Cost‐Effectiveness of Coronary Artery Calcium Scoring for Cardiovascular Disease Prevention in Diabetes: An Analysis From MESA

Abstract

Background

Coronary artery calcium scoring (CACS) by computed tomography could enhance risk assessment and decision making for preventive medication in patients with diabetes. We performed a microsimulation study to compare costs and health outcomes of guideline‐based periodic cardiovascular risk assessment with and without CACS.

Methods

We modeled various US guideline‐based preventive approaches based on periodically assessed 10‐year risk by pooled cohort equations with and without CACS. We predicted cumulative costs and quality‐adjusted life years (QALYs) until age 100 years from the US health care sector perspective in MESA (Multi‐Ethnic Study of Atherosclerosis) participants aged 45 to 84 years with diabetes (n=853), who were weighted to represent the US general patient population. Probabilistic and deterministic sensitivity analyses were performed to address uncertainty.

Results

Initiating high‐intensity statins regardless of risk and low‐dose aspirin if 10‐year risk ≥10% led to the largest QALY gains with incremental cost‐effectiveness ratios of $35 000 to $40 000/QALY. When omitting such universal approaches, allocating high‐intensity statins and low‐dose aspirin if CACS ≥100 led to incremental cost‐effectiveness ratios around $50 000/QALY. Ranking of strategies by cost‐effectiveness was generally robust against parameter uncertainty. The incremental cost‐effectiveness ratio of the CACS ≥100 strategy fell below $50 000/QALY if the fee of CACS fell below $75 or when statin continuation was assumed to significantly improve with nonzero CAC scores.

Conclusions

Broadening the use of high‐intensity statins and low‐dose aspirin in patients aged 45 to 84 years with diabetes can be considered cost‐effective. If broad‐scale use of intensive preventive treatment is either not feasible or not desired, then CACS may be cost‐effective in refining preventive treatment decisions.

Nonstandard Abbreviations and Acronyms

ADA

American Diabetes Association

AHA

American Heart Association

CACS

coronary artery calcium scoring

ICER

incremental cost‐effectiveness ratio

MESA

Multi‐Ethnic Study of Atherosclerosis

PCE

pooled cohort equations

PREVENT

Predicting Risk of Cardiovascular Disease EVENTs

QALY

quality‐adjusted life year

Clinical Perspective.

What Is New?

• Conducting empirical studies to assess the value of enhancing guideline‐based periodic cardiovascular risk assessment by coronary artery calcium scoring in diabetes is challenging.

• This microsimulation study shows that, if feasible, initiating high‐intensity statins regardless of cardiovascular risk and low‐dose aspirin if 10‐year risk is ≥10% in patients aged 45 to 84 years with diabetes is most cost‐effective; otherwise, allocating intensive treatments if coronary artery calcium scoring ≥100 could be considered cost‐effective, especially when the fee for performing coronary artery calcium scoring fell below $75.

What Are the Clinical Implications?

• Broadening the use of high‐intensity statins and low‐dose aspirin in patients aged 45 to 84 years with diabetes is effective and economically attractive, but if broad‐scale use of intensive preventive treatment is either not feasible or not desired, then coronary artery calcium scoring may be cost‐effective in refining intensive preventive treatment decisions.

Diabetes has traditionally been considered a coronary heart disease (CHD) risk equivalent, that is, a condition that puts individuals at high risk for cardiovascular disease (CVD) independent of other cardiovascular risk factor levels. Therefore, for individuals at middle age or older with diabetes, US cardiovascular prevention guidelines recommend moderate‐intensity statins and blood pressure–lowering treatment regardless of assessment of total cardiovascular risk. Recommendations for cardiovascular risk assessment are restricted to guide preventive treatment with high‐intensity statins and low‐dose aspirin when the predicted cardiovascular risk is high. ¹ , ² , ³

However, recent studies have challenged the concept of diabetes as CHD risk equivalent. For example, in >1.5 million Kaiser patients, CHD was associated with a hazard ratio of 2.80 (95% CI, 2.70–2.85) for subsequent CHD events versus a much lower hazard ratio of 1.70 (95% CI, 1.66–1.74) for diabetes alone. ⁴ It can be shown that cardiovascular risk is heterogeneously distributed in patients with diabetes, and risk can be explained by established cardiovascular risk factors as well as more novel markers including coronary artery calcium scoring (CACS) by computed tomography. ⁵ Epidemiological studies have shown cardiovascular risk predictions can be enhanced with CACS regardless of diabetes status. ⁶ , ⁷ , ⁸ , ⁹ In addition, communication of elevated CAC scores has been associated with increased uptake of preventive interventions ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ and could potentially enhance continuation of statin therapy once initiated. ¹³ , ¹⁵ Finally, obtaining a CAC score of zero could help guide decision making about withholding preventive treatments that have not been initiated yet. ¹⁶ Consequently, also in individuals with diabetes, CACS could be used to improve guidance of intensive preventive treatments.

