Abstract
Background
There is considerable evidence that medication non-adherence is associated with higher healthcare costs. Payers and providers often target adherence quality measures for intervention, but the impact may depend on the medication. This study sought to assess differences in the relationship between medication adherence and medical (i.e., non-pharmacy) costs across three quality measures in a Medicare Advantage sample.
Methods
An observational study was conducted among cohorts of Medicare Advantage beneficiaries using the 2018-19 Optum’s de-identified Clinformatics® Data Mart Database. Cohort assignment was based on inclusion in one or all three of the Pharmacy Quality Alliance’s adherence measures for (1) renin-angiotensin-system antagonists, (2) statin, and (3) diabetes medications. Medication adherence was measured in year 1 and payer medical costs were measured in year 2. Generalized linear modeling (Gamma distribution and a log link) with interaction terms and coefficient contrasts were used to assess the relationship between adherence and subsequent payer medical costs, and to evaluate differences in this relationship across the three measures. Analyses were adjusted for sociodemographic, clinical, prescription-related, and insurance-related variables.
Results
The single-measure cohort included 1,001,316 beneficiaries, and the three-measure cohort consisted of 284,137 beneficiaries. There were negative relationships (p < 0.0001) between adherence and payer medical costs for all medication classes. These associations were stronger for diabetes medications (0.9808; p < 0.0001), followed by renin-angiotensin-system antagonists (0.9874; p < 0.0001) and statin medications (0.9919; p < 0.0001) in the single-measure cohort. The findings were similar among beneficiaries using all three medication classes.
Conclusions
Better adherence was associated with lower payer medical costs across several therapeutic areas, providing additional evidence of the importance of adherence in managing health care costs. The relationship between adherence and subsequent year medical costs appears stronger for diabetes medications relative to renin-angiotensin-system antagonists or statins. Further research may explore interventions to increase adherence to diabetes medications to improve diabetes management.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12913-025-13331-2.
Keywords: Medication adherence, PDC, Medical costs, Medicare advantage
Background
Despite the positive effects of medications on both short and long-term health, it is estimated that up to 50% of patients with chronic conditions in the United States (US) are non-adherent to their medications [1–5]. The economic consequences associated with non-adherence are substantial, with estimates of the cost to the US health care system ranging from $100 to $300 billion annually [3, 4, 6]. The drug-related morbidity and mortality costs associated with all non-optimized medication therapy in the US, which includes the impact of non-adherence, have been estimated to be over $500 billion annually [7]. The annual impact of non-adherence to the U.S. Medicare fee-for-service program alone, based on an analysis of beneficiaries aged 65 years or older, was recently estimated to be close to $30 billion [8]. Despite increased drug costs, higher medication adherence has been shown to be associated with reduced health care utilization and lower medical costs in different populations and in patients with various conditions [9–15], including among older adults.
As the US health care system continues to move toward value-based care delivery and payment [16], those accountable for improving the quality and affordability of health care seek to close care gaps, such as those associated with medication non-adherence. The Pharmacy Quality Alliance’s (PQA’s) three adherence measures are often targeted for intervention by payers and providers and since 2012 have been included in the Centers for Medicare & Medicaid Services (CMS) Medicare Part D Star Ratings program [17]. The Medicare Part D program provides prescription drug coverage for Medicare enrollees (such as older adults and those with disabilities) in the United States and the star ratings program provides financial incentives to health plans in exchange for initiatives that can improve healthcare quality. These measures assess adherence to three groups of medications that treat chronic conditions: (1) renin-angiotensin-system antagonists (RASAs), (2) statins, and (3) diabetes medications (except insulin). The focus on these quality measures has led US health plans to implement interventions such as medication therapy management, targeted at improving medication adherence among older adults, who comprise most of the Medicare beneficiaries. These efforts have led to meaningful improvements in medication adherence in these chronic conditions among Medicare beneficiaries with corresponding cost avoidance. For example, CMS estimates that during 2013–2018, adherence improvements associated with these three measures led to costs avoided of up to $13.7 billion for statins, up to $7.2 billion for diabetes medications, and up to $25.7 billion for RASAs [18].
