Accurate Medication Adherence Measurement Using Administrative Data for Frequently Hospitalized Patients

Abstract

Background: Pharmacy administrative claims data remain an accessible and efficient source to measure medication adherence for frequently hospitalized patient populations that are systematically excluded from the landmark drug trials. Published pharmacotherapy studies use medication possession ratio (MPR) and proportion of days covered (PDC) to calculate medication adherence and usually fail to incorporate hospitalization and prescription overlap/gap from claims data. To make the cacophony of adherence measures clearer, this study created a refined hospital-adjusted algorithm to capture pharmacotherapy adherence among patients with end-stage renal disease (ESRD). Methods: The United States Renal Data System (USRDS) registry of ESRD was used to determine prescription-filling patterns of those receiving new prescriptions for oral P2Y12 inhibitors (P2Y12-I) between 2011 and 2015. P2Y12-I-naïve patients were followed until death, kidney transplantation, discontinuing medications, or loss to follow-up. After flagging/censoring key variables, the algorithm adjusted for hospital length of stay (LOS) and medication overlap. Hospital-adjusted medication adherence (HA-PDC) was calculated and compared with traditional MPR and PDC methods. Analyses were performed with SAS software. Results: Hospitalization occurred for 78% of the cohort (N = 46 514). The median LOS was 12 (interquartile range [IQR] = 2-34) days. MPR and PDC were 61% (IQR = 29%-94%) and 59% (IQR = 31%-93%), respectively. After applying adjustments for overlapping coverage days and hospital stays independently, HA-PDC adherence values changed in 41% and 52.7% of the cohort, respectively. When adjustments for overlap and hospital stay were made concurrently, HA-PDC adherence values changed in 68% of the cohort by 5.8% (HA-PDC median = 0.68, IQR = 0.31-0.93). HA-PDC declined over time (3M-6M-9M-12M). Nearly 48% of the cohort had a ≥30 days refill gap in the first 3 months, and this increased over time (P < .0001). Conclusions: Refill gaps should be investigated carefully to capture accurate pharmacotherapy adherence. HA-PDC measures increased adherence substantially when adjustments for hospital stay and medication refill overlaps are made. Furthermore, if hospitalizations were ignored for medications that are included in Medicare quality measures, such as Medicare STAR program, the apparent reduction in adherence might be associated with lower quality and health plan reimbursement.

Introduction

Pharmacy administrative claims data remain an accessible and efficient source to measure postmarketing drug’s effectiveness and safety in real-world patients. It is specifically important for patient populations that are systematically excluded from the landmark drug trials or groups of patients who are at excessive risk of adverse drug events and postmarketing safety data become important, for example, patients with end-stage renal disease (ESRD) on chronic dialysis. Commonly, hospital inpatient length of stay, refill overlap, and gaps are not always captured to capture pharmacotherapy exposure among patients who are more frequently hospitalized. Although at least 11 different medication adherence measures have been identified in claims-based adherence studies alone,^{1 -10} they failed to take into consideration the length of stay, overlap in refill, and gaps. Besides, frequent and prolonged hospitalizations may have gaps in treatment that can be misattributed to nonadherence. The most common measures of drug adherence are the medication possession ratio (MPR) and proportion of days covered (PDC).^6,11 The numerator in MPR is the sum of the days supply of a drug and the denominator is the number of days in the period. The numerator in PDC is the number of days in the period that were covered. PDC is preferred to MPR ¹² as the latter reflects a patient’s overall accordance with the prescribed medication while disregards the timeliness of refills and can overestimate adherence. PDC overcomes this limitation of MPR. ⁹

Neither of these methods adjusts the medication adherence calculation for hospital stays or overlaps in prescription refills, and all events are frequently observed when analyzing claims data, particularly among very sick patients. These adjustments may provide more accurate estimates of medication adherence for at-risk populations.