However, conducting empirical studies to assess the effectiveness of CACS‐based preventive strategies in diabetes is challenging. In addition, potential improvements in health outcomes should be evaluated in the context of economic costs. Modeling studies can be used to determine which strategies add best value to provide guidance to clinicians and policymakers. Recently published economic evaluations of CACS‐based preventive strategies ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ generally did not focus on patient populations with diabetes. One Canadian modeling study evaluated cost and effectiveness outcomes of CACS‐guided risk assessment in adults with diabetes, but these were all assumed not to be on prior statin therapy. Only differences in initiation of low‐dose statin therapy were modeled, which depended on elevated CAC scores. ²⁴ However, trade‐offs for intensified preventive treatments including high‐intensity statin therapy should be considered for patients with diabetes in the context of enhancing risk assessment by CACS. Therefore, we performed a microsimulation study to compare the long‐term health and economic outcomes expected from CACS‐based approaches versus more traditional preventive strategies.

Methods

Study Design and Population

We updated an established microsimulation state‐transition model ⁵ in R version 4.3.1 (R Foundation for Statistical Computing; http://www.r‐project.org/) and TreeAge Pro 2023 software (TreeAge Software, Inc., Williamstown, MA). The model was constructed using data from 6770 MESA (Multi‐Ethnic Study of Atherosclerosis) ²⁵ , ²⁶ participants aged 45 to 84 years (mean age, 62 years, n=853 with diabetes). Within the model, individuals could transition to a health state for having experienced 1 of 3 CVD events (CHD, stroke, or heart failure), and death (Figure S1). Transitions took into account competing risks and were individualized using cause‐specific Cox regression and logistic regression for 30‐day case fatality. For the current study, we additionally incorporated longitudinal data on CACS, risk factors, and medication use. Validity of the updated model was reassessed within MESA participants with diabetes (see Data S1 and Tables S1 through S4).

MESA data can be obtained under a data use agreement from the MESA Coordinating Center or National Heart, Lung, and Blood Institute Biologic Specimen and Data Repository Information Coordinating Center. For details on MESA, see Data S1. All MESA participants gave informed consent, and the study protocol was approved by the institutional review board at each site. The Mount Sinai Hospital institutional review board deemed the study protocol for our modeling study as exempt from institutional review board review.

Modeled Preventive Strategies

We defined strategies by assessing statin, blood pressure–lowering, and low‐dose aspirin treatment recommendations within US CVD prevention guidelines by group discussions during monthly calls. We considered recommendations issued by the American Diabetes Association (ADA) and American Heart Association (AHA), which were identical (“ADA/AHA”), as well as the US Preventive Services Task Force (“USPSTF”). ¹ , ² , ³ All investigators agreed with the final strategies to model before running the analyses.

For the base case analysis, we evaluated preventive practice as defined by modeled data without modification (“observed practice”), and modified practice strategies based on systematic implementation of “ADA/AHA” recommendations, “USPSTF” recommendations, as well as recommendations for using CACS available within ADA and AHA guidelines (“CACS, ADA/AHA”). ¹ , ²⁷ An “intensive treatment” strategy reflected modified practice using the broadest eligibility criteria possible for initiation of intensive pharmacological treatment. We considered the latter to be a benchmark strategy and potentially infeasible to implement and therefore presented cost‐effectiveness outcomes with and without its inclusion.

In each modified practice strategy, periodic cardiovascular risk assessment using pooled cohort equations (PCE) ²⁸ was systematically offered to assess 10‐year atherosclerotic CVD (ASCVD) risk from the age of 45 through 75 years or until all preventive medications were initiated. Intervals for cardiovascular risk assessment were every 5 years, and in the intensive treatment strategy, periodicity was increased to annual. We assumed preventive drugs were prescribed for patients aged >75 years as observed in MESA. For “CACS, ADA/AHA”, CACS was offered only when it could potentially alter decision making for preventive medication beyond the traditional criteria. We modeled full uptake of periodic risk assessment and CACS, assuming additional tests could be integrated within routine medical examinations.

Moderate‐ and high‐intensity statins were prescribed on the basis of 10‐year ASCVD risk and low‐density lipoprotein (LDL) cholesterol threshold recommendations. Annual blood pressure measurement and prescription of blood pressure–lowering therapy was modeled on the basis of criteria outlined in “ADA/AHA” guidelines ¹ , ²⁹ and did not differ across the 4 modified practice strategies. Within scenario analyses, we modeled recommendations for low‐dose aspirin and scenarios in which preventive treatments would be withheld when CACS indicated absence of coronary artery calcium (“CAC zero”) (see Table 1). ¹ , ²⁷ , ³⁰ We did not include a “CAC zero” strategy in which statin dose would be decreased after obtaining a zero CAC score for 2 reasons. First, very few MESA subjects with diabetes had both high‐intensity statin use at baseline and a zero CAC score (n=2). Second, development of a zero CAC score after a nonzero CAC score was not possible.

Table 1.