These estimates suggest that the impact of targeted interventions on these adherence quality measures by payers and providers may depend on the medication class. However, the overall cost avoided estimates provided by CMS are somewhat reflective of the number of beneficiaries receiving medications from each class. As the number of quality measures grows, a natural question is whether the per beneficiary relationship between medication adherence and healthcare costs differs across these three quality measures. Empirical evidence about the differences among these quality measures can help health plans prioritize which measure to focus on to achieve the most cost savings. In addition, given increases in adherence over the years in medications falling into these classes, it is important to evaluate this relationship on an ongoing basis. Thus, the purpose of this study was to estimate differences in the relationship between medication adherence and subsequent year payer medical costs across the three medication adherence quality measures used in the Medicare Part D Star Ratings program. Given that the potential for cost savings may be related to the number of chronic medications used by an individual, differences across medication classes were assessed among beneficiaries on sole therapy (i.e., being included in a single quality measure), and among beneficiaries on medications from all three medication classes (i.e., being included in all three quality measures).
Methods
Study design, data source, and sample
This observational study used de-identified administrative claims data from Optum’s de-identified Clinformatics® Data Mart Database (CDM). The CDM database contains healthcare service billing claims for beneficiaries enrolled in commercial and Medicare Advantage (MA) plans. Medicare Advantage is a federal health insurance program for Medicare beneficiaries (such as older adults, individuals with disabilities, etc.) who choose to receive their healthcare services through private insurers. The data are captured from various settings of care such as hospital, outpatient care, pharmacy, and laboratory services. All datasets are linked by an encrypted identifier and can be linked to data with individual enrollment details. Four groups of Medicare Advantage beneficiaries within the CDM were formed based off their eligibility for inclusion in the three medication adherence quality measures: [18–23] [1] beneficiaries only eligible for Proportion of Days Covered: RASA (RASA measure only group), [2] beneficiaries only eligible for Proportion of Days Covered: Statins (statin measure only group), [3] beneficiaries only eligible for Proportion of Days Covered: Diabetes All Class (diabetes measure only group), and [4] beneficiaries eligible for all three adherence measures (hereafter referred to as the three-measure cohort). All four groups were mutually exclusive.
CDM data from July 2017 to December 2019 were used for the study. Beneficiaries were evaluated for 30 months, which includes a 6-month lookback period, a 12-month assessment period for medication adherence measurement, and a 12-month follow-up period for assessment of payer medical costs. These time frames were selected to reflect the way adherence measures are used in the marketplace and how health plans are evaluated in the CMS Star Ratings Program. Thus, the study was designed from the perspective of a health plan. In addition, using subsequent year payer medical costs (i.e., lagging the adherence ◊ payer medical costs relationship by one year) may help to reduce the potential consequences associated with reverse causality. While this one-year lag period may not be ideal in all circumstances, it does have meaning from the payer perspective based on how these measures are used in practice.
Because eligibility for group inclusion was adapted from the adherence quality measure specifications, individuals were included if they were 18 years or older on the date of the first prescription fill in 2018 for a medication included in the adherence measures, continuously enrolled in MA medical and pharmacy insurance coverage in 2018, did not have a recorded date of death during the assessment period (i.e., 2018), and had at least two prescription claims in 2018 for a medication included in the adherence measures. Individuals who had end-stage renal disease (ESRD), or received hospice care during the assessment period were excluded. Measure-specific exclusion criteria were also implemented – i.e., individuals who had claims for insulin products were excluded from the diabetes measure, and those with sacubitril/valsartan claims were excluded from the RASA measure. Institutional review board approval was not required because the commercially available data source used for the study (the CDM database) only contains de-identified health information consistent with a U.S. Health Insurance Portability & Accountability Act (HIPAA) expert de-identification determination. Therefore, the study was not considered human subjects research and it was not necessary to obtain informed consent for use of this non-identifiable data, consistent with U.S. 45 CFR 46.102.
Study measures
The primary independent variable was medication adherence in the assessment period (i.e., 2018). Proportion of Days Covered (PDC) [24], as adapted from the quality measures [17], was used to measure medication adherence. The medications included in the adherence calculation for each quality measure was consistent with the lists provided by CMS [17] and were identified in the CDM data using national drug codes (NDCs). PDC is estimated as the proportion of days covered by prescription claims for a medication over a specified period of time. PDC, and other claims-based adherence measures, may not perfectly represent actual medication consumption, but they are feasible, use standardized data that are widely available, and are the dominant mechanism to characterize medication-taking behavior for quality measures [22, 23]. The limitation that PDC does not fully reflect an individual’s consumption of medication is known; however, despite these limitations, PDC as a measure of adherence has been consistently associated with beneficial health and economic outcomes [14, 15, 25]. PDC was measured from the date of the index prescription to the end of the assessment period and was treated as a continuous variable in the current analysis.