There are several knowledge gaps in the methods of calculating medication exposure. Previous pharmacoepidemiological studies failed to address adjustments in the calculation of MPR or PDC for hospitalization, few address prescription refill overlaps and gaps in prescription refills, and fewer still examine those patients who discontinue treatment after their first prescription.^{13 -15} Furthermore, recently, the European Society for Patient Adherence, Compliance, and Persistence (ESPACOMP) guidelines acknowledged these knowledge gaps and propose reporting standards in adherence research. ¹⁶ The Medication Adherence Reporting Guidelines (EMERGE) suggest enhancing reporting quality of methods in medication adherence research to standardize approaches, reduce research waste, accelerate progress in this field, and ultimately improve patient outcomes. ¹⁶

To make the cacophony of adherence measures clearer and to address the above-mentioned knowledge gaps, and following the ESPACOMP and EMERGE, we propose a new method for calculating medication adherence. Our measure, the hospital-adjusted PDC (HA-PDC), assesses the implementation component, where implementation is defined as the degree to which drug dosing in practice matches the prescribed dosing regimen, from treatment initiation until discontinuation. ¹⁷ Our aims are thus to (1) compare medication adherence as measured by MPR and traditional PDC to HA-PDC, (2) categorize gaps in prescription refill gaps and determine the timing of the refill gaps >30 days and to what extent they impact HA-PDC, and (3) determine to what extent HA-PDC changes over time in different landmark time points. To accomplish our objectives, we used a real-world cohort of ESRD patients from the United States Renal Data System (USRDS) registry data.

Methods

Data Source

The USRDS is a national database that tracks Medicare beneficiaries with ESRD. All adult ESRD patients are eligible for Medicare coverage, regardless of their age. Furthermore, nearly 80% of these patients use Medicare Part D benefits compared with <70% of the general Medicare population. ¹⁸ The Core Patients data set provides details on patient demographics, including death data. Separate data sets contain information on hospitalizations and Medicare Part D prescription claims at the individual patient level. ¹⁹ Administrative claims data are an accessible, reliable, and valid source to measure drug effectiveness in real-world patients. It is especially important for patient populations that are systematically excluded from landmark drug trials or groups of patients who are at excessive risk of adverse drug events. Within the USRDS, we used Medicare Part A and Part D pharmacy claims to capture the entire prescription-filling history for new prescriptions of oral P2Y₁₂ inhibitors (P2Y12-I) administered to this patient population. Prescription claims (Medicare Part D) are a separate component of the Medicare claims data in the USRDS registry and include information such as the drug name and strength, refill number, out-of-pocket costs, days of medication supplied, and quantity dispensed. Data source files are linked with a patient-specific USRDS identifier. We used this rich data source to capture the hospital admissions and the entire prescription-filling history for oral P2Y₁₂ inhibitors (P2Y12-I) for our population.

Patient Population

The prevalence of nonadherence among ESRD patients ranges from 12.5% to 98.6%. ² The large range in results represents the heterogeneity of methods assessing adherence: Patients undergoing chronic dialysis for ESRD have one of the highest pill burdens, taking a median of 19 pills a day. ²⁰ One of the more common drug classes are oral P2Y₁₂-Inhibitors (P2Y12-I)—antiplatelet agents (clopidogrel, prasugrel, and ticagrelor). These agents are the cornerstone of management of acute coronary syndrome, one of the leading causes of mortality in these patients. It has been estimated that 32% of ESRD patients are prescribed these agents, ²¹ and despite making up less than 3.5% of the population, chronic kidney disease (CKD) patients account for 49.3% to 60.4% of all Veterans Health Administration prescriptions for the 3 P2Y12-I agents. ²² These complicated regimens, together with the necessity of routine dialysis treatments and checkups, increase the risk of drug-related adverse effects, hospitalizations, and medication nonadherence in ESRD patients.^2,23 Given the high disease burden^20,24 and cost of treating these patients, pharmacoepidemiologic research on drug prescribing and utilization has the potential to improve public health at the national level.