Overview of Assumptions in Modeled Preventive Strategies

Strategy	Statin therapy	BP‐lowering therapy	Low‐dose aspirin therapy (in scenario analysis only)
Observed practice	Cardiovascular risk assessment for statins done opportunistically Statin use based on MESA Costs reflected by MEPS	BP measurement done opportunistically Antihypertensive med use based on MESA Costs reflected by MEPS	Aspirin use based on MESA Costs reflected by MEPS
ADA/AHA	5‐yearly cardiovascular risk assessment with 10‐yr PCE ages 40–75 y if not optimally treated Moderate‐intensity statins if risk <20% and LDL‐C 70–190 mg/dL High‐intensity statins if LDL‐C ≥190 mg/dL or PCE risk ≥20% and LDL‐C 70–190 mg/dL	BP measurement annually Antihypertensive meds if BP ≥130/80 with BP target <130/80 to achieve by titration N meds	Low‐dose aspirin if PCE risk ≥20% and age ≤ 70 y
USPSTF	5‐yearly cardiovascular risk assessment with 10‐y PCE ages 40–75 y if not optimally treated Moderate‐intensity statins if PCE risk ≥10% and LDL‐C 70–190 mg/dL High‐intensity statins if LDL‐C ≥190 mg/dL	As modeled for ADA/AHA	Low‐dose aspirin if PCE risk ≥20% and age ≤59 y
CACS, ADA/AHA	As modeled for ADA/AHA, perform CACS if eligible for statins, not on high‐intensity statins, and no prescription indication already based on PCE risk and LDL‐C High‐intensity statins if CAC score ≥100 Alternative strategy: No statin if zero CAC score	As modeled for ADA/AHA	As modeled for ADA/AHA, perform CACS if eligible for low‐dose aspirin, not on aspirin and no prescription indication already based on PCE risk and age Aspirin if CAC score ≥100 Alternative strategies: Statin as modeled for ADA/AHA, no aspirin if zero CAC score No statin, no aspirin if zero CAC score
Intensive treatment	Annual cardiovascular risk assessment with 10‐y PCE ages 40–75 y if not optimally treated High‐intensity statins if LDL‐C ≥70 mg/dL	As modeled for ADA/AHA	Aspirin if PCE risk ≥10% and age ≤70 y

ADA indicates American Diabetes Association; AHA, American Heart Association; BP, blood pressure; CAC, coronary artery calcium; LDL‐C, low‐density lipoprotein cholesterol; MEPS, Medical Expenditure Panel Survey; MESA, Multi‐Ethnic Study of Atherosclerosis; PCE, pooled cohort equations; and USPSTF, US Preventive Services Task Force.

Treatment Uptake, Efficacy, and Safety

For uptake of preventive medication prescriptions, we used “time to new uptake” and “time to discontinuation” Cox regression models of longitudinal prescription medication use data from MESA exams 1 through 5 (see Data S1 and Table S4 for hazard ratios). To account for parameter uncertainty, all Cox regression models and baseline hazard functions were refitted in bootstrap data sets. Individuals who became nonadherent to newly prescribed medication could restart once within a later screening visit. When adherent, relative treatment effects based on randomized trials were applied assuming uptake of prescriptions in these randomized trials was practically optimal. When applicable, the relative treatment effects for statin therapy were based on the expected absolute change in LDL cholesterol, and for blood pressure lowering on the expected absolute change in systolic blood pressure. Changes in cholesterol and blood pressure levels were assessed at the time of prescription. We incorporated adverse drug event rates when evidence from randomized trials indicated significant differences for those with and without preventive drug usage. These were defined as myopathy for statin use, intolerable and serious adverse drug events for blood pressure lowering, and gastrointestinal bleeding for low‐dose aspirin. We assumed a similar relative increase of myopathy rates with moderate versus high‐intensity statins. ³¹ For details see Data S1 and Tables S5 and S6.

Cost and Utility Weight Data

We included cost and utility weight data from a US health care sector perspective. The Second Panel on Cost‐Effectiveness in Health and Medicine recommends using community‐based preferences for health states as the most appropriate source of preferences for reference case analyses using the health care sector perspective. ³² We used multivariable regression modeling of weighted 2012 to 2020 Medical Expenditure Panel Survey data to include individual‐level long‐term annual cost and community‐based utility weights. ³³ Models included age; sex; race and ethnicity; education; insurance; diabetes; other comorbidity; and history of CHD, stroke, or heart failure. The independent impact of eliminating or inducing CVD events was assumed to be causal and represented by the adjusted regression coefficients of the latter 3 variables. Thus, while modeling counterfactual CVD rates under different preventive strategies, the impact of CVD on long‐term annual cost and utility weights was determined by switching these variables on/off in the model. Uncertainty was included by refitting all prediction equations in bootstrap data sets accounting for the survey design. For modeled individuals that were newly prescribed statin, antihypertensive, or aspirin medication, drug prices were added on the basis of prior analyses and Medical Expenditure Panel Survey data. ³⁴ , ³⁵ , ³⁶ We also included a disutility of daily preventive pill use with values obtained for the US general population using time trade‐off methods. ³⁷ Preferences from the community may differ from those obtained in individuals with experience taking preventive pills daily and do not reflect heterogeneity in adaptation to the experience of preventive pill use. When modeled individuals experienced an adverse event, a short‐term cost and disutility was applied on the basis of published data. ³⁴ , ³⁵ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² Medical expenditure data were used from different calendar years and thus potentially represented different amounts of purchasing power. We assumed that the all‐payer reimbursements from Medical Expenditure Panel Survey data and expenditures from other sources were a proxy for the underlying resource costs and adjusted for price inflation using the Personal Consumption Expenditures Health Index with the financial year 2022 as the base year. ³³ , ⁴³ For details, see Data S1 and Tables S7 and S8.