Because this analysis was from the perspective of the payer and because higher PDCs are generally associated with higher payer prescription costs (even when using subsequent year costs), payer medical costs was the key dependent variable. Total medical costs were estimated as the sum of all-cause costs for each medical service category (e.g., inpatient, outpatient) during the follow-up period (i.e., 2019). Any claim with a negative claim line amount for copays, deductibles, coinsurance, and total cost was excluded. The total medical cost estimate included both patient out-of-pocket (OOP) and payer costs. Payer medical costs were then estimated as patient OOP medical costs subtracted from total medical costs. For those beneficiaries not enrolled in the follow-up period for a full 12 months, costs were annualized.
Covariates adjusted for in the statistical models were selected based on the Andersen Behavioral Model [26]. The Andersen Behavioral Model suggests that individual factors affecting health outcomes may be classified as predisposing factors, need factors, and enabling factors. Guided by this framework, the following groups of covariates were measured during the assessment period: Predisposing factors – age, sex, race/ethnicity, geographic region; Need factors – treatment naïve status (based on whether the medication included in the measure was dispensed during the six months prior to the index prescription), medication burden (i.e., total number of unique medications with prescription days’ supply greater than or equal to 28 during the year), comorbidity burden (assessed using the Deyo-Charlson Comorbidity Index); Enabling factors – plan type (i.e., Health Maintenance Organization, Preferred Provider Organization, other), status as receiving either low-income subsidy or being dual eligible for Medicare and Medicaid, use of 90-day prescription fills, and mail-order pharmacy use.
Statistical analysis
Demographic and clinical characteristics of each study group during the assessment period were described using frequencies (and percentages) for categorical variables and means (and standard deviation) for continuous variables. Multivariable generalized linear modeling was used to assess the adjusted association between medication adherence and subsequent year payer medical costs and to assess differences in this relationship among the three adherence quality measures. Based on results from the modified Park test, the Gamma distribution along with the log link were used for all models. For data from the single-measure cohort (i.e., beneficiaries qualifying for RASA measure only, statin measure only, or diabetes measure only), a single data set was created with a categorical variable indicating group membership (i.e., 1, 2, or 3). To compare across multiple quality measures (i.e., assess for differences in the magnitude of coefficients across medication groups to examine whether relationship between PDC in the assessment period and payer medical costs in the follow-up period vary across RASA, statin, and diabetes medication groups), interaction terms between PDC in the assessment period and the multicategorical variable indicating group membership were utilized. The following covariates were allowed to be group-specific by including interaction terms: treatment naïve status, mail-order pharmacy use, and 90-day prescription fill use. For beneficiaries in the three-measure cohort, all three adherence measures were used as independent variables and payer medical costs during the follow up was used as the dependent variable. In this case, coefficient contrasts were utilized to assess for differences in the relationship between PDC in the assessment period and payer medical costs in the follow-up period across three adherence measures (see Additional File 1 for more information about the statistical models).
To aid in interpretation, the independent variable was scaled to represent a 5% change in PDC rather than a unit change (i.e., 1%) in the value for all models. Thus, effects are expressed as a mean ratio of the dependent variable when comparing two different values of PDC 5% points apart, adjusted for covariates. In other words, how much does the adjusted mean payer medical costs in the follow-up period multiply by for each 5%-point increase in PDC in the assessment period?
Regression models excluded beneficiaries from the study population who were not enrolled at any time during the follow up period (January – December 2019) or had a lack of continuous enrollment at the beginning of the follow up period. Missing values on categorical covariates were assigned an additional category for that variable (and an additional dummy variable in the set of dummy variables used to represent that covariate). In all regression models, beneficiaries with missing data on all other variables (i.e., other than categorical covariates) were excluded from analysis (i.e., listwise deletion). Overall, less than 5% of cases were excluded because of missing data in either of the multivariable analyses. All analyses were conducted using SAS version 9.4 (SAS Inc, Cary, NC). All hypothesis tests were 2-sided with an a priori significance level of 0.05.
Results
After applying all study inclusion and exclusion criteria, a total of 47,091 beneficiaries were included in the diabetes only group, 443,831 in the RASA onl y group, and 510,394 in the statins only group; thus, the single-measure cohort included 1,001,316 beneficiaries. The patient attrition table is presented in Additional File 2. There were 284,137 beneficiaries in the three-measure (i.e., RASA, statin, and diabetes) cohort. Across cohorts, a majority of the study population was female (range: 50.1–60.7%) and non-Hispanic White (range: 56.3–70.9%), with a mean age of 73.6 years. The median costs (and interquartile range) for individuals in the diabetes-only cohort, RASA-only cohort, statins-only cohort, and the multiple-measure cohorts were $6,397.50 (IQR: $2,011.90-$20,216.00), $5,595.50 (IQR: $1,811.60-$17,604.00), $6,106.20 (IQR: $2,148.90-$18,283.00), and $5,835.40 (IQR: $1,962.70-$15,561.00), respectively. Additional descriptive statistics are provided in Table 1 for the single-measure cohort and in Additional File 3 for the three-measure cohort.