ESRD patients are systematically excluded from the landmark clinical trials of commonly prescribed drugs so administrative data are commonly used to assess a drug’s performance for sicker patients like ESRD. Finally, USRDS is a registry of ESRD patients in the United States that tracks >80% of all patients on chronic dialysis in the United States and provides readily available, reliable, national patient data. We identified all adult ESRD dialysis patients (age = 18-100) in the USRDS who had corresponding records in the Medicare Part D data files and initiated P2Y12-I treatment between July 20, 2011, and December 30, 2015. Although 2 or more claims are commonly used for verifying prescription-dispensing, this strategy is more appropriate for longer term treatments like statins and antipsychotics.^6,25 Since P2Y12-I are commonly prescribed for 6 to 12 months only, we included patients with at least one P2Y12-I claim in our analyses. We chose our start date to coincide with the US Food and Drug Administration (FDA) approval date for ticagrelor to allow for the analysis of all 3 P2Y12-I drugs. Our end date was based on data availability. The index date was the first claim for one of the target agents, and patients were required to have at least 6 months of continuous enrollment prior to the index date. Given the high mortality in ESRD patients, ²⁶ we followed other pharmacoepidemiological studies and only included patients who began P2Y12-I treatment at least 6 months after their first dialysis service date. ²⁷ Exclusion criteria included missing date of first USRDS-recorded service, gaps in chronic maintenance dialysis treatments, gaps in Medicare enrollment or more than one type of P2Y12-I prescribed on the index date.

Study Design

Defining the observation period is critical for accuracy and affects the denominator value used to calculate hospital-adjusted PDC (HA-PDC). We defined our observation start date as the first medication fill date (index date). In our study, the first dialysis service date was used to identify claims for only those ESRD patients actively receiving dialysis. We also used modified Liu Index to report comorbidity burden. As patients may not be due for a refill on the same day as their first dialysis service date, P2Y12-I prescriptions filled in the previous 90 days (maximum days supply allowed by Medicare) were also included when identifying new prescriptions. The observation or service end date is often defined as the last refill date, including or excluding the last refill days supply, or is defined based on a predetermined study length time, eg, 1 year. The observation period was primarily influenced by the study goals. In our study, the patients were followed as long as possible until death, kidney transplantation, a change in P2Y12-I, or lost to USRDS follow-up. This resulted in patients with median follow-up of 367 (interquartile range [IQR] = 147-1627) days.

Calculating Medication Adherence

To calculate MPR, PDC and HA-PDC, we performed the following steps related to the prescription claims for the defined study cohort (N = 46 514) (Figure 1).

Figure 1.

Flagging days of medication exposure: Adjustment made to accurately capture medication exposure.

Identifying claims of interest

Initially, all prescription claims submitted for our cohort for all years were retrieved from the USRDS Statistical Analysis Files (SAFs). Each Medicare Part D prescription claim in the USRDS data set comes with USRDS patient ID, drug name and National Drug Code (NDC) codes, strength, dosage form, days supply, service start date, refill, and service end date. We identified P2Y12-I claims using the nonproprietary names (clopidogrel, prasugrel, and ticagrelor). Only prescriptions filled during ongoing dialysis treatment were considered. Claims for the P2Y12-I were directly identified by the list of generic names from the raw data set. We also used nonproprietary names to identify the number of medications these patients were taking on index date. While following these beneficiaries, we also captured dates of hospitalizations.

Defining new prescription accurately

The new prescription for a P2Y12-I was identified in 2 steps. First, all prescriptions for P2Y12-Is were identified using nonproprietary drug names from Medicare Part D. Second, any P2Y12-I prescription identified after a 6-month period of no P2Y12-I prescription was defined as a new prescription for a P2Y12-I (Figure 1). Because these patients were verified to be not on P2Y12-I for 6 months prior to the first fill date, we identified that day of the prescription dispensed to be the index date. The earliest possible index date was July 20, 2011. We focused on identifying new prescriptions to reduce prevalence bias. The claims data set was then limited to create the final cohort of new users only. All P2Y12-I prescription claims submitted on or after the index date were included in calculating HA-PDC variable.

Censoring events

Patients were followed until one of the following events happened: death, receipt of kidney transplantation, switching to a different P2Y12-I, initiation of oral anticoagulants, lost to USRDS follow-up or loss of Medicare Part D. If none of these events occurred, follow-up continued until the last date available in the data (December 31, 2015).