Analyses of Average Cost and Health Outcomes

For each preventive strategy, we modeled lifetime outcomes through age 100 within probabilistic sensitivity analyses on the basis of 500 bootstrap MESA data sets and randomly sampled parameters for non‐MESA data. Virtual life courses (N=250 000) were generated for 500 sampled MESA subjects aged 45 to 84 years with diabetes at baseline. To make the simulations representative of the US patient population with diabetes, MESA participants were sampled by weighting their risk profiles according to distributions observed in pooled National Health and Nutrition Examination Survey 2017 to March 2020 data. Weights were developed using multivariable inverse odds estimates (see Data S1). ⁴⁴

We calculated aggregated outcomes for clinical events, health care resource use, costs, and quality‐adjusted life years (QALYs). Costs and QALYs were discounted at the recommended annual rate of 3%. ³² Incremental cost‐effectiveness ratios (ICERs), defined as the difference in costs divided by the difference in QALYs, were calculated for undominated scenarios after ordering strategies by increasing average cost. Strategies were “absolutely dominated” when they were less effective and more costly than another strategy or “extendedly dominated” when they were less efficient due to smaller QALY gains for the same incremental costs. ICERs <$50 000/QALY were considered indicative of high value, whereas ICERs ≥$150 000/QALY were considered indicative of low value. ⁴⁵

Uncertainty and Additional Scenario Analyses

To report overall parameter uncertainty in results, we calculated 95% “credible” or “uncertainty” intervals (95% UIs) using results aggregated at the bootstrap/sampled parameter set with the percentile method. We also calculated the likelihood of being most cost‐effective (“acceptability”), and the expected loss per strategy at different cost‐effectiveness (“willingness to pay”) thresholds up to $200 000/QALY. ⁴⁶ For these measures, we used the net monetary benefit, defined as $\mathrm{Net}\text{monetary benefit}=\text{QALY}\times \text{Willingness to}\mathrm{pay}-\text{Cost}$ as outcome, which combines effectiveness and costs into 1 outcome facilitating interpretation of the results. ⁴⁷ To investigate the impact of single parameters on cost‐effectiveness results, we used metamodeling of net monetary benefit results. In a separate threshold analysis, we varied the fee of performing CACS. Finally, in scenario analyses, we assumed (1) significantly improved continuation of statin therapy following communication of a nonzero CAC score with improvement depending on the CAC category ¹⁵ ; and (2) full uptake of any new prescription medication and no change in uptake of existing prescription medication, regardless of the preventive strategy. For details, see Data S1.

Results

Validity and Generalizability

Predicted risks of cardiovascular and mortality outcomes were similar to cumulative incidences in 15‐year MESA data. In addition, incidence of CAC corresponded with incident CAC observed at examinations 2 and 3 (Table S9). After weighting data, risk factor distributions in MESA participants with diabetes approximated those estimated for the general US population with diabetes (Table S10).

Base Case Analysis

Implementation of “intensive treatment” resulted in the lowest overall lifetime CVD risk: 45.4% (95% UI, 38.4–52.4). The next optimal strategies for reducing CVD risk were “CACS, ADA/AHA” and “ADA/AHA”, which resulted in lifetime CVD risk of 45.6% (95% UI, 38.4–52.6) and 45.7% (95% UI, 38.6–52.5), respectively. In the “intensive treatment” strategy, more individuals received a statin prescription: 81.9% (95% UI, 78.2–85.3) versus 79.5% (95% UI, 75.8–83.0) and 76.3% (95% UI, 72.3–80.2) in “CACS, ADA/AHA” and “ADA/AHA” strategies, respectively, and also earlier, with generally fewer risk assessments required. The proportion receiving high‐intensity statins in the latter 2 strategies was 61.9% (95% UI, 57.2–66.2) and 53.0% (95% UI, 48.0–51.6). Proportions receiving antihypertensive medication and experiencing drug‐related adverse events were similar across preventive strategies (Table 2).

Table 2.

Disaggregated Health Outcomes Until Age 100 Years (95% UI)