Table 1.
Descriptive statistics for single-measure cohort
| Variable | Group | ||
|---|---|---|---|
| RASA Measure Only (N = 443,831) |
Statin Measure Only (N = 510,394) |
Diabetes Measure Only (N = 47,091) |
|
| n (%) | n (%) | n (%) | |
| Age (mean, SD) | 74.0 (8.7) | 73.9 (8.2) | 72.1 (9.9) |
| Sex | |||
| Female | 269,557 (60.7) | 287,244 (56.3) | 25,722 (54.6) |
| Male | 161,017 (36.3) | 207,052 (40.6) | 19,988 (42.4) |
| Unknown | 13,257 (3.0) | 16,098 (3.2) | 1,381 (2.9) |
| Race/ethnicity | |||
| Non-Hispanic White | 293,188 (66.1) | 361,887 (70.9) | 27,783 (59.0) |
| Non-Hispanic Black | 61,279 (13.8) | 54,941 (10.8) | 7,593 (16.1) |
| Hispanic | 54,021 (12.2) | 50,744 (9.9) | 7,348 (15.6) |
| Asian | 13,591 (3.1) | 17,923 (3.5) | 2,089 (4.4) |
| Unknown | 21,752 (4.9) | 24,899 (4.9) | 2,278 (4.8) |
| LIS/DE Status | |||
| Yes | 88,412 (19.9) | 96,424 (18.9) | 11,754 (25.0) |
| No | 355,419 (80.1) | 413,970 (81.1) | 35,337 (75.0) |
| Geographic Region | |||
| Midwest | 72,663 (16.4) | 93,437 (18.3) | 6,974 (14.8) |
| Northeast | 49,408 (11.1) | 64,262 (12.6) | 5,412 (11.5) |
| South | 204,447 (46.1) | 219,097 (42.9) | 23,311 (49.5) |
| West | 114,859 (25.9) | 131,152 (25.7) | 11,179 (23.7) |
| Unknown | 2,454 (0.6) | 2,446 (0.5) | 215 (0.5) |
| Plan Type | |||
| HMO | 126,096 (28.4) | 142,438 (27.9) | 12,570 (26.7) |
| PPO | 52,588 (11.8) | 57,254 (11.2) | 4,485 (9.5) |
| Other | 262,826 (59.2) | 308,410 (60.4) | 29,833 (63.4) |
| Unknown | 2,321 (0.5) | 2,292 (0.4) | 203 (0.4) |
| Treatment Naïve Status | |||
| Yes | 63,677 (14.3) | 81,749 (16.0) | 9,963 (21.2) |
| No | 380,154 (85.7) | 428,645 (84.0) | 37,128 (78.8) |
| Use of Mail Order | |||
| Yes | 80,053 (18.0) | 119,055 (23.3) | 7,183 (15.3) |
| No | 363,778 (82.0) | 391,339 (76.7) | 39,908 (84.7) |
| Use of 90-day Prescription Fills | |||
| Yes | 395,778 (89.2) | 457,031 (89.5) | 39,069 (83.0) |
| No | 48,053 (10.8) | 53,363 (10.5) | 8,022 (17.0) |
| DCCI (mean, SD) | 1.9 (2.3) | 2.1 (2.5) | 3.4 (2.6) |
| Medication Burden (mean, SD) | 6.2 (4.0) | 6.5 (4.3) | 7.4 (4.5) |
| PDC in Assess. Period (mean, SD) | 89% (16%) | 89% (16%) | 84% (20%) |
| Medical cost in the follow up period [Median (Interquartile range)] | $5,595.50 ($1,811.60-$17,604.00) | $6,106.20 ($2,148.90-$18,283.00) | $6,397.50 ($2,011.90-$20,216.00) |
DCCI Deyo-Charlson Comorbidity Index, LIS/DE Low-Income Subsidy/Dual-Eligibility
Results of the multivariable model for the single-measure cohort are provided in Table 2; for the three-measure cohort, the results are presented in Table 3. In general, there is a negative relationship between PDC in the assessment period and payer medical costs in the follow-up period regardless of medication class or cohort (i.e., higher PDC levels are associated with lower payer medical costs). For example, a 5% increase in PDC for diabetes medications in year 1 is associated with about a 2% reduction (p < 0.0001) in mean payer medical costs in year 2 [(0.9808–1)*100 ≅ −2%]. Similarly, a 5% increase in PDC resulted in a 1.3%, and 0.8% reduction in payer medical costs for RASA and statin medications, respectively. The associations tend to be stronger for diabetes medications, followed by RASA medications, and then statin medications (all p < 0.0001; see Table 2). Results were consistent in the three-measure cohort, where reduction in payer medical costs were observed to be 1.8%, 0.8%, and 0.2% for each 5% increase in PDC for diabetes medications, RASA medications, and statin medications, respectively. Associations were once again stronger for diabetes medications, followed by RASA, and statin medications (all p < 0.0001; Table 3).