Flagging days covered

Each day of the observation period was flagged by an indicator variable: days covered by P2Y12-I prescription were flagged as “1” and days not covered flagged as “0”.

Final calculation of MPR and PDC

The equations below demonstrate the traditional way to calculate MPR and PDC:

$$MPR=\frac{Total\#DaysSupply}{Total\#DaysinStudyPeriod}$$

$$PDC=\frac{TotalDaysCovered}{Daysoffollow-up}$$

For example, to calculate PDC for a patient who was followed for 61 days from the index date of March 1 through April 30, one would flag each day as either 0 or 1. The sum of 0s and 1s for the period determines the total number of days covered by a P2Y12-I. The total number of days covered is then divided by the number of days of follow-up to provide the PDC value. In contrast, instead of days covered, MPR simply sums the days supply from each claim for the numerator, which could result in an adherence level greater than 100%.

Calculating HA-PDC after adjustments in PDC for various events

We developed an algorithm taking care of hospital admission and length of stay, overlapping coverage, and refill gaps on the traditional adherence measures, and to calculate the adjusted PDC, ie, HA-PDC value. Similar to MPR and PDC, larger values of HA-PDC indicate greater medication intake. The mean HA-PDC and IQR are better measures of the typical exposure as the distribution is skewed. ²⁸

Let i be the number for refills and D represent the days supply, then Σni=1Di = D1 + D2 + D3 + . . . + Dn = Total days covered.

$$\begin{array}{l}HospitalAdjusted-Proportionofdayscovered\\ =TotalDaysCovered/Daysoffollow-up\\ (considerhospitalization,overlap,refillgaps)\end{array}$$

Adjusting for refill overlap coverage

Service dates and days supply frequently need adjustments for overlapping coverage days as it is common occurrence in claims data. Overlap in coverage occurs when a patient fills a prescription ahead of the refill date. This allows a patient with a medication stock that lasts longer than the predicted duration based solely on refill dates. Therefore, overlap in coverage was assessed before calculating total days of supply. We report 2 different scenarios for this adjustment:

• Scenario 1: If the medication strength of the second overlapping prescription was the same as the first prescription and the second prescription was filled within the last 7 days of the end date of previous prescription, the second prescription service start date was adjusted to 1 day after the end date of the first prescription. Subsequently, the second prescription service end date was extended to account for the number of overlapping days (Figure 1). Subsequent prescriptions were similarly adjusted. The assumption here is that the patient will finish the first prescription before beginning the second.

• Scenario 2: If the second prescription was for a different strength of the same medication, the first prescription’s calculated end date (service date + days supply) was shortened to 1 day prior to the service date of the second prescription. (Figure 1). This change assumes that a patient stops the first prescription early due to change in therapy. No adjustment was made to the end date of the second prescription.

Adjusting for hospital admissions

To account for hospital admissions, it was assumed that patients continued to take P2Y12-Is during their hospital stay, and adherence is 100%. Days supply and service end dates were thus extended by the length of the hospital stay for prescriptions spanning hospitalization (Figure 1). We captured inpatient hospitalizations, length of stay for each hospitalization, and patient refill trajectory before and after hospitalization during the study period.

Adjusting for refill gaps using adherence levels

Patterns of gaps in prescription refills in the study cohort over time are shown in Figure 2. There were 3 major categories of medication adherence levels among patients who were followed and did not get censored at landmark time points: (1) 30+ day gap users (any HA-PDC plus refill gaps of ≥30 consecutive days), (2) continuous users (HA-PDC ≥0.80 plus no refill gaps of ≥30 consecutive days), (3) infrequent users (HA-PDC <0.80 plus no refill gaps of ≥30 consecutive days). These 3 groups were categorized as all 3 levels of medication adherence indicating unique medication adherence pattern. The distribution and patterns of HA-PDC in the 3 adherence level groups at 3, 6, 9, and 12 months landmarks are shown in Figure 4.

Figure 2.

First appearance of 30+ day gap in patients who had a 30+ day gap in the first year.

Figure 4.

Prescription refill patterns of P2Y12-I, measured using medication adherence levels at landmark time periods.

