Reductions in Mortality Among Medicare Beneficiaries Following the Implementation of Medicare Part D

Published Online: July 16, 2015
April P. Semilla, MS; Fang Chen, PhD; Timothy M. Dall, MS
Medicare Part D is a prescription drug program that provides seniors and disabled individuals enrolled in Medicare with outpatient drug coverage benefits. Part D has been shown to increase access to medicines and improve medication adherence; however, the effect of Part D on health outcomes has not yet been extensively studied. In this study, we used a published and validated Markov-based microsimulation model to quantify the relationships among medication use, disease incidence and severity, and mortality. Based on the simulation results, we estimate that since the implementation of Part D in 2006, nearly 200,000 Medicare beneficiaries have lived at least 1 year longer. Reductions in mortality have occurred because of fewer deaths associated with medication-sensitive conditions such as diabetes, congestive heart failure, stroke, and myocardial infarction. Improved access to medication through Medicare Part D helps patients improve blood pressure, cholesterol, and blood glucose levels, which in turn can prevent or delay the onset of disease and the incidence of adverse health events, thus reducing mortality.

Am J Manag Care. 2015;21:S165-S172

In 2003, Medicare Part D, a prescription drug program providing outpatient drug coverage benefits to seniors and disabled individuals enrolled in Medicare, was signed into law. A large body of literature documents the positive impact that Part D has had on increasing access to medicines and improving medication adherence; however, the effect of Part D on health outcomes has not yet been extensively studied.1-9

While life expectancy among the population 65 years and older has increased since 2006,10 4 studies that have examined Part D’s impact on mortality report mixed findings.5,11-13 Kaestner et al and Briesacher et al found no statistically significant association between Part D and mortality.5,11 Huh and Reif estimated that Part D reduced overall mortality among beneficiaries by 2.2% annually, with the effect driven primarily by a decrease in cardiovascular-related deaths.12 Similarly, Dunn and Shapiro found that Part D reduced cardiovascular-related mortality, but that mortality associated with noncardiovascular disease remained unchanged.13 Dunn and Shapiro estimate that between 21,800 and 25,500 individuals were still alive in mid-2007 because of Part D’s implementation in 2006. These 4 studies were limited to a narrow time frame of analysis or made strong assumptions regarding the continuation of historical trends pre- and post Part D. The goal of this study is to further explore Part D’s effect on mortality from 2006 to the present.


This study used a published and validated Markov-based microsimulation model to quantify the relationships among medication use, disease incidence and severity, and mortality. Information on the model design, data, assumptions, methods, and validation results have been published elsewhere.14,15 A microsimulation approach was chosen based on its ability to model changes in health and disease states longitudinally based on individual demographics (age, sex, race, ethnicity), biometrics (body mass index [BMI], systolic blood pressure [SBP], diastolic blood pressure [DBP], total cholesterol, high-density lipoprotein cholesterol [HDL-C], and glycated hemoglobin [A1C]), smoking status, and presence of approximately 30 diseases to predict mortality as an individual ages. The model incorporates the expected impact of medication use on changes in blood pressure, cholesterol, and A1C levels, and simulates the resulting implications on disease incidence and severity that in turn impact the cause and timing of beneficiary mortality.


Population We created a representative sample of the Medicare population using data from the 2005 to 2012 National Health and Nutrition Examination Survey (NHANES) files.16 The combined files contain data for 3621 Medicare beneficiaries 65 years and older, including demographic and biometric information required for use in the microsimulation model. Using NHANES sample weights to dictate the probability of selection, we used Monte Carlo simulation, which involves repeated random sampling of the population, to produce a nationally representative simulation sample of 100,000 observations representative of the estimated 22.5 million beneficiaries enrolled in Medicare Part D in 2006.17 We added 20,000 additional observations in each year of the simulation to represent new enrollees in Part D. This annual number of observations added to the simulation population reflected that net enrollment in Part D grew by approximately 6.5% annually between 2006 and 2014 after factoring in mortality, and the assumption that once enrolled in Part D beneficiaries did not drop Part D coverage unless death occurred.18

Using this model, we simulated outcomes for this population under 2 scenarios over the 2006 to 2014 period:

  1. Medicare Part D scenario: An estimated 25% of Part D enrollees would not have had comprehensive drug coverage in the absence of Part D.19,20 Beneficiaries who gained comprehensive coverage under Part D increased their use of medication.3,4,6,21,22 For modeling purposes, we used findings from Zhang et al, which showed that drug possession rates increased by 29%, 31%, and 22%, respectively, for people diagnosed with hyperlipidemia, diabetes, and hypertension.23
  2. Status Quo scenario: In this scenario, there were no changes in drug coverage among the Medicare population. As a result, there were no changes in medication use, and natural progression of disease and mortality was maintained.