Table 2.
Relationship between medication adherence and payer medical costs across three adherence measures: Single-Measure cohort
| Relationship | RASA Measure Only Estimate (95% CI) P value |
Statin Mesaure Only Estimate (95% CI) P value |
Diabetes Measure Only Estimate (95% CI) P value |
Omnibus P value for differences in the relationship among the groups | P values for pairwise differences in the relationship between the groups |
|---|---|---|---|---|---|
| Year 1 PDC ➔ Year 2 Payer Medical Costs |
0.9874 (0.9864, 0.9884) < 0.0001 |
0.9919 (0.9910, 0.9929) < 0.0001 |
0.9808 (0.9784, 0.9832) < 0.0001 |
< 0.0001 |
RASA only v. Statin only: <0.0001 RASA only v. Diabetes only: <0.0001 Statin only v. Diabetes only: <0.0001 |
For this analysis, patients were in only one of the 3 groups. PDC = Proportion of Days Covered. Covariates in all models: age, sex, race, geographic region of country, LIS/DE (low-income subsidy/dual eligible) status, insurance plan type, Deyo-Charlson Comorbidity Index (DCCI), medication burden, as well as group-specific treatment naïve status, mail-order pharmacy use, and 90-day prescription fill use. Models were estimated with PROC GENMOD in SAS using the log link and Gamma distribution. To aid in interpretation, the independent variable was scaled to represent a 5% change in PDC rather than a unit change (i.e., 1%) in the value for all models. Thus, the effect is expressed as a mean ratio of payer medical costs when comparing two different values of PDC 5% points apart, adjusted for covariates (i.e., how much does the adjusted mean payer medical costs in the follow-up period multiply by for each 5%-point increase in PDC in the assessment period?)
Table 3.
Relationship between medication adherence and payer medical costs across three adherence measures: Three-Measure cohort
| Relationship | RASA Estimate (95% CI) P value |
Statin Estimate (95% CI) P value |
Diabetes Estimate (95% CI) P value |
P values for pairwise differences in the relationship between medication classes |
|---|---|---|---|---|
| Year 1 PDC ➜ Year 2 Payer Medical Costs |
0.9922 (0.9905, 0.9940) < 0.0001 |
0.9981 (0.9965, 0.9997) 0.0184 |
0.9823 (0.9808, 0.9839) < 0.0001 |
RASA v. Statin: <0.0001 RASA v. Diabetes: <0.0001 Statin v. Diabetes: <0.0001 |
For this analysis, patients were taking medications from all three medication classes. PDC = Proportion of Days Covered. Covariates in all models: age, sex, race, geographic region of country, LIS/DE (low-income subsidy/dual eligible) status, insurance plan type, Deyo-Charlson Comorbidity Index (DCCI), medication burden, as well as medication-specific treatment naïve status, mail-order pharmacy use, and 90-day prescription fill use. Models were estimated with PROC GENMOD in SAS using the log link and Gamma distribution. To aid in interpretation, the independent variable was scaled to represent a 5% change in PDC rather than a unit change (i.e., 1%) in the value for all models. Thus, the effect is expressed as a mean ratio of payer medical costs when comparing two different values of PDC 5% points apart, adjusted for covariates (i.e., how much does the adjusted mean payer medical costs in the follow-up period multiply by for each 5%-point increase in PDC in the assessment period?)
Discussion
This study conducted a systematic investigation of Medicare Advantage beneficiaries who were eligible for one (or more) of three key medication use quality measures [27]. Given the characteristics of the source data and the size of the study cohorts, these findings include a large portion of the Medicare Advantage population. As expected, the study population was older and had significant medication burden. Their performance on the three adherence quality measures examined in this study was fairly high, with the average PDC for all three medication classes being near or above 85%. These high rates of adherence may be reflective of a combination of the unique nature of the Medicare Advantage population and the sustained efforts by health plans to improve quality measure performance.