Statistical Analysis

The unit of analysis was the individual patient. Demographics and clinical characteristics were assessed. The number of distinct medications prescribed per patients was calculated, and a modified Liu Index was used as a measure of comorbidity burden. ²⁹ Continuous variables were summarized by mean and standard deviation when distributions were symmetric; otherwise, median and IQR were provided. For categorical variables, counts and percentages are provided. Kruskal-Wallis and χ² tests were used to compare baseline characteristics and to compare HA-PDC across the 3 P2Y12-I drugs. A 5-number summary is provided for the number of observed hospitalizations, hospitalization length of stay, length of follow-up, number of medications observed at baseline, all-cause deaths, and days of P2Y12-I exposure. Furthermore, refill gap and exposure levels are reported at 3-month, 6-month, 9-month, and 12-month periods for those who were followed at least 3, 6, 9, or 12 months, respectively (Figures 2 and 3). Analyses were generated with SAS software, Version 9.4 of the SAS System for Windows (SAS Institute Inc, Cary, NC).

Figure 3.

Patterns of gaps in prescription refill in the study cohort over time.

Results

The final study cohort included 46 514 patients with 95% on clopidogrel (n = 44 280), 3% on prasugrel (n = 1237), and 2% on ticagrelor (n = 997) (Table 1). The CONSORT diagram (Figure A1 in Supplemental Material) describes the derivation of our study cohort from the USRDS data set. Median study follow-up was 367 (IQR = 147-728) days. The median age for patients receiving P2Y12 with was 64 (IQR = 55-73) years. Among them, 54% were men, 41% were Caucasians. The median modified Liu Index, a validated measure of comorbidity disease burden, was 7 (IQR = 4-10).

Table 1.

Characteristics of the Study Population.

	All	Clopidogrel	Prasugrel	Ticagrelor	P-value a
N	46 514	44 280	1237	997
Age, b median (IQR)	64.0 (55.0-73.0)	64.0 (55.0-73.0)	60.0 (52.0-68.0)	64.0 (56.0-72.0)	<.0001
Categories of age, No. (%)
18-64	23 489 (50.5%)	22 175 (50.1%)	804 (65.0%)	510 (51.2%)
65-75	14 748 (31.7%)	14 074 (31.8%)	360 (29.1%)	314 (31.5%)
>75	8277 (17.8%)	8031 (18.1%)	73 (5.9%)	173 (17.4%)
Sex, male, No. (%)	25 112 (54.0%)	23 811 (53.8%)	751 (60.7%)	550 (55.2%)	<.0001
Ethnicity, Hispanic/Latino, No. (%)	8673 (18.6%)	8265 (18.7%)	244 (19.7%)	164 (16.4%)	.3259
Race, No. (%)					<.0001
African American, Black	16 685 (35.9%)	16 053 (36.3%)	328 (26.5%)	304 (30.5%)
Caucasian	18 901 (40.6%)	17 853 (40.3%)	583 (47.1%)	465 (46.6%)
Hispanic, White	8296 (17.8%)	7897 (17.8%)	241 (19.5%)	158 (15.8%)
Other race	2530 (5.4%)	2380 (5.4%)	82 (6.6%)	68 (6.8%)
Residence location, No. (%)					.0014
Metro 1mill+	24 637 (53.0%)	23 510 (53.1%)	612 (49.5%)	515 (51.7%)
Metro <1mill or adjacent Metro	19 186 (41.2%)	18 208 (41.1%)	564 (45.6%)	414 (41.5%)
Rural or urban, not adjacent Metro	2613 (5.6%)	2484 (5.6%)	61 (4.9%)	68 (6.8%)
Unknown	78 (0.2%)	78 (0.2%)
LIS at any time, No. (%)	8212 (17.7%)	7973 (18.0%)	217 (17.5%)	22 (2.2%)	<.0001
Continuously eligible for Parts A, B, D, without LIS, without Medicare advantage, c No. (%)	19 247 (41.4%)	18 333 (41.4%)	540 (43.7%)	374 (37.5%)	.0123
Medicare advantage at any time pre-index, No. (%)	65 (0.1%)	60 (0.1%)	5 (0.4%)	0 (0.0%)	.0218
History of smoking tobacco, No. (%)	3213 (6.9%)	3072 (6.9%)	79 (6.4%)	62 (6.2%)	.5111
Dialysis-related factors
Etiology of end-stage renal disease, No. (%)					.0178
Diabetes	25 919 (55.7%)	24 618 (55.6%)	721 (58.3%)	580 (58.2%)
Glomerulonephritis/chronic kidney disease	3279 (7.0%)	3110 (7.0%)	104 (8.4%)	65 (6.5%)
Hypertension	11 643 (25.0%)	11 131 (25.1%)	265 (21.4%)	247 (24.8%)
Other	5348 (11.5%)	5111 (11.5%)	138 (11.2%)	99 (9.9%)
Dialysis modality, No. (%)					<.0001
Hemodialysis	43 234 (92.9%)	41 244 (93.1%)	1087 (87.9%)	903 (90.6%)
Peritoneal dial	3280 (7.1%)	3036 (6.9%)	150 (12.1%)	94 (9.4%)
Dialysis vintage, y, median (IQR)	3.8 (1.9-6.7)	3.9 (1.9-6.7)	3.6 (1.7-6.3)	3.6 (1.8-6.2)	.0070
Modified Liu Index (180), median (IQR)	7.0 (4.0-10.0)	7.0 (4.0-10.0)	7.0 (4.0-9.0)	7.0 (4.0-10.0)	.0581
Medication exposure level, median (IQR)	0.68 (0.31-0.93)	0.68 (0.30-0.93)	0.69 (0.34-0.94)	0.80 (0.43-1.00)	<.0001

