APPENDIX A.Assigning Statin Use in the Future Elderly Model

APPENDIX A.Assigning Statin Use in the Future Elderly Model

A.1Data

To estimate the probability that individuals are statin users within FEM simulations, we utilize estimates based on Medical Expenditure Panel Survey (MEPS) data. The MEPS is a set of large-scale surveys of families and individuals, their medical providers and employers across the United States. The MEPS collects data on the specific health services that Americans use, how frequently they use them, the cost of these services, and how they are paid for, as well as data on the cost, scope, and breadth of health insurance held by and available to U.S. workers.

We use two components of the MEPS. The first is the Household Component (HC), which collects data from a sample of families and individuals in selected communities across the United States, drawn from a nationally representative sample of the civilian non-institutionalized population of the United States. These constitute a subsample of households that participated in the prior year's National Health Interview Survey (conducted by the National Center for Health Statistics). During the household interviews, MEPS collects detailed information for each person in the household, including demographic characteristics and health conditions. The second is the Prescribed Medicines files, which provide detailed information on household-reported prescribed medicines, which can be used to make estimates of prescribed medicine utilization and expenditures. Each record on this event file represents a unique prescribed medicine event; that is, a prescribed medicine reported as being purchased by the household respondent.

By merging the two components for years 2009-2011, we obtained an individual-level database including individual characteristics and statin use. A MEPS respondent is defined as a statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase inhibitor” therapeutic subclassduring a given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which combine statins with other active ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.The years 2009-2011 are chosen to surround 2010, the first year of our simulations. We also used MEPS 2008 data to find the lag of statin use in 2009.

A.2Estimation

Since we define statin use as a dichotomic variable, we probit regressions on our MEPS dataset to find the probability that individuals are using statins at each period in the FEM simulations. We estimate these probabilities in two steps. The first regression finds the factors influencing the probability that 50 to 53 year-old individuals use statins, based on information common to the MEPS and the FEM: demographics, education, health conditions and BMI. The regression estimates are used within FEM to assign the probability of statin use at ages 51-52, in the first period of the simulations. The second regression uses the MEPS 2-year overlapping panel design to find the probability of using statins for the population aged 53 and over. This regression is limited to individuals in their second interview year, and adds age splines and the use of statins in the previous year (by far the largest predictor of statin use) to the list of regressors. After several specification checks, we chose nodes at age 70 and BMI 30 for the age and log of BMI splines in order to maximize the fit. The estimates of this regression are used to assign the transitions in and out of using statins in each period of the simulations. The results of both regressions are presented in Table A2.

A.3Statin Assignment

The use of these estimates in the assignment of statin use follows the same methodology as the other non-absorbing binary outcomes in the FEM, which are detailed in the complete technical appendix provided in the Electronic Supplementary Material. Since the FEM periods are two-year long and the MEPS only follows respondents longitudinally for a second year, we first assign statin use during the next year, which corresponds to the FEM’s mid-period. This mid-period statin use then becomes the lag of statin use, based on which we assign statin use for the next FEM period.

Thus, MEPS data is used to assign statin use to our FEM cohort, which is based on HRS data. While the two surveys are representative of the American population they study, it is initially unclear whether our assignment of statin use will be consistent with observed data. To verify this, we use our methodology to assign a probability of statin use to each individual of the FEM cohort of age 51-52 in 2010.[1] We then compare the average probability of using statins in this cohort to observed statin use in the MEPS for the same ages in years 2009 to 2011. We focus on statin use by BMI and chronic condition diagnostic (Figure A3).

We find that our assigned probability of using statins mirrors quite well the observed prevalence of statin use in the MEPS. We observe some divergence for individuals with very high BMIs, for which statin use probability in FEM is about 10% lower than prevalence in the MEPS. The MEPS however counted less than 35 respondents for these BMI values. It is of note that we cannot compare statin use by age between the FEM simulations and the MEPS data because of cohort effects, since we simulate the health of the cohort aged 51-52 in 2010. For instance, our FEM cohort at age 71-72 (in 2030) is quite different from respondents of the same age in the available MEPS data.