Expected outcome	Observed practice	ADA/AHA	USPSTF	CACS, ADA/AHA	Intensive treatment
Clinical outcome risks
CVD, first event, %	52.9% (45.5 to 60.0)	45.7% (38.6 to 52.5)	46.4% (39.3 to 53.7)	45.6% (38.4 to 52.6)	45.4% (38.4 to 52.4)
Fatal first CVD event, %	16.3% (12.0 to 20.6)	14.6% (10.4 to 19.1)	14.8% (10.4 to 19.2)	14.5% (10.4 to 18.9)	14.5% (10.4 to 18.8)
First CHD event, %	23.7% (18.1 to 29.3)	20.9% (15.6 to 26.4)	21.4% (16.2 to 26.9)	20.8% (15.4 to 26.3)	20.7% (15.2 to 26.2)
First stroke event, %	15.8% (11.0 to 21.2)	13.1% (8.8 to 18.4)	13.2% (8.9 to 18.5)	13.1% (8.6 to 18.5)	13.0% (8.6 to 18.3)
First HF hospitalization, %	21.8% (15.6 to 28.3)	17.1% (11.6 to 22.9)	17.4% (11.9 to 23.2)	17.1% (11.6 to 22.7)	17.1% (11.6 to 22.8)
Life expectancy (undiscounted), y	23.64 (22.28 to 25.19)	24.25 (22.91 to 25.72)	24.19 (22.83 to 25.65)	24.27 (22.91 to 25.70)	24.29 (22.92 to 25.74)
Δ risk of drug‐related events (vs observed practice strategy)
Myopathy, %	…	0.03% (−0.20 to 0.40)	0.02% (−0.20 to 0.20)	0.03% (−0.20 to 0.40)	0.03% (−0.20 to 0.40)
Intolerable ADE from antihypertension meds, %		0.36% (0.00 to 1.00)	0.36% (0.00 to 1.00)	0.36% (0.00 to 1.00)	0.36% (0.00 to 1.00)
Serious ADE from antihypertension meds, %	…	0.26% (0.00 to 0.80)	0.26% (0.00 to 0.80)	0.26% (0.00 to 0.80)	0.26% (0.00 to 0.80)
Gastrointestinal bleeding, %	…	0.19% (−0.60 to 1.00)	0.19% (−0.60 to 1.00)	0.20% (−0.60 to 1.10)	0.21% (−0.60 to 1.10)
Δ health care use outcomes (vs observed practice strategy)
Cardiovascular risk assessment, count, mean	…	2.12 (1.97 to 2.25)	2.44 (2.29 to 2.59)	1.92 (1.80 to 2.05)	1.84 (1.60 to 2.11)
CT for CACS, count, mean	…	…	…	1.12 (1.00 to 1.23)	…
Statins prescribed/switched, %	…	76.3% (72.3 to 80.2)	52.6% (47.4 to 58.3)	79.5% (75.8 to 83.0)	81.9% (78.2 to 85.3)
High‐intensity statins prescribed, %	…	53.0% (48.0 to 57.4)	2.3% (1.2 to 3.8)	61.9% (57.2 to 66.2)	81.9% (78.2 to 85.3)
Antihypertensive meds prescribed, %	…	46.6% (41.5 to 51.6)	46.5% (41.4 to 51.6)	46.6% (41.5 to 51.6)	46.6% (41.6 to 51.6)
Antihypertensive meds, number of meds added, mean	…	1.19 (1.03 to 1.34)	1.19 (1.03 to 1.34)	1.19 (1.03 to 1.34)	1.19 (1.03 to 1.34)
Aspirin prescribed, %	…	…	…	…	…

Simulated outcomes with 95% UIs based on 500 bootstrap data sets are given for simulations of reweighted risk profiles from MESA participants with diabetes at baseline (N=831). ADA indicates American Diabetes Association; ADE indicates adverse drug event; AHA, American Heart Association; CACS, coronary artery calcium scoring; CHD, coronary heart disease; CT, computed tomography; CVD, cardiovascular disease; HF, heart failure; and USPSTF, US Preventive Services Task Force.

Average costs and QALYs were highest with the “intensive treatment” strategy. Its ICER was $39 136/QALY compared with the next undominated option, “ADA/AHA”, which had an ICER of $34 992/QALY. When “intensive treatment” was omitted, “CACS, ADA/AHA” was most effective, with gains of 0.005 QALY (95% UI, −0.034 to 0.050) versus “ADA/AHA”. It was also a potentially cost‐effective scenario at a threshold slightly higher than $50 000/QALY (Table 3 and Figure 1A).

Table 3.

Cost‐Effectiveness Outcomes (95% UI)

Strategies ranked by increasing costs	Costs	QALYs	Incremental costs*	Incremental QALYs*	ICER $/QALY*
Base case analysis
Observed practice	$299 151 (262 914 to 337 593)	10.371 (9.931 to 10.862)	…	…	…
USPSTF	$300 917 (264 271 to 338 709)	10.552 (10.129 to 11.062)	$1766 (−7013 to 9453)	0.181 (0.017 to 0.354)	$9772
ADA/AHA	$301 678 (265 515 to 339 325)	10.574 (10.130 to 11.067)	$761 (−3020 to 4465)	0.022 (−0.059 to 0.107)	$34.992
CACS, ADA/AHA	$301 976 (265 958 to 340 019)	10.579 (10.130 to 11.073)	Extendedly dominated*($298 [−1636 to 1860])	Extendedly dominated*(0.005 [−0.034 to 0.050])	Extendedly dominated*($55 254)
Intensive treatment	$302 211 (266 286 to 340 310)	10.587 (10.123 to 11.076)	$533 (−2033 to 3008)	0.014 (−0.044 to 0.073)	$39 136
Scenario analyses
Observed practice	$299 151 (262 914 to 337 593)	10.371 (9.931 to 10.862)	…	…	…
USPSTF	$300 917 (264.271 to 338 709)	10.552 (10.129 to 11.062)	$1766 (−7013 to 9453)	0.181 (0.017 to 0.354)	$9772
USPSTF, aspirin	$300 947 (263 900 to 338 926)	10.554 (10.129 to 11.053)	$30 (−703 to 716)	0.002 (−0.017 to 0.024)	$18 544
CACS, ADA/AHA: no statin|CACS zero	$301 337 (263 890 to 339 702)	10.555 (10.092 to 11.057)	Extendedly dominated	Extendedly dominated	Extendedly dominated
CACS, ADA/AHA, aspirin: no statin, aspirin|CACS zero	$301 388 (263 890 to 339 702)	10.558 (10.090 to 11.052)	Extendedly dominated	Extendedly dominated	Extendedly dominated
ADA/AHA	$301 678 (265 515 to 339 325)	10.574 (10.130 to 11.067)	Extendedly dominated	Extendedly dominated	Extendedly dominated
ADA/AHA, aspirin	$301 862 (264 642 to 340 114)	10.582 (10.150 to 11.087)	$915 (−3079 to 4844)	−0.029 (−0.055 to 0.117)	$31 956
CACS, ADA/AHA, aspirin: no aspirin|CACS zero	$301 954 (265 114 to 339 594)	10.578 (10.124 to 11.069)	Absolutely dominated	Absolutely dominated	Absolutely dominated
CACS, ADA/AHA	$301 976 (265 958 to 340 019)	10.579 (10.130 to 11.073)	Absolutely dominated	Absolutely dominated	Absolutely dominated
CACS, ADA/AHA, aspirin	$302 158 (265 218 to 340 170)	10.588 (10.145 to 11.091)	Extendedly dominated*($296 [−1596 to 1907])	Extendedly dominated*(0.006 [−0.035 to 0.051])	Extendedly dominated*($53 309)
Intensive treatment	$302 211 (266 286 to 340 310)	10.587 (10.123 to 11.076)	Absolutely dominated	Absolutely dominated	Absolutely dominated
Intensive treatment, aspirin	$302 468 (265 353 to 340 515)	10.600 (10.162 to 11.069)	$606 (−2233 to 3686)	0.017 (−0.052 to 0.090)	$34 896