Note. LIS = low income subsidy; IQR = interquartile range.

^aP-values were calculated by Kruskal-Wallis test or χ² test.

^bAges were calculated as at the P2Y12-I Index date.

^cAll eligibility was tabulated for 6 months prior to the Index date.

Hospitalization occurred in 78% of the cohort, and the median length of stay was 12 (IQR = 2-34) days. The median number of concurrent medications on the index date was 7 (IQR = 5-10) (Table 2). A total of 15 684 (33.7%) people died, and the median time to death was 280 (IQR = 113 558) days. Besides antiplatelet agents, the most common concomitant medications were phosphate binders, calcium channel blockers, beta blockers, ACE inhibitors, and statins (Table A2 in Supplemental Material).

Table 2.

Profile of Patients With End-Stage Renal Disease Cohort (N = 46 415).

Variable	No. of eligible counts	Minimum	25th percentile	Median	75th percentile	Maximum
Number of hospitalizations per patient	35 403	0	1	2	5	88
Length of stay hospital, d	35 403	0	2	12	34	958
Study follow-up, d	46 514	1	147	367	728	1625
No. of medications a	46 514	1	5	7	10	35
All cause death, b d	15 684	1	113	280	558	1593
Medication exposure, d	46 514	1	53	143	351	1583

^aNumber of medications patients were taking prior on index date.

^bAll-cause death occurred from index date.

The median (IQR) MPR was 61% (29%-94%) and median (IQR) PDC was 59% (31%-93%). Patients had higher adherence values when adjusted for overlapping coverage and hospital stays (Table 3). For the 19 129 patients with overlapping prescription fill dates, HA-PDC was higher by a median of 1.08% from traditional PDC. Similarly, the HA-PDC value was higher by a median of 5% in patients who were hospitalized (n = 24 547). When adjusting for overlap coverage and hospital stay simultaneously, there was a median increase in HA-PDC of 5.8% in the 67% (IQR= 30%-92%) of patients affected. Median HA-PDC was 68% (IQR = 31%-93%) compared with a PDC of 59% (IQR = 31%-93%). There was a decline in median HA-PDC of the overall cohort over time: 91.11% at 3 months, 75.00% at 6 months, 68.52% at 9 months, and 65.56% at 12 months (Figure A2 in Supplemental Material).

Table 3.

Comparison Among MPR, PDC, and Hospital-Adjusted PDC (HA-PDC).