Results of this study affirm prior research showing that increases in medication adherence can help reduce payer medical costs [9–15]. Evidence of this relationship in a variety of therapeutic areas and in this specific population lends support to the continued use of medication adherence-based quality measures within value-based reimbursement programs. Most importantly, results of this study extend prior research by comparing the strength of this relationship across different therapeutic areas. To our knowledge, formal tests of differences in the adherence-payer medical costs relationship across medication classes have not been presented in the literature. The relationship between medication adherence and future medical costs was strongest for diabetes medications compared to RASA or statin medications. Interestingly, this finding is also consistent among individuals using medication therapy from all three classes. This finding is likely a function of the nature of the condition itself, as inadequate diabetes management is associated with a variety of short and long-term adverse consequences with significant financial impact. While such a relationship is also true for hypertension or hypercholesterolemia, these findings show that the penalties for inadequate adherence management maybe greater in the case of diabetes. Given the growing prevalence and burden of diabetes in the United States, these results underscore the importance of the need for a coordinated, strategic, and interdisciplinary response for the management of diabetes. Within the context of a limited resource environment, these findings can help guide priorities for interventions for payers that are looking to maximize the impact of patient management services.
Another consideration is that the diabetes measure is considered an “all-class measure,” consisting of a number of different classes of medications indicated for the treatment of diabetes, with the exception of insulin. The RASA measure includes several similar drug classes, but certainly not all classes of medications indicated for hypertension (or other diseases for which RASAs may be used). The statin measure is for a specific medication class and does not include all medication classes indicated for hyperlipidemia. It is possible that multi-drug class adherence measures for hypertension or hyperlipidemia may have different results than those observed in the current analysis. Future research might explore such other multi-drug class adherence measures and their relationship with subsequent medical costs. It is important to note that all three conditions have guidelines that recommend add-on therapies if patients are not at the target health goals. This suggests a need for comprehensive disease state management rather than focusing solely on specific medications included in the measures as a pathway to ensure patients achieve optimal outcomes.
There are several limitations associated with the current study. First, as with any administrative claims-based analysis of medication adherence, prescription claims data are an indirect measure of medication-taking behavior, and the presence of a prescription claim does not indicate the medication was actually taken. Second, since CDM data are administrative health claims data collected primarily for administrative purposes, billing and coding errors and omissions cannot be ruled out. In addition, medical services not processed through insurance would not be captured in this analysis. It is important to mention that these findings do not necessarily extend to fee-for-service Medicare beneficiaries, who often have lower levels of medication adherence than Medicare Advantage beneficiaries. Third, this analysis chose to annualize healthcare costs in order to account for individuals who were followed for less than 12 months during the follow up. While this approach is fairly common, it does not account for the possibility that healthcare spending is often concentrated in specific periods due to acute events. Fourth, the analysis assumed a log-linear functional form in the relationship between cost and medication adherence. It is possible that other functional forms better capture this relationship and that the functional form may differ across these three quality measures. Future might explore these questions. Finally, although this study uses subsequent year payer medical costs as the outcome in an effort to reduce the impact associated with reverse causality between adherence and costs and also controls for a number of potential confounders, it is possible that important variables have been omitted from the regression models (i.e., residual confounding, for example, healthy adherer bias).
Conclusions
This study provides evidence that adherence to medications for chronic disease management continues to provide a significant and meaningful benefit to the payer. Targeting interventions to improve medication adherence can improve patient outcomes and reduce healthcare expenditures. There are differences in the relationship between adherence and payer medical costs across the three medications classes examined, with larger effects being observed for diabetes medications, followed by RASA medications, and then statin medications.
Supplementary Information
Additional file 1: Table 1. Patient Attrition Table. The additional table shows the number of individuals excluded from the analysis for each step of inclusion/exclusion requirements.
Additional file 2: Table 2. Descriptive Statistics for Beneficiaries Qualifying for All Three Adherence Measures.The additional table shows descriptive statistics for the beneficiaries that were included the three-measure cohort.
Additional file 3. Statistical models used in the analysis. The additional file shows the statistical models used to answer the primary research question.
Acknowledgements
Not applicable.
Abbreviations
- MA
Medicare Advantage
- RASA
Renin-angiotensin-system antagonists
- US
United States
- PQA
Pharmacy Quality Alliance
- CMS
Centers for Medicare & Medicaid Services
- CDM
Clinformatics® Data Mart Database
- ESRD
End-stage renal disease
- PDC
proportion of days covered
- OOP
out-of-pocket
Authors’ contributions
PC, TK, HB, JPB, and MP were involved in study design and protocol. PC, HB, and TK were responsible for data acquisition and access. MP and PC were responsible for acquisition of funding. SR, IN, MP were responsible for study programming and data analysis. JPB developed the statistical analysis protocol and was responsible for oversight of the analysis. SR and JPB wrote the first draft of the manuscript and all authors provided substantive edits and approved the final manuscript.
