Cost-Effectiveness of Lipid-Lowering Treatments in Young Adults

Ciaran N. Kohli-Lynch, PHD; Brandon K. Bellows, PHARMD; Yiyi Zhang, PHD; Bonnie Spring, PHD; Dhruv S. Kazi, MD; Mark J. Pletcher, MD; Eric Vittinghoff, PHD; Norrina B. Allen, PHD; Andrew E. Moran, MD


J Am Coll Cardiol. 2021;78(20):1954-1964. 

In This Article


We obtained Columbia University Institutional Review Board approval to perform secondary analysis on individual participant data from National Institutes of Health observational cohort studies. All other data sources were publicly accessible or were published summary data.

Number of Treatment-eligible Young Adults

To project the number of U.S. young adults eligible for lipid-lowering therapies, we examined the age-stratified distribution of LDL-C (<100 mg/dL, 100–129 mg/dL, 130–159 mg/dL, 160–189 mg/dL, and ≥190 mg/dL) among ASCVD-free U.S. adults in the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2014, scaled to 2020 U.S. population estimates.[10,11] We also examined self-reported cholesterol screening rates to contextualize the awareness of raised LDL-C (Supplemental Appendix).

CVD Policy Model

A microsimulation version of the CVD Policy Model was developed in TreeAge Pro, version 2020 (TreeAge Software, Inc). The model estimated lifetime health-related quality of life and direct health care costs (2020 US$) for LDL-C–reduction strategies in a nationally representative cohort of ASCVD-free U.S. young adults. The CVD Policy Model uses time-varying ASCVD risk factor exposures to simulate an individual's lifetime risk of ASCVD. As individuals progress through the model, they accumulate health care costs and quality-adjusted life years (QALYs) (Supplemental Tables 1 to 5).

Individuals start the simulation without ASCVD and, each year, are at risk of coronary heart disease (CHD), stroke, combined CHD and stroke, and ASCVD or non-ASCVD death. Incident ASCVD and non-ASCVD mortality risks are estimated using competing risk survival models developed in the NHLBI-PCS (National Heart, Lung, and Blood Institute Pooled Cohorts Study) dataset (Supplemental Figure 1).[12–15]

The probabilities of the first ASCVD event and the probability of non-ASCVD mortality are operationalized in the model using logistic risk functions (Supplemental Table 1, equation 1). Predictors include race (African American or non–African American), current age, LDL-C, high-density lipoprotein cholesterol, systolic blood pressure (SBP), body mass index, smoking status (current, former, never), cigarettes per day, diabetes status, and estimated glomerular filtration rate. Both LDL-C and SBP are incorporated in the risk functions as time-weighted averages (TWAs), which are updated each year from age 18 years to present. Individuals with higher lifetime exposure to LDL-C are at greater risk of experiencing CHD and stroke (Central Illustration). Interaction effects with age reduce the effect of the TWA variables over time:

Central Illustration.

Conceptual Diagram, Cardiovascular Disease Policy Model, and Time-Weighted Average Risk Factors
(A) Potential LDL-C trajectories for a 20-year-old young adult with raised LDL-C (172 mg/dL) are shown: no treatment (blue solid line), later life statin treatment (red dotted line), and lifetime statin treatment (gray dashed line). (B) Results are shown for 100,000 simulations of the 3 LDL-C trajectories in A while holding other risk factors constant. The mean number of ASCVD events (left axis) was lower, and the mean ASCVD-free years (right axis) was greater with treatment initiation in young adulthood. ASCVD = atherosclerotic cardiovascular disease; LDL-C = low-density lipoprotein.

This equation shows the logistic risk function predicting event k, incorporating the underlying rate of event k in a population (α), the coefficients determining the risk factor effect on risk (β), the time-weighted average of LDL-C and SBP from age 18 years to the present, and the remaining ASCVD risk factors (RFs).

Risk functions were recalibrated so that modeled event rates matched contemporary U.S. ASCVD incidence and total event, ASCVD-related mortality, and all-cause mortality rates (Supplemental Figure 2). Results were recalibrated with data from a previously validated population simulation version of the CVD Policy Model and U.S. Centers for Disease Control and Prevention data. Recalibration was conducted separately for women and men.

Simulation Cohort

The simulated cohort was created by sampling ASCVD-free young adult participants from U.S. NHANES 1999 to 2014 examinations. Participants were excluded if they reported a diagnosis of heart failure, stroke, or CHD. Lifetime (aged 18–89 years) ASCVD risk factor trajectories were assigned to each NHANES participant by randomly matching them 1:1 to NHLBI-PCS participants for whom trajectories were previously defined (Supplemental Table 6, Supplemental Figure 3).[1,13,14] We repeatedly sampled with replacement the NHANES–NHLBI-PCS matched participants to create 100 cohorts, each consisting of 250,000 young adult men and 250,000 young adult women.

Simulated Treatment Strategies

Every individual was simulated through the model from their age at NHANES observation until age 89 years or death. Lipid screening commenced at the start of the simulation and repeated every 5 years throughout an individual's life while untreated and without ASCVD. To avoid unnecessary inconvenience and testing costs for patients unlikely to require lipid management, screening was not continued for young adults with an LDL-C level of ≥10 mg/dL below the treatment initiation threshold at first screening. For these individuals, screening recommenced at age 40 years.