Statistics	MPR a	PDC a	HA-PDC a
All P2Y12 patients, N	46 514	46 514	46 514
Median (IQR)	0.61 (0.29-0.94)	0.59 (0.31-0.93)	0.68 (0.31-0.93)
Change in HA-PDC after sequential adjustments
Adjustments made	Population impacted, No. (%)		HA-PDC, median (IQR)
Overlap adjusted for HA-PDC	19 129 (41%)		0.65 (0.29-0.91)
Hospitalization adjusted for HA-PDC	24 547 (53%)		0.67 (0.30-0.92)
Hospitalization and supply overlap adjusted concurrently for HA-PDC	31 218 (67%)		0.68 (0.31-0.93)

Note. Overall median HA-PDC: 91.11 at 3 months, 75.00 at 6 months, 68.52 at 9 months, and 65.56 at 12 months. MPR = medication possession ratio; PDC = proportion of days covered; HA-PDC = hospital-adjusted medication adherence; PDC = proportion of days covered; MPR = medication possession ratio.

^aUsing PDC, MPR, and HA-PDC formula shown in “Methods” section.

A refill gap of ≥30 days was observed in more than 50% of the cohort (n = 29 245, 63%) during this study period. For patients who had a ≥30 days refill gap within the first year of starting a new prescription for P2Y12-I, nearly half of them (48% of the entire cohort) had a ≥30 days refill gap in the first 3 months (Figure 2).

Based on adherence levels, we categorized patients into 3 groups: (1) continuous users were those who <HA-PDC ≥0.80 plus no refill gaps of ≥30 consecutive days >, (2) infrequent users were defined as <HA-PDC <0.80 plus no refill gaps of ≥30 consecutive days >, and (3) 30+ day gap users < any HA-PDC plus refill gaps of ≥30 consecutive days >. Figure 2 shows the 3 level of adherence patterns over time. There was a decline in proportion of patients in the continuous users group: from 60% (n = 23 173) at 3 months, to 31% (n = 7286) at 12 months, P < .0001. Finally, the proportion of patients in the 30+ day gap user group increased over time: 29% (n = 11 186) at 3 months to 68% (n = 15 999) at 12 months, P < .0001 (Figure 3).

The distribution and patterns of HA-PDC in the 3 adherence level groups (continuous, infrequent, 30+ day gap) at 3, 6, 9, and 12 months landmarks are shown in Figure 4. Box and Whisker Plots display the 5-number summary of HA-PDC with the minimum, quartiles, median differences with full ranges to representation over landmark points.

Discussion

Our use of the USRDS database provided the opportunity to evaluate the effects of hospitalizations, prescription refill overlaps, and gaps on traditional measures of medication adherence. Based on our enhanced precision and reproducibility in calculation for medication exposure, accurate HA-PDC for P2Y12-I was 68%. HA-PDC value was higher than the PDC value by a median of 5.8% after adjusting for overlap coverage and length of hospital stay. We developed a novel analytical algorithm with hospitalization to calculate medication adherence from prescription claims data and report our methods in depth. We applied our method to oral P2Y₁₂ inhibitors as this is a critical drug for the ESRD population which experiences rates of cardiovascular mortality that are 10 to 20 times higher than matched controls without ESRD. ⁸

Our study advances the current understanding of methods for measuring drug exposure by taking common details such as overlaps in prescriptions and hospital stays into account which has not been well defined previously using administrative data.^6,30,31 As ESRD patients commonly take >7 prescription medications on a single day, are hospitalized frequently (78%) with a median length of hospital stay of 7 days, and have high mortality rate, our methods are particularly more comprehensive than previously reported adherence matrix. Therefore, if a study fails to incorporate frequent and prolonged hospitalizations in adherence measures, outcomes can be falsely misattributed to nonadherence. Furthermore, if hospitalizations were ignored for medications that are included in Medicare quality measures, such as Medicare STAR program, the apparent reduction in adherence might be associated with lower quality and health plan reimbursement.