Simulated outcomes with 95% UIs based on 500 bootstrap and parameter samples are given for simulations of reweighted risk profiles from MESA participants with diabetes at baseline (N=831). ADA indicates American Diabetes Association; AHA, American Heart Association; CE indicates cost‐effectiveness; ICER, incremental cost‐effectiveness ratio; QALY, quality‐adjusted life year; and USPSTF, US Preventive Services Task Force.

^*Calculated by comparison with the preceding undominated scenario, relevant if the intensive treatment strategy is not an option.

Figure 1. Cost‐Effectiveness Graphs of Base Case and Scenario Analyses.

Shown are mean estimates of costs (in US dollars) and QALYs obtained from the base case (A) and scenario analysis including aspirin‐based preventive strategies (B). The dotted line reflects the efficiency frontier with slopes equal to the incremental cost‐effectiveness ratios for undominated strategies. ADA indicates American Diabetes Association; AHA, American Heart Association; CACS, coronary artery calcium scoring; QALY, quality‐adjusted life year; and USPSTF, US Preventive Services Task Force.

Consideration of Additional Preventive Strategies

Strategies incorporating aspirin led to further small decreases in CVD rates, but also to excess gastrointestinal bleeding risk (Table S11). Strategies that used a zero CAC score to justify withholding preventive medication led to lower proportions of prescription of statin and aspirin than their equivalent strategies ignoring zero CAC scores (Table S12). The “intensive treatment, aspirin” strategy would be considered most attractive with an ICER of $34 896/QALY. When excluding this strategy, the “CACS, ADA/AHA, aspirin” strategy would be considered economically attractive if decision makers are willing to pay slightly more than $50 000/QALY (Table 3 and Figure 1B).

Uncertainty and Scenario Analyses

The likelihood of being most cost‐effective became highest for “intensive treatment” when thresholds ≥$42 500/QALY were used: 35% to 45% (Figure 2A). When leaving this strategy out, “CACS, ADA/AHA” had the highest likelihood for thresholds ≥$52 500/QALY: 35% to 46% (Figure 2B). Similarly, when alternative preventive scenarios of aspirin and CACS were considered, the “intensive treatment, aspirin”, and “CACS, ADA/AHA, aspirin” strategies achieved the highest probability of being cost‐effective if thresholds were ≥$35 000/QALY (21%–36%) and ≥$90 000/QALY (21%–25%), respectively (Figures 2C and 2D).

Figure 2. Cost‐Effectiveness Acceptability Curves.

Curves demonstrate the probability (%) of each strategy being the most cost‐effective strategy defined by its net monetary benefit for the base case (A with B excluding the intensive treatment strategy) and scenario analysis including aspirin‐based preventive strategies (C with D excluding intensive treatment strategies). Probabilities equal the obtained percentage of probabilistic sensitivity analysis iterations in which the strategy of interest's net monetary benefit was highest given the selected cost‐effectiveness threshold on the x axis. ADA indicates American Diabetes Asscociation; AHA, American Heart Association; CACS, coronary artery calcium scoring; CE, cost‐effectiveness; QALY, quality‐adjusted life year; and USPSTF, United States Preventive Services Task Force.

Expected monetary loss was lowest and second lowest for these intensive treatment and CACS‐based strategies, particularly for thresholds just above $50 000/QALY (Figure S2). When considering all possible preventive strategies, the expected minimum loss across strategies (equivalent to the expected value of obtaining perfect information through further research) was $491 at a threshold of $50 000/QALY, $1263 at $100 000/QALY, and $2300 at $150 000/QALY.

Ranking of strategies by cost‐effectiveness was generally robust, especially at higher thresholds of $100 000/QALY and $150 000/QALY (Figures S3 through S8). At a threshold of $100 000/QALY, the cost‐effectiveness of “CACS, ADA/AHA” versus “ADA/AHA” was only sensitive to uncertainty around the efficacy of LDL cholesterol reduction for stroke and background annual health care use costs. In addition, when considering a threshold of $50 000/QALY, “CACS, ADA/AHA” was cost‐effective versus “ADA/AHA” if the price of performing CACS fell below $75 (Figure S9). Assuming significantly improved adherence to statin therapy following communication of nonzero CAC scores increased QALY gains for the “CACS, ADA/AHA” strategy against only a small amount of additional costs rendering its cost‐effectiveness outcomes comparable to those of intensive treatment (Figure S10). Finally, optimal uptake of recommended prescription medication regardless of the preventive strategy did not change conclusions about cost‐effectiveness (Figure S11).