Results for both scenarios were then scaled to simulate total Part D enrollment.

Model Parameters and Disease States

For each beneficiary, the previous year’s health status was used to predict the current year’s outcomes, with this process repeated through 2014 or until death. Prediction equations dictating the transition probabilities for various cardiovascular conditions (congestive heart failure [CHF], ischemic heart disease, stroke, myocardial infarction [MI]), diabetes, chronic kidney disease, renal failure, and other diseases modeled were obtained from published studies including the Framingham Heart Study, the United Kingdom Prospective Diabetes Study, and various clinical trials and observational studies.24-30 Annual changes in body weight were estimated based on national trends by age, sex, and current weight.14,15 Annual changes in DBP, SBP, HDL-C and total cholesterol were determined based on each beneficiary’s age, sex, and change in BMI,14,15,24,31,32 while changes in A1C were predicted based on BMI, age, and total cholesterol (if beneficiary did not have diabetes) and years since diagnosis and previous year’s A1C (if diabetic).14,15,28

Annual mortality rates associated with the chronic diseases modeled, acute events such as MI and stroke, and “all other causes” were estimated from published sources and based on demographics, biometrics, smoking status, and presence of adverse medical conditions.14,15,28,33,34

Coverage and Drug Effect

To simulate the impact of gaining comprehensive drug coverage under the Part D scenario, enrollees with high cholesterol (total cholesterol >240 mg/dL), high blood pressure (SBP >140 mm Hg or DBP ≥90 mm Hg), and high A1C levels (≥6.5%) were assumed to increase their medication use, resulting in improvements on these measurements. We performed sensitivity analyses using published 95% CIs for key parameters used in the model. We modeled that statins would reduce total blood cholesterol by 34.42 mg/dL (CI, 22.04-46.40), and anti-hypertensives would reduce SBP by 14.5 mm Hg (CI, 14.2-14.8) and DBP by 10.7 mm Hg (CI, 10.5-10.8).35,36 These reductions were assumed to occur entirely within the first year of therapy. Oral diabetes medications were assumed to reduce A1C by 1 percentage point (CI, 0.5-1.25) annually across every year of the analysis until diabetes control was reached at A1C of 7.5%.37 (Most patients with high A1C levels see their A1C levels fall to 7.5% or lower within 1 year.) Medication use and drug effect were assumed to remain unchanged in the status quo scenario.


A nationally representative sample of Medicare beneficiaries 65 years and older served as the baseline simulation population in this study. The initial (2006) sample was 55.5% female, with an average BMI of 28.4, and an average age of 73.6 years (Table 1). Controlled or uncontrolled hypertension (74.4%) and hypercholesterolemia (58.6%), prediabetes (60.0%), and diabetes (25.1%) were highly prevalent among this population. The CDC reports similar characteristics for Medicare beneficiaries in 2006, but reports a greater proportion of individuals 85 years and older (14.0% vs 2.2% in our sample) and a lower prevalence of hypertension (62.8%).38

In the status quo scenario, the prevalence of uncontrolled hypertension remained fairly constant at about 41% from 2006 to 2014. We observed a lower prevalence under the Part D scenario (ranging from 38.8% to 39.6%), which translates into an estimated 550,000 fewer beneficiaries with uncontrolled hypertension between 2006 and 2014 (Table 2). Cumulative over this period, there were 180,000 fewer new cases of CHF, 210,000 fewer MIs, and 830,000 more people with tightly controlled diabetes (defined by A1C level <7%).

The results of the simulation suggest that about 198,800 Part D beneficiaries lived at least 1 year longer following the implementation of the program. The average increase in longevity was 3.3 years. On average, approximately 22,100 lives were saved annually between 2006 and 2014. The number of lives saved each year increased from 2006 to 2009, but declined in the later years as deaths that might have occurred sooner eventually occurred in later years (Figure).

Mortality reductions were primarily attributable to fewer deaths from medication-sensitive conditions, particularly diabetes and cardiovascular diseases (Table 3). We estimate that between 2006 and 2014, total deaths associated with diabetes declined by 100,400, deaths associated with CHF declined by 53,100, deaths from stroke declined by 40,400, and deaths from MI declined by 25,100.

PDF is available on the last page.
Clinical Pathways Compendium
COPD Compendium
Diabetes Compendium
GI Compendium
Lipids Compendium
MACRA Compendium
Oncology Compendium
Rare Disease Compendium
Reimbursement Compendium
Know Your News
HF Compendium