APPENDIX B.Effects of Statin Use and Obesity on Health Dynamics and Mortality in the Future Elderly Model

To clarify the mechanisms at play in the simulations, we refer to Figure 1, which synthesizes FEM transitions into illness and death. This figure simultaneously applies to all chronic conditions in the FEM. An individual has probability p of already suffering from a chronic condition at a given age, and probability 1-p of being free of that condition. Conditional on him being healthy, he has a probability q1 of contracting the disease, a probability q2of dying and a probability 1-q1-q2 of staying healthy in the next period. If the individual already has the disease, he has a probability u of dying and a probability 1-u of surviving with the illness.

Obesity has two effects on these probabilities. As summarized in Figure A4, obesity has an intricate relationship with the health conditions in the FEM. BMI is a significant predictor of key health conditions including diabetes, stroke, heart diseases, and hypertension, which have important dynamic consequences for the life trajectories of individuals. As shown in the figure, having one or several chronic conditions increases the probability of contracting additional ones, which contribute to the probability of mortality. Thus, obesity increases the probability that an individual is unhealthy in a given year (p) and the probability that a healthy individual contracts a new condition (q1). The widespread use of statins has four impacts: it increases the probability that an individual is healthy in a given year (1-p), decreases the probability of contracting heart diseases and stroke (q1), and lowers the probability of dying if healthy (q2) or afflicted with heart disease (u).

The net effect of the interaction of obesity and statins is complex. First, for otherwise identical individuals free of heart disease and stroke, the reduction of q1 due to statin use has more impact on the obese, since their base probability of contracting the conditions is higher.

Second, since obesity does not directly enter the FEM’s mortality prediction and only affects mortality through the onset of chronic conditions, the effect of statins on mortality (q2 and u) is identical for individuals differentiated only by their BMI.

Third, since obese people are more likely to be chronically ill in a given year (higher p), obese statin users benefit relatively more from the secondary prevention of mortality (reduction of u) than healthy-weight statin users. For the same reason, obese people gain relatively less from the primary prevention effects of statins (q1 and q2 reductions). Thus, the worse health of obese people as we simulate their health outcomes interacts in an a priori ambiguous manner with the health effects of statins.

Finally, a factor that is absent from Figure 1 is the probability of using statins. As can be seen in Table A2, statin use increases significantly with BMI until it reaches 30. This effect unequivocally increases the health impact of statins for obese individuals.

APPENDIX C.Sensitivity analyses

We consider two sensitivity analyses for the effect of statins on the costs of obesity: 1) without reassigning the BMI of the cohort at age 51, and 2) simulating only the secondary prevention effect of statins. The effects of these alternative approaches on life expectancy and quality-of-life are detailed for all BMIs in Table A4. The baseline difference-in-difference results of Table 3 are compared with these approaches in Table A5. For simplicity, we present only results for the obesity class 1 category, which accounts for over 50% of the US obese population, in Table A5. The difference-in-difference in the total cost of obesity is presented graphically for all obesity classes in Figure A5.

Appendix A

Appendix B

Appendix C

C.1Running simulations without reassigning BMI

Throughout this article, we consider the cost of being obese as the consequences of an average person of age 51 going from a healthy BMI to an obese BMI. Thus, we reassigned the BMI of each individual in our cohort to specific BMIs, while keeping their other states (comorbidities) unchanged (see Section 2.2.2). The difference in expected health outcomes and medical spending between these synthetic cohorts revealed the cost of obesity. By construction, this strategy removes the difference in other health variables linked to obesity. In other words, our strategy considers that a person of average health faces the choice between obese and non-obese BMIs, rather than a choice between the average health of obese andnon-obese people.

To test whether statins have a larger effect when taking into account the lower health status of obese people, we conduct an additional set of simulations. Instead of reassigning BMI values, we remove from the simulations all individuals outside of the BMI category (BMI under 24.9 for “Healthy” and between 30 and 34.9 for “Obese class 1”, etc.).

As could be expected, the differential health of obese individuals has serious consequences on longevity and quality of life. In Table A4, we find that the difference in life expectancy and expected QALYs by BMI is much greater without reassigning BMI (the second set of results in the table) than our baseline results (the first set of results). We also find that, under this definition, the additional gain of obese individuals – the difference-in-difference - is higher. According to our results, obese individuals can expect to gain 0.18-0.19 more years of life expectancy than healthy-weight individuals, about three times more than in our baseline simulations.