Discussion

We found that systematically implementing guideline‐based recommendations for periodic cardiovascular risk assessment followed by preventive treatment in diabetes generally results in small decreases in CVD risks and gains in QALYs, while adverse drug event risks and cost increases were small. Initiating intensive preventive treatment including high‐intensity statins and low‐dose aspirin across wider ranges of cardiovascular risk led to the largest health benefits with ICERs that are generally considered high value. When omitting such universal intensive preventive treatment approaches, more restrictive strategies of initiating high‐intensity statin and low‐dose aspirin treatments when CAC scores are ≥100 in addition to 10‐year cardiovascular risk criteria can be considered useful if decision makers are willing to pay slightly more than $50 000/QALY.

QALY gains and cost savings as compared with “observed practice” were mainly driven by CVD rate reductions that follow from replacing moderate with high‐intensity statins (and adding low‐dose aspirin) in patients at sufficiently high risk. Because more than half of the modeled patient population already used statin therapy at baseline and due to the background trend of statin initiation, the window of opportunity for improved delivery was small. Even though health benefits from the modeled preventive strategies remained small, our results on their incremental cost‐effectiveness were generally robust. We identified 2 mechanisms by which the CACS‐based preventive strategy with prescription of high‐intensity statins if CAC scores were ≥100 could become more cost‐effective. The incremental cost‐effectiveness of this strategy would approximate the optimal “intensive treatment” strategy's outcome if (1) the strategy were less expensive by decreasing the fee of performing CACS below $75; or (2) statin adherence significantly improves after communication of nonzero CAC scores. With respect to the latter, few prospective studies have demonstrated higher statin continuation rates for nonzero CAC compared with zero CAC scores. However, an isolated causal effect of communication of CACS on medication continuation remains uncertain due to risk of bias. Further research is needed to ascertain causality. ¹³ Although we demonstrated the potential cost‐effectiveness of using CACS to select additional patients for intensive treatment, a more restrictive approach by reclassifying patients to low risk on the basis of absence of CAC, and withholding treatment did not appear attractive. Prior studies indicated that event rates in those with zero CAC are generally low, especially when zero CAC is detected at multiple occasions. ⁵ , ⁴⁸ , ⁴⁹ , ⁵⁰ Yet our results indicated that even though event rates are lower, the gains of avoiding prescription medication costs and pill disutility, as well as adverse drug events, do not outweigh the benefits from the small absolute number of cardiovascular events that still can be prevented.

Recent cost‐effectiveness studies of CACS ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ concluded that CACS can be considered a reasonable alternative to traditional preventive approaches, especially when the up‐front costs of CACS and disutility and costs of subsequent preventive treatment are low. However, none of these studies focused on assessing the value of CACS in diabetes. Recently, one economic evaluation based on a Markov model was published that evaluated the cost‐effectiveness of CACS‐guided risk counseling among adults with diabetes not on statin therapy. ²⁴ Results showed that, compared with traditional risk factor counseling, CACS‐guided counseling was cost‐effective, particularly in individuals aged 60 to 69 years.

Compared with this latter analysis, we used individual‐level simulations of the heterogeneous patient population with diabetes instead of focusing on a uniform patient population. This methodology allowed us to incorporate individualized CVD and mortality event rates, longitudinal risk factor trends, costs, and utility weights. We also considered patients who were already on statins and the value of increasing statin dose on the basis of elevated risk or CACS. We incorporated cardiovascular benefits of different statin doses that depend on the expected change in LDL cholesterol. In addition, we modeled the additional effect of cardiovascular risk assessment on blood pressure lowering due to identification of undiagnosed hypertension. Even though criteria for blood pressure management were assumed to be equal across preventive strategies, the decrease in cardiovascular event rates due to blood pressure lowering affects the absolute cardiovascular benefits expected from statin and aspirin, and thus cost‐effectiveness. Finally, we weighted MESA participant risk profiles to make our simulations more generalizable to the eligible US general population with diabetes.

