of actual medication-taking behavior. In particular, adherence will be overstated using MPR if patients are on multiple concomitant AEDs. Thirdly, the selected outcomes of interest in epilepsy could only be approximated using claims data, recognizing that actual clinical outcomes, such as the occurrence of seizures, are not reliably observable in these data. Until linkable electronic medical records containing full text from all payers—health insurance, automobile insurance and worker's compensation—are developed, it will never be possible to fully ascertain seizure outcomes. Using proxy measures means we assume that incorrect ascertainment of outcomes is not correlated with the types of measures. Finally, these adherence-based quality measures only report on an intermediate process outcome, not an actual clinically measured outcome. Until available databases and reporting systems allow more sophisticated measurement, proxy measures such as these are the only alternative. Fifthly, since we performed a cross-sectional analysis, we did not assess the timing of non-adherence relative to possible events. This would be very difficult to implement in a quality measure.

Conclusion

This article explores the performance of different population-level measures of adherence in epilepsy for use as quality measures. Both PDC and gaps in therapy produced significant adjusted ORs in the expected direction for the negative outcomes of admissions, ER visits, head injuries and fractures. More research needs to be conducted on the dynamics of MPR in epilepsy, given the results observed here and the recognized bias in MPR for cases of concomitant AED therapy.

Conflict of interest

M.D. and J.F. are employed by Ortho-McMeil Janssen Pharmaceuticals. M.J.G. and D.I.B. have conducted projects funded by the pharmaceutical industry.

Funding

This research was funded by Ortho-McNeil Janssen Pharmaceuticals Inc.

References

1
Institute of Medicine (U.S.). Committee on Quality of Health Care in America
Crossing the Quality Chasm: A New Health System for the 21st Century
 , 
2001
Washington, DC
National Academy Press
2
Hirtz
D
Thurman
DJ
Gwinn-Hardy
K
, et al.  . 
How common are the ‘common’ neurologic disorders?
Neurology
 , 
2007
, vol. 
68
 (pg. 
326
-
37
)
3
National Epilepsy Foundation
Epilepsy and seizure statistics
 
4
Sander
JW
The use of antiepileptic drugs—principles and practice
Epilepsia
 , 
2004
, vol. 
45
 
Suppl. 6)
(pg. 
28
-
34
)
5
Davis
KL
Prevalence and cost impact of nonadherence with antiepilepsy drugs among adults in a managed care population
Academy of Managed Care Pharmacy 19th Annual Meeting Showcase
 , 
2007
San Diego, CA
6
Jones
RM
Butler
JA
Thomas
VA
, et al.  . 
Adherence to treatment in patients with epilepsy: associations with seizure control and illness beliefs
Seizure
 , 
2006
, vol. 
15
 (pg. 
504
-
8
)
7
Cramer
JA
Glassman
M
Rienzi
V
The relationship between poor medication compliance and seizures
Epilepsy Behav
 , 
2002
, vol. 
3
 (pg. 
338
-
42
)
8
Manjunath
R
Davis
KL
Candrilli
SD
, et al.  . 
Association of antiepileptic drug nonadherence with risk of seizures in adults with epilepsy
Epilepsy Behav
 , 
2009
, vol. 
14
 (pg. 
372
-
8
)
9
Faught
E
Duh
MS
Weiner
JR
, et al.  . 
Nonadherence to antiepileptic drugs and increased mortality: findings from the RANSOM Study
Neurology
 , 
2008
, vol. 
71
 (pg. 
1572
-
8
)
10
Davis
KL
Candrilli
SD
Edin
HM
Prevalence and cost of nonadherence with antiepileptic drugs in an adult managed care population
Epilepsia
 , 
2008
, vol. 
49
 (pg. 
446
-
54
)
11
Hess
LM
Raebel
MA
Conner
DA
, et al.  . 
Measurement of adherence in pharmacy administrative databases: a proposal for standard definitions and preferred measures
Ann Pharmacother
 , 
2006
, vol. 
40
 (pg. 
1280
-
8
)
12
Vanderpoel
DR
Hussein
MA
Watson-Heidari
T
, et al.  . 
Adherence to a fixed-dose combination of rosiglitazone maleate/metformin hydrochloride in subjects with type 2 diabetes mellitus: a retrospective database analysis
Clin Ther
 , 
2004
, vol. 
26
 (pg. 
2066
-
75
)
13
Ettinger
AB
Manjunath
R
Candrilli
SD
, et al.  . 
Prevalence and cost of nonadherence to antiepileptic drugs in elderly patients with epilepsy
Epilepsy Behav
 , 
2009
, vol. 
14
 (pg. 
324
-
9
)
14
Pharmacy Quality Alliance
PQA approved measures
2010
 
15
Martin
BC
Wiley-Exley
EK
Richards
S
, et al.  . 
Contrasting measures of adherence with simple drug use, medication switching, and therapeutic duplication
Ann Pharmacother
 , 
2009
, vol. 
43
 (pg. 
36
-
44
)
16
Weiden
PJ
Kozma
C
Grogg
A
, et al.  . 
Partial compliance and risk of rehospitalization among California Medicaid patients with schizophrenia
Psychiatr Serv
 , 
2004
, vol. 
55
 (pg. 
886
-
91
)
17
Karve
S
Cleves
MA
Helm
M
, et al.  . 
An empirical basis for standardizing adherence measures derived from administrative claims data among diabetic patients
Med Care
 , 
2008
, vol. 
46
 (pg. 
1125
-
33
)
18
Vink
NM
Klungel
OH
Stolk
RP
, et al.  . 
Comparison of various measures for assessing medication refill adherence using prescription data
Pharmacoepidemiol Drug Saf
 , 
2009
, vol. 
18
 (pg. 
159
-
65
)

Supplementary data