We found that PDC values, when adjusted for overlapping coverage and length of hospital stay, increased by a median of 5.8%. In addition, while adherence levels tended to fall over time, patients were most likely to have a refill gap of greater than 30 days in the first 3 months after treatment initiation. It is imperative for these patients to not disrupt P2Y12-I in excess of 7 days based on the previous study results and pharmacokinetics of P2Y12-Is. ³² Platelet aggregation and bleeding time gradually return to baseline after discontinuation: for clopidogrel, it is 5 days; for prasugrel, 5 to 9 days; and for ticagrelor, 3 days postdiscontinuation.^32,33

Previous studies, which are used for calculating traditional MPR and PDC values, did not include full adjustments for overlapping refill periods, gaps in refill, and hospital stays. Some studies included hospitalization as a covariate in the final analyses only;^{4,10,31,34,35} others adjusted for overlapping coverage days by moving the start date of overlapping prescriptions but failed to address the issue of refills of different strengths.^{36 -38} Only a select few studies attempted to account for hospital/institutional stays.^{36 -38} They did this by shortening the study period by the corresponding length of hospital stay instead of adjusting the number of days covered by medication.^39,40 Medication oversupply affects the medication exposure numerator. We made rigorous attempts in refining our algorithm to accurately calculate medication adherence levels from prescription event files and, hospitalization information, prescription overlap information, days supply, and gap data to reflect precise patients’ medication adherence trajectory. Standardized methods needed in calculating medication exposure in pharmacoepidemiological studies. Particularly, this is also very relevant to special patient populations where randomized controlled trials are limited due to their systemic exclusion, and pharmacoepidemiological studies are the main source of evidence-based patient management. Future studies should determine whether changes in algorithms to calculate medication exposure level affect outcomes analyses.

Our study has several strengths. Medicare Part D claims with Part A event files were used to calculate medication exposure level of P2Y12-I prescriptions using refined algorithms to incorporate medication overfill coverage days and length of hospital stay. Prior studies have failed to address these common occurrences. Moreover, our findings apply to almost all ESRD patients in the United States as we used USRDS data set where patients with ESRD are eligible for Medicare coverage based on ESRD status and use Medicare Part D benefits more (80%) compared with general Medicare population (<70%). Our study has several limitations. First, patients with ESRD are included; therefore, this measurement may not be applicable to other population with different medication insurance coverage or to populations outside of the United States where hospitalization and prescription patterns may be very different. Second, for the purpose of our study, we censored patients if they switched from one P2Y12-I to another to reflect medication-specific adherence. Although this technique is common in comparative effectiveness research, it may not be generalizable to other countries as hospitalization and drug use pattern differ. Third, although 2 or more claims are commonly considered for dispensing-verification using prescription claims data for medications used for chronic conditions (eg, statin and antipsychotics),^6,21 we included patients with one P2Y12-I prescription claim since P2Y12-I prescriptions are given for short term (6-12 months) compared with other long-term and potentially lifelong therapies. This allowed us to quantify the number of patients never refilling their prescriptions. As this is often due to high costs or adverse effects, ³⁷ it may indicate an avenue for intervention. Last, a claims-based adherence measure makes the assumption that the patient is taking the drugs and taking them correctly—we have no way of verifying medication intake.

Similar to other studies,^37,38 we have shown that adherence decreases over time. A patient is most likely to have a gap of greater than 30 days between refills in the first 3 months of treatment. In future research, we would like to compare the HA-PDC measures across different drugs taken by these patients to check for internal consistency. We would also like to test our method on different patient groups such as those with chronic obstructive pulmonary disease. Planned future research will determine to what extent HA-PDC predicts effectiveness and safety outcomes compared with traditional PDC.

Conclusion

Our results indicate that HA-PDC, ie, traditional PDC adjusted for hospitalization and overlapping coverage, provides a more accurate determination of medication adherence using administrative data set. Hospitalization days, overlapping coverage of prescription refills, and refill gaps should be described in depth by researchers until guidelines standardize the methods of medication adherence ascertainment using prescription claims data. Standardization of medication exposure measurements and transparent algorithm reporting will make future studies more comparable and also allow for the creation of meaningful intervention strategies that improve medication adherence, reduce costs, and ultimately, improve health outcomes.