Despite the many strengths of our analysis, a few important limitations should be mentioned. First, the AHA recently developed the Predicting Risk of Cardiovascular Disease EVENTs (PREVENT) equations as improvement to the PCE. ⁵¹ Risk assessment by the PCE potentially overestimates risk, and issues have been identified regarding omission of important risk factors and inclusion of Black race. Risk assessment with PREVENT can potentially replace risk assessment with the PCE in the existing American College of Cardiology/AHA prevention guideline framework. ⁵² However, it has been demonstrated that the application of PREVENT equations together with the currently recommended treatment thresholds is expected to significantly reduce eligibility for statins, blood pressure–lowering therapy, and aspirin. The AHA does not currently recommend replacing PCE with PREVENT, and inclusion of PREVENT equations in future American College of Cardiology/AHA guidelines may thus require revised risk thresholds. ⁵³ Moreover, there are various versions of PREVENT equations available that include different sets of predictors. Therefore, including strategies of PREVENT‐based risk assessment was considered beyond the scope of our study. Second, we did not model initiation of ezetimibe and other nonstatin therapies to optimize lipid management without the need for increasing statin dose or to intensify therapy after inadequate lowering of LDL cholesterol by statin therapy. ⁵⁴ However, from the LDL cholesterol threshold levels required for initiation of statin therapies (Table 1) and the expected relative reduction of LDL cholesterol (Table S5), it can be concluded that typically recommended LDL cholesterol goals are met with statins alone. Also, we did not model preventive treatment strategies with glucose‐lowering medications such as glucagon‐like peptide 1 receptor agonists and sodium–glucose cotransporter 2 inhibitors. These treatments have been shown to be associated with lower ASCVD and heart failure risk but can currently be very costly and are associated with adverse drug events. It has been suggested that CACS could potentially help prioritizing allocation to those most likely to benefit. ⁵⁵ However, to appropriately evaluate scenarios of incorporating nonstatin and glucose‐lowering agents in the CACS‐based preventive strategies, the traditional (non–CACS‐based) preventive strategies should also be expanded with inclusion of recommendations for their usage. Moreover, there are various scenarios possible for delivering these additional agents to patients with diabetes when ASCVD, heart failure, and chronic kidney disease are not established. As for all decision modeling studies, the model that we used is a simplified representation of hypothetical, scenarios aiming to capture the most important trade‐offs for the decision making. As such, we could not consider all possible preventive scenarios. Third, our analysis was restricted to the US health care sector perspective, and we did not consider informal health care and societal consequences that may differ across preventive strategies. Fourth, we did not incorporate a small effect of worsening glycemic control following statin initiation and replacing moderate by high‐intensity statins. In a recent meta‐analysis it was shown that hemoglobin A_1c values may increase by 0.06% (95% CI, 0.00–0.12) with initiation of moderate‐intensity and by 0.08% (95% CI, 0.07–0.09) with high‐intensity statin therapy when compared with placebo. ⁵⁶ The potential consequences of a required increased use of diabetes medication were not incorporated, as our model focused on modeling risk and adverse events not related to glycemic management. Finally, we did not incorporate potential associations of risk factors including elevated CACS with myopathy and gastrointestinal bleeding event rates. For example, harms of low‐dose aspirin may have been underestimated in those that were allocated to aspirin in our model. A recent study showed that when those with elevated CACS of ≥100 are treated with low‐dose aspirin, absolute reductions of 10‐year ASCVD risk are potentially offset by increases in bleeding risk, unless PCE risk is ≥20%. ⁵⁷ In contrast, our analysis took on a lifetime perspective and incorporated the consequences of these outcomes by weighting for costs and preferences.

If feasible, intensive treatment strategies appeared most cost‐effective in the eligible US patient population with diabetes. However, whether these can be implemented on a broad scale also depends on patient and provider attitudes toward earlier and more aggressive pharmacological approaches for CVD prevention. ⁵⁸ Results from preference elicitation studies on preventive pill use are in agreement with those from survey studies, and show that a large proportion of individuals have strong preferences against preventive pill use in general. ³⁷ , ⁵⁹ , ⁶⁰ Our study shows that the allocation to more intensive preventive treatments in diabetes can be guided by CACS with improved health outcomes at reasonable additional costs and without significantly increasing pill usage. More empirical studies that evaluate CACS versus risk score–only strategies ¹⁴ as well as broad‐scale intensive treatment approaches will need to be conducted in patients with diabetes to address the remaining uncertainty in feasibility, as well as effectiveness for health and economic outcomes.

Conclusions

Broadening the use of high‐intensity statins and low‐dose aspirin in patients aged 45 to 84 years with diabetes can be considered cost‐effective according to generally acceptable cost‐effectiveness thresholds. If broad‐scale use of intensive preventive treatment is either not feasible or not desired, then CACS may be cost‐effective in refining preventive treatment decisions.

Sources of Funding

This study was supported by American Diabetes Association grant number 1‐18‐ICTS‐041 (Drs Ferket, Hunink, Masharani, Max, and Fleischmann). MESA was supported by contracts 75N92020D00001, HHSN268201500003I, N01‐HC‐95159, 75N92020D00005, N01‐HC‐95160, 75N92020D00002, N01‐HC‐95161, 75N92020D00003, N01‐HC‐95162, 75N92020D00006, N01‐HC‐95163, 75N92020D00004, N01‐HC‐95164, 75N92020D00007, N01‐HC‐95165, N01‐HC‐95166, N01‐HC‐95167, N01‐HC‐95168, and N01‐HC‐95169 from the National Heart, Lung, and Blood Institute, and by grants UL1‐TR‐000040, UL1‐TR‐001079, and UL1‐TR‐001420 from the National Center for Advancing Translational Sciences. The American Diabetes Association had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.

Disclosures

Dr Hunink receives (or received in the past 3 years) royalties from Cambridge University Press for a textbook on medical decision making, until 2022 reimbursement of travel expenses from the European Society of Radiology for work on the European Society of Radiology guidelines for imaging referrals, and additional research funding from the Netherlands Organization for Health Research and Development, the German Innovation Fund, Netherlands Educational Grant (“Studie Voorschot Middelen”), and the Gordon and Betty Moore Foundation. All other authors declare that they have no conflicts of interest.

Supporting information

Data S1

Tables S1–S12

Figures S1–S11

References 61–72