However, the costs of obesity are also much larger with this definition than they are under our baseline approach. In Table A5, our baseline difference-in-difference results for obese class 1 are shown in the first column,[2]and those obtained without reassigning BMI are presented in the second column. We note that the larger confidence intervals for all statistics in the table are caused by the smaller number of individuals going through the simulations for each BMI category. While we find a net reduction of $5,000 in the costs of obesity with this specification, it represents only 0.8% of the “Without Statins” costs of obesity and is not significantly different from zero, as was the case in our baseline simulation.

The Panel B of Figure A5 shows the effect of statins on the total costs of obesity for the three obese categories visually. When compared to Panel A (our baseline results), the cost of obesity are larger, while the differences between the “Without Statins” and the “With Statins” remain small – and statistically insignificant – for all obese classes.

C.2Simulating only the secondary prevention effect of statins

Our second sensitivity analysis considers only the secondary prevention effects of statins. The “Without Statins” scenario of this analysis removes only the secondary prevention effect of statins, and thus corresponds to a scenario in which statins are used, but do not reduce the risk of mortality after the onset of heart disease. This analysis serves two purposes. First, the small impact of statin therapy on the cost of obesity can plausibly be due to its important primary prevention effects, which may benefit healthy-weight individuals more than obese individuals. The primary prevention effects can thus be too effective for statins to lower the costs of obesity. Considering only the secondary prevention effects can thus better illustrate the potential of medical innovation to reduce the costs of obesity when its benefits are targeted towards the obese. Second, this exercise serves to decompose the overall effect of statins (the baseline results) into its primary and secondary prevention components.

As we saw in Table 2, when considering both primary and secondary preventions, individuals with class 1 obesity gain 6% more life expectancy than healthy-weight individuals (1.24 vs. 1.17 years). However, most of the difference-in-difference is associated with some level of disability, such that QALY gains are only 2% higher (0.90 vs 0.88 QALYs). When considering only the secondary prevention effect (third set of results of Table A4), we find that the gains of obese individuals expressed as a percent of the gains of healthy weight individual increase. For instance, life expectancy and QALY gains of obese class 1 individuals are 14% and 8% higher than healthy-weight individuals, respectively. Thus, this analysis confirms the notion that the secondary prevention effect of statins benefits obese individuals more, while the primary prevention effect benefits healthy-weight individuals more.

However, the difference-in-difference in QALYs is quite small (0.02-0.03 QALYs) relative to the costs of obesity, and is partially countered by higher medical costs. Thus, when compared to the baseline results, the net reduction in the costs of obesity due to the secondary prevention effect also does not significantly differ from zero (shown for class 1 obesity in Table A5). When presented graphically (Panel C of Figure A5), the costs of obesity appear virtually identical to our baseline results (Panel A).

