Who needs laboratories and who needs statins?: comparative and cost-effectiveness analyses of non-laboratory-based, laboratory-based, and staged primary cardiovascular disease screening guidelines

Abstract

Background: Early detection and treatment of cardiovascular disease (CVD) risk factors produces significant clinical benefits, but no consensus exists on optimal screening algorithms. This study aimed to evaluate the comparative and cost-effectiveness of staged laboratory-based and non-laboratory-based total CVD risk assessment.

Methods and results: We used receiver operating characteristic curve and cost-effectiveness modeling methods to compare strategies with and without laboratory components and used single-stage and multistage algorithms, including approaches based on Framingham risk scores (laboratory-based assessments for all individuals). Analyses were conducted using data from 5998 adults in the Third National Health and Nutrition Examination Survey without history of CVD using 10-year CVD death as the main outcome. A microsimulation model projected lifetime costs, quality-adjusted life years (QALYs), and incremental cost-effectiveness ratios for 60 Framingham-based, non-laboratory-based, and staged screening approaches. Across strategies, the area under the receiver operating characteristic curve was 0.774 to 0.780 in men and 0.812 to 0.834 in women. There were no statistically significant differences in area under the receiver operating characteristic curve between multistage and Framingham-based approaches. In cost-effectiveness analyses, multistage strategies had incremental cost-effectiveness ratios of $52,000/QALY and $83,000/QALY for men and women, respectively. Single-stage/Framingham-based strategies were dominated (higher cost and lower QALYs) or had unattractive incremental cost-effectiveness ratios (>$300,000/QALY) compared with single-stage/non-laboratory-based and multistage approaches.

Conclusions: Non-laboratory-based CVD risk assessment can be useful in primary CVD prevention as a substitute for laboratory-based assessments or as the initial component of a multistage approach. Cost-effective multistage screening strategies could avoid 25% to 75% of laboratory testing used in CVD risk screening with predictive power comparable with Framingham risks.

Keywords: diagnosis; economics; primary prevention.

Figure 1

A hypothetical two-staged primary CVD screening strategy that incorporates non-laboratory-based risk assessment In a multistage screening framework, all individuals are assessed using non-laboratory-based risk assessment initially, and those at intermediate risk from the first stage are ultimately assessed using laboratory-based (Framingham) risk. xT is the laboratory-based treatment threshold, xL is the lower bound for laboratory testing (based on non-laboratory-based risk assessment), xU is the upper bound for laboratory testing (based on non-laboratory-based risk assessment). Compared to laboratory-based risk assessment strategy for all individuals, a multistage strategy would only result in different treatment decisions for individuals in regions I and VI. -Regions I and II would be characterized as high-risk and recommended for statin treatment but not recommended for laboratory testing from Stage 1 -Region III would be characterized as intermediate-risk and recommended for laboratory testing from Stage 1, but not recommended for statin treatment based on Stage 2 -Region IV would be characterized as intermediate-risk from and recommended for laboratory testing from Stage 1, and recommended for statin treatment based on Stage 2 -Regions V and VI would characterized as low-risk and not be recommended for either laboratory testing or statin treatment from Stage 1

Figure 2

Simplified depiction of the cardiovascular disease model In the micro-simulation model, all individuals begin in the “Disease Free without Treatment” state. Transitions to “Disease Free with Treatment” depend on the type of screening strategy being evaluated. All other transitions are based on published estimates, with adjustments made for statin treatment when applicable.

Figure 3

a. ROC curves (10-year CVD death outcome) for multistage and Framingham CVD risk scores, men. b. ROC curves (10-year CVD death outcome) for multistage and Framingham CVD risk scores, women. Receiver operator characteristic (ROC) curves for the three versions of the multistage screening strategy (with 25%, 50%, and 75% of the population receiving laboratory-based testing) and the Framingham CVD (“fram cvd”) scores, with 10-year CVD death as the outcome of interest, for individuals with complete data. For men (Figure 3a), the performances in risk discrimination, as assessed by AUC (i.e., area under the ROC curve) were 0.776, 0.774, 0.778, and 0.780 for the Framingham CVD and multistage (75%, 50%, and 25% of adults receiving laboratory-based risk assessments) risk scores, respectively, with a p-value for the differences compared to the Framingham score of 0.71, 0.74, and 0.57. For women (Figure 3b), the corresponding AUC results were 0.834, 0.827, 0.819, 0.812, with p-values for the differences compared to the Framingham score of 0.15, 0.14, and 0.06.

Figure 3

a. ROC curves (10-year CVD death outcome) for multistage and Framingham CVD risk scores, men. b. ROC curves (10-year CVD death outcome) for multistage and Framingham CVD risk scores, women. Receiver operator characteristic (ROC) curves for the three versions of the multistage screening strategy (with 25%, 50%, and 75% of the population receiving laboratory-based testing) and the Framingham CVD (“fram cvd”) scores, with 10