Funding
This study was funded by Merck Sharp & Dohme, LLC, a subsidiary of Merck & Co., Inc., Rahway, NJ, USA. The study was designed, analyzed, and the preparation of the manuscript was completed in collaboration with the funder.
Data availability
The data that support the findings of this study are available from Optum but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available upon reasonable request and with permission of Optum.
Declarations
Ethics approval and consent to participate
The conduct of the research was performed in compliance with the Declaration of Helsinki.
Consent for publication
Not applicable.
Competing interests
SR and JB report consulting fees from Pharmacy Quality Alliance, Inc. PC, HB, and TK are full-time employees of Merck & Co., Inc., Rahway, NJ, USA. HB owns stock in Merck & Co., Inc., Rahway, NJ, USA.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Cramer JA, Rosenheck R. Compliance with medication regimens for mental and physical disorders. Psychiatr Serv. 1998;49(2):196–201. 10.1176/ps.49.2.196. [DOI] [PubMed] [Google Scholar]
- 2.Sabaté E, editor. Adherence to Long-term Therapies: Evidence for Action. Geneva, Switzerland: World Health Organization, 2003. Available: https://apps.who.int/iris/handle/10665/42682. Accessed 5 June 2023.
- 3.Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005;353(5):487–97. 10.1056/NEJMra050100. [DOI] [PubMed] [Google Scholar]
- 4.New England Healthcare Institute (NEHI). Thinking Outside the Pillbox: A System-Wide Approach to Improving Patient Medication Adherence for Chronic Disease. Cambridge, MA, New England Healthcare Institute.: 2009. Available: http://psnet.ahrq.gov/issue/thinking-outside-pillbox-system-wide-approach-improving-patient-medication-adherence-chronic. Accessed 3 June 2023.
- 5.Naderi SH, Bestwick JP, Wald DS. Adherence to drugs that prevent cardiovascular disease: meta-analysis on 376,162 patients. Am J Med. 2012;125(9):882. 10.1016/j.amjmed.2011.12.013. [DOI] [PubMed] [Google Scholar]
- 6.Viswanathan M, Golin CE, Jones CD, et al. Interventions to improve adherence to self-administered medications for chronic diseases in the United States: a systematic review. Ann Intern Med. 2012;157(11):785–95. 10.7326/0003-4819-157-11-201212040-00538. [DOI] [PubMed] [Google Scholar]
- 7.Watanabe JH, McInnis T, Hirsch JD. Cost of prescription drug-related morbidity and mortality. Ann Pharmacother. 2018;52(9):829–37. 10.1177/1060028018765159. [DOI] [PubMed] [Google Scholar]
- 8.Lloyd JT, Maresh S, Powers CA, Shrank WH, Alley DE. How much does medication nonadherence cost the Medicare fee-for-service program? Med Care. 2019;57(3):218–24. 10.1097/MLR.0000000000001067. [DOI] [PubMed] [Google Scholar]
- 9.Aubert RE, Yao J, Xia F, Garavaglia SB. Is there a relationship between early statin compliance and a reduction in healthcare utilization? Am J Manag Care. 2010;16(6):459–66. [PubMed] [Google Scholar]
- 10.Roebuck MC, Liberman JN, Gemmill-Toyama M, Brennan TA. Medication adherence leads to lower health care use and costs despite increased drug spending. Health Aff. 2011;30(1):91–9. 10.1377/hlthaff.2009.1087. [DOI] [PubMed] [Google Scholar]
- 11.Hansen RA, Maciejewski M, Yu-Isenberg K, Farley JF. Adherence to antipsychotics and cardiometabolic medication: association with health care utilization and costs. Psychiatr Serv. 2012;63(9):920–8. 10.1176/appi.ps.201100328. [DOI] [PubMed] [Google Scholar]
- 12.Roebuck MC. Medical cost offsets from prescription drug utilization among medicare beneficiaries. J Manag Care Pharm. 2014;20(10):994–5. 10.18553/jmcp.2014.20.10.994. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Stuart BC, Dai M, Xu J, Loh F-HE, Dougherty S. Does good medication adherence really save payers money? Med Care. 2015;53(6):517–23. 10.1097/MLR.0000000000000360. [DOI] [PubMed] [Google Scholar]
- 14.Boye KS, Curtis SE, Lage MJ, Garcia-Perez L-E. Associations between adherence and outcomes among older, type 2 diabetes patients: evidence from a medicare supplemental database. Patient Prefer Adherence. 2016;10:1573–81. 10.2147/PPA.S107543. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Roebuck MC, Kaestner RJ, Dougherty JS. Impact of medication adherence on health services utilization in medicaid. Med Care. 2018;56(3):266–73. 10.1097/MLR.0000000000000870. [DOI] [PubMed] [Google Scholar]
- 16.Burwell SM. Setting value-based payment goals–HHS efforts to improve U.S. health care. N Engl J Med. 2015;372(10):897–9. 10.1056/NEJMp1500445. [DOI] [PubMed] [Google Scholar]
- 17.Centers for Medicare & Medicaid Services. Medicare 2022 Part C & D Star Ratings Technical Notes. Available: https://www.cms.gov/files/document/2022-star-ratings-technical-notes-oct-4-2022.pdf. Accessed 6 June 2023.