Table A1 Health Impact of Statins

Source / Population / Effects of statin therapy compared to placebo
Cochrane Collaboration 2013[12] / Primary prevention of cardiovascular disease / Reduced all-cause mortality (OR 0.86, 95% CI 0.79-0.94)
Reduced fatal (RR 0.83, 95% CI 0.72-0.96) and non-fatal (RR 0.77, 95% CI 0.62-0.96) CVD
Reduced fatal (RR 0.82, 95% CI 0.7-0.96) and non-fatal (RR 0.67, 95% CI 0.59-0.76) CHD
Reduced non-fatal stroke (RR 0.69, 95% CI 0.58-0.83)
Chen 2012[7] / Primary prevention of cardiovascular disease among diabetics / Reduced incidence of MACCE (RR 0.79, 95% CI 0.66-0.95)
Reduced the risk of stroke (RR 0.71, 95% CI 0.54-0.94)
No significant effect on all-cause mortality (RR 0.79, 95% CI 0.58-1.08).
Cholesterol Treatment Trialists 2012[8] / Primary prevention of cardiovascular disease / Reduced risk of major vascular events (RR 0.79, 95% CI 0.77-0.81, per 1.0 mmol/L reduction in LDL)
Reduced risk of vascular mortality (RR 0.85, 95% CI 0.77-0.95, per 1.0 mmol/L reduction in LDL)
Reduced all-cause mortality (RR 0.91, 95% CI 0.85-0.97 per 1.0 mmol/L reduction in LDL).
de Vries 2012[9] / Primary prevention of cardiovascular disease among diabetics / Reduced major cardiovascular or cerebrovascular events (RR 0.75, 95% CI 0.67-0.85)
Reduced fatal/non-fatal stroke (RR 0.69, 95% CI 0.51-0.92)
Reduced fatal/non-fatal MI (RR 0.70, 95% CI 0.54-0.90)
No significant effect on all-cause mortality (RR 0.84, 95% CI 0.65-1.09)
Mills 2011[11] / Primary and secondary prevention of cardiovascular disease / Reduced all-cause mortality (RR 0.90, 95% CI 0.86-0.94)
Reduced CVD mortality (RR 0.80, 95% CI 0.74-0.87)
Reduced fatal (RR 0.82, 95% CI 0.75-0.91) and non-fatal (RR 0.74, 95% CI 0.67-0.81) MI
Reduced revascularization (RR 0.76, 95% CI 0.70-0.81)
Reduced fatal/non-fatal strokes (RR 0.86, 95% CI 0.78-0.95)
Gutierrez 2012[10] / Secondary prevention of cardiovascular disease / Reduced CVD events in women (RR 0.81, 95% CI 0.74-0.89) and men (RR 0.82, 95% CI 0.78-0.85)
Reduced all-cause mortality (RR 0.79, 95% CI 0.72-0.87) in men
Reduced stroke (RR 0.81, 95% CI 0.72-0.92) in men
No significant effect on all-cause mortality (RR 0.92, 95% CI 0.76-1.13) in women
No significant effect on stroke (RR 0.92, 95% CI 0.76-1.10) in women
Wei 2005[25] / Secondary prevention of cardiovascular disease / Significant reduction in all-cause mortality among overall community population (RR 0.69, 95% CI 0.59-0.80), women (RR 0.63, 95% CI 0.49-0.80), and age ≥ 65 (RR 0.72, 95% CI 0.61-0.84)
Significant reduction in fatal/non-fatal MI among overall community population (RR 0.82, 95% CI 0.71-0.95), women (RR 0.69, 95% CI 0.54-0.88), and age ≥ 65 (RR 0.84, 95% CI 0.71-0.99)
RR: Relative risk; OR: Odds ratio; CVD: Cardiovascular disease; CHD: Coronary heart disease; MI: Myocardial infarction; MACCE: Major cardiovascular and cerebrovascular events

Table A2 Factors Influencing the Probability of Purchasing Statins in MEPS, 2009 to 2011: Probit Estimates

1. Initialization Model / 2. Transition Model
Pop. aged 50 – 53 / Pop. aged 53+
Coefficient / Standard error / Marginal effect / Coefficient / Standard error / Marginal effect
Demographics and education
Black / -0.19 / ** / (0.08) / -0.039 / -0.08 / (0.06) / -0.028
Hispanic / -0.24 / *** / (0.09) / -0.049 / -0.10 / (0.07) / -0.033
Male / 0.08 / (0.10) / 0.018 / 0.03 / (0.06) / 0.010
Less than high school / 0.13 / * / (0.07) / 0.030 / -0.02 / (0.06) / -0.006
College education / 0.26 / *** / (0.09) / 0.058 / 0.08 / (0.07) / 0.027
Male and less than high school / -0.09 / (0.13) / -0.020 / -0.16 / (0.10) / -0.054
Male and college / 0.01 / (0.10) / 0.003 / -0.04 / (0.08) / -0.013
Male and black / -0.23 / ** / (0.12) / -0.045 / -0.06 / (0.10) / -0.021
Male and Hispanic / -0.02 / (0.12) / -0.005 / -0.02 / (0.10) / -0.007
Health Conditions
Cancer / 0.22 / ** / (0.11) / 0.055 / -0.07 / (0.06) / -0.024
Diabetes / 0.91 / *** / (0.06) / 0.272 / 0.27 / *** / (0.05) / 0.099
Heart diseases / 0.40 / *** / (0.06) / 0.103 / 0.21 / *** / (0.04) / 0.076
High blood pressure / 0.51 / *** / (0.05) / 0.117 / 0.18 / *** / (0.04) / 0.064
Lung diseases / 0.05 / (0.09) / 0.011 / 0.05 / (0.06) / 0.017
Stroke / 0.04 / (0.11) / 0.009 / 0.10 / * / (0.06) / 0.037
Weight (loglinear splines)
Log of BMI under 30 / 1.36 / *** / (0.22) / 0.297 / 0.63 / *** / (0.15) / 0.221
Log of BMI over 30 / -0.12 / (0.21) / -0.026 / -0.24 / (0.19) / -0.083
Age (linear splines)
Less than 70 years old / 0.02 / *** / (0.00) / 0.008
70 years and older / -0.01 / *** / (0.00) / -0.004
Statin use in the previous period / 2.67 / *** / (0.04) / 0.818
Constant / -6.00 / *** / (0.72) / -5.05 / *** / (0.55)
Observations / 5,234 / 11,458
Pseudo R-squared / 0.154 / 0.594

*** p<0.01, ** p<0.05, * p<0.1. BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the square of height, expressed in meters. A MEPS respondent is defined as a statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase inhibitor” therapeutic subclass during a given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which combine statins with other active ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.

Table A3 95% Confidence Interval of the Impact of Statins on Lifetime Medical Costs after Age 50, $2009 thousands

With Statins / Without Statins / Difference
Healthy weight (BMI < 25)
Total Medical Costs / [459.3 - 462.4] / [422.9 - 448.7] / [12.1 - 38.0]
Excluding Statins / [454.5 - 457.5] / [422.9 - 448.7] / [7.2 - 33.2]
Free of disability / [288.1 - 289.6] / [280.9 - 286.7] / [2.7 - 7.8]
Disabled / [96.2 - 97.4] / [88.2 - 95.3] / [1.6 - 8.7]
In a nursing home / [69.2 - 71.5] / [52.9 - 68.2] / [2.6 - 17]
Statins / [4.8 - 4.8] / [0 - 0] / [4.8 - 4.8]
Obese class 1 (BMI = 30)
Total Medical Costs / [476 - 479.3] / [435.3 - 464.2] / [13.3 - 41.9]
Excluding Statins / [470.5 - 473.7] / [435.3 - 464.2] / [7.7 - 36.4]
Free of disability / [283.5 - 285.0] / [276.2 - 282.0] / [2.6 - 8.1]
Disabled / [115.8 - 117.1] / [105.8 - 114.4] / [2.1 - 10.6]
In a nursing home / [70.2 - 72.3] / [52.7 - 69.2] / [2.5 - 18.2]
Statins / [5.6 - 5.6] / [0 - 0] / [5.6 - 5.6]
Obese class 2 (BMI = 35)
Total Medical Costs / [483.8 - 487.1] / [441.6 - 471.5] / [13.5 - 42.9]
Excluding Statins / [478.3 - 481.5] / [441.6 - 471.5] / [8 - 37.3]
Free of disability / [277.1 - 278.5] / [269.7 - 275.5] / [2.6 - 7.9]
Disabled / [130.5 - 131.9] / [119.3 - 128.8] / [2.1 - 11.7]
In a nursing home / [69.8 - 71.7] / [51.8 - 68.9] / [2.4 - 18.5]
Statins / [5.5 - 5.6] / [0 - 0] / [5.5 - 5.6]
Obese class 3 (BMI = 40)
Total Medical Costs / [489.0 - 492.1] / [445.8 - 476.9] / [13.2 - 44]
Excluding Statins / [483.5 - 486.7] / [445.8 - 476.9] / [7.8 - 38.6]
Free of disability / [267.2 - 268.7] / [260.0 - 265.6] / [2.5 - 7.7]
Disabled / [146.5 - 148.1] / [134.0 - 144.7] / [2.5 - 13.1]
In a nursing home / [68.7 - 70.8] / [51.0 - 68.1] / [2.1 - 18.8]
Statins / [5.4 - 5.4] / [0 - 0] / [5.4 - 5.4]

This table shows 95% confidence intervalswith regards to the uncertainty of the effectiveness of statins for the first three columns of Table 1. All amounts are in present value at age 51, computed with a 3% interest rate. BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the square of height, expressed in meters.The first two columns present lifetime medical costs at age 51 in the “With Statins” and “Without Statins” scenarios. The third column presents the difference between the scenarios, and corresponds to the additional medical costs due to the existence of statins. The last three columns decompose the additional costs due to statins by spending source. In the rows, we decompose medical costs by type (total medical costs excluding statins prescriptions and statin prescriptions) and by functional status.A “disabled” functional status refers to reporting at least one instrumental activity of daily living (IADL) or activity of daily living (ADL) limitation. The “free of disability” status refers to reporting .no IADL or ADL limitation and not living in a nursing home. “In a nursing home” indicates the most severe functional status impairment.