Lifetime ASCVD, survival, and cost-effectiveness outcomes were projected for different LDL-C treatment strategies. The reference "standard care" strategy was defined according to the 2018 AHA/ACC guideline (Supplemental Appendix).[7] Standard care was statin treatment for adults aged ≥40 years based on LDL-C, ASCVD risk, or diabetes plus young adults with LDL-C ≥190 mg/dL. Each young adult treatment strategy was supplemental to standard care.

Four treatment strategies were compared with standard care in the primary analysis. The first 2 initiated moderate-intensity statins in young adults with LDL-C of ≥160 mg/dL or ≥130 mg/dL. The second 2 initiated intensive lifestyle modification in young adults with LDL-C of ≥160 mg/dL or ≥130 mg/dL. All individuals were treated according to ACC/AHA 2018 guidelines after age 40 years. A secondary analysis included moderate-intensity statins plus intensive lifestyle intervention in young adults with LDL-C of ≥160 mg/dL and LDL-C of ≥130 mg/dL as a comparator.

Treatment Inputs

Treatment input parameters are outlined in Table 1.

For statin treatment strategies, the magnitude of clinical benefit was a function of the magnitude of LDL-C reduction. Moderate- and high-intensity statin therapy reduced LDL-C by 29% and 43%, respectively.[16] For adults aged ≥40 years, we applied the relative risk (RR) of statins on ASCVD per 1.0-mmol/L (38.67-mg/dL) reduction in LDL-C from clinical trials.[5] For young adults, the lipid-lowering benefit was modeled through the counterfactual of lower ASCVD risk with lower time-weighted average LDL-C observed in longitudinal cohort data.[13]

For lifestyle interventions, the magnitude of clinical benefit corresponded to LDL-C and SBP reductions.[6] Intensive lifestyle interventions reduced LDL-C by 0.05 mmol/L (2.1 mg/dL) and SBP by 1.9 mm Hg, which were derived from a U.S. Preventive Services Task Force meta-analysis of behavioral counseling for ASCVD prevention.[6] Lifestyle interventions also reduced risk of incident diabetes (RR: 0.67) throughout the course of treatment.

Clinical benefit was greater for individuals receiving lipid-lowering therapy in young adulthood because earlier intervention results in lower cumulative LDL-C exposure over the adult life course. To benchmark this assumption, we compared the simulated CHD RRs of young-adult and later-life LDL-C lowering with the RR of 0.78 per 17 mg/dL lifetime lower LDL-C observed in a mendelian randomization study (PCSK9 R46L loss-of-function mutation) (Supplemental Table 7).[17,18] As expected, the simulated RRs for CHD per 17 mg/dL of LDL-C lowering in young adulthood and in later life were higher, at 0.86 and 0.92, respectively.

Adherence to statin therapy (ie, the proportion continuing with treatment beyond the persistence observed in clinical trials) was assumed to be 67%, 53%, and 50% in the first, second, and subsequent years of treatment, respectively.[19] Persistent statin users experienced slight increases in diabetes risk, annual pill-taking utility decrements, and follow-up physician visits.[7,20,21] Although statins are generally well tolerated,[22] costs were also included for minor and major statin-related adverse events, which occurred in 4.7% and 0.006% of statin users, respectively.[21] Based on expert opinion, treatment effect for intensive lifestyle interventions did not persist past 5 years. Low adherence or abbreviated treatment attenuated LDL-C reduction, side effect risks, and treatment-related costs. In a sensitivity analysis, we examined outcomes when the duration of treatment effect with full adherence was varied from 1 to 22 years.

The annual cost of statins was calculated as the median cost of statin therapy to all payers, including patient out-of-pocket costs, in the 2017 Medical Expenditure Panel Survey (Supplemental Table 8).[23] The cost of lifestyle intervention was the median cost of group therapy from a systematic review of economic evaluations of diet and physical activity promotion programs for the prevention of type 2 diabetes.[24] Few behavioral intervention studies extend beyond 1 year, and the optimal regularity of long-term behavioral visits with physicians has not been established. Based on expert opinion, we determined that patients receiving lifestyle modification treatment would incur costs for 4 annual cardiovascular prevention behavioral visits in years subsequent to treatment initiation.

Main Analysis

The primary model outcomes were lifetime QALYs gained; direct health care costs in 2020 U.S. dollars; and incremental cost-effectiveness ratios (ICERs), stratified by sex. The mean and 95% uncertainty interval (95% uncertainty interval; ie, the 2.5th to 97.5th percentile interval of the means) were calculated for model outcomes from 100 probabilistic simulations in which model inputs were sampled from prespecified distributions. Strategies were classified as highly cost-effective (ICER: <US$50,000/QALY gained), intermediately cost-effective (ICER: ≥US$50,000/QALY but <US$150,000/QALY gained), or not cost-effective (ICER: ≥US$150,000/QALY gained), in accordance with the "ACC/AHA Statement on Cost/Value Methodology in Clinical Practice Guidelines and Performance Measures".[25,26] We performed all analyses from a U.S. health care sector perspective, which includes all formal health care costs (ie, treatment related, ASCVD, and non-ASCVD health care costs) regardless of payer (eg, patients, insurance companies, and so on). Future QALYs and costs were discounted 3.0% annually.[27]

Sensitivity Analysis

One-way sensitivity analyses estimated the effect of changing the model inputs (Table 1) one at a time across the upper and lower uncertainty bounds while holding all other parameters constant (Supplemental Appendix). In a further sensitivity analysis, we examined outcomes when the duration of the treatment effect with full adherence was varied from 1 to 22 years.

This study followed the Consolidated Health Economic Evaluation Reporting Standards reporting guideline (Supplemental Table 9).