- 18.Centers for Medicare & Medicaid Services, U.S. Department of Health and Human Services. 2021 National Impact Assessment of the Centers for Medicare & Medicaid Services (CMS) Quality Measures Report. Baltimore, MD:. 2021. Available: https://www.cms.gov/files/document/2021-national-impact-assessment-report.pdf. Accessed 6 June 2023.
- 19.Pharmacy Quality A. 2022 PQA Measure Manual: Proportion of Days Covered: Renin Angiotensin System Antagonists (PDC-RASA). 2022. https://www Accessed 8 Apr 2024.
- 20.Pharmacy Quality Alliance. 2022 PQA Measure Manual: Proportion of Days Covered: Statins (PDC-STA). 2022. https://www.pqaalliance.org/measures-overview#pdc-sta. Accessed 8 Apr 2024.
- 21.Pharmacy Quality A. 2022 PQA Measure Manual: Proportion of Days Covered: Diabetes All Class (PDC-DR). 2022. https://www.pqaalliance.org/measures-overview#pdc-dr. Accessed 8 Apr 2024.
- 22.Lam WY, Fresco P. Medication adherence measures: an overview. Biomed Res Int. 2015;2015:217047. 10.1155/2015/217047. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Peterson AM, Nau DP, Cramer JA, Benner J, Gwadry-Sridhar F, Nichol M. A checklist for medication compliance and persistence studies using retrospective databases. Value Health. 2007;10(1):3–12. 10.1111/j.1524-4733.2006.00139.x. [DOI] [PubMed] [Google Scholar]
- 24.Nau DP. Proportion of days covered (PDC) as a preferred method of measuring medication adherence. Alexandria, VA: Pharmacy Quality Alliance; 2012. [Google Scholar]
- 25.Korhonen MJ, Ruokoniemi P, Ilomäki J, Meretoja A, Helin-Salmivaara A, Huupponen R. Adherence to Statin therapy and the incidence of ischemic stroke in patients with diabetes. Pharmacoepidemiol Drug Saf. 2016;25(2):161–9. 10.1002/pds.3936. [DOI] [PubMed] [Google Scholar]
- 26.Babitsch B, Gohl D, Von Lengerke T. Re-revisiting andersen’s behavioral model of health services use: a systematic review of studies from 1998–2011. GMS Psycho-Social-Medicine. 2012;9(Doc11). 10.3205/psm000089. [DOI] [PMC free article] [PubMed]
- 27.Parikh M, Nsiah I, Campbell P, Ramachandran S, Karmarkar T, Castora-Brinkley M, Black H, Bentley JP. Medication Adherence and Subsequent Year Payer Medical Costs Among Medicare Advantage Beneficiaries: Differences across Quality Measures. ISPOR Annual Meeting. 2023. Available: https://www.ispor.org/conferences-education/conferences/past-conferences/ispor-2023/program/program/paper/intl2023-3682/16030/medication-adherence-and-subsequent-year-payer-medical-costs-among-medicare-advantage-beneficiaries-differences-across-quality-measures. Accessed 29 July 2025.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Additional file 1: Table 1. Patient Attrition Table. The additional table shows the number of individuals excluded from the analysis for each step of inclusion/exclusion requirements.
Additional file 2: Table 2. Descriptive Statistics for Beneficiaries Qualifying for All Three Adherence Measures.The additional table shows descriptive statistics for the beneficiaries that were included the three-measure cohort.
Additional file 3. Statistical models used in the analysis. The additional file shows the statistical models used to answer the primary research question.
Data Availability Statement
The data that support the findings of this study are available from Optum but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available upon reasonable request and with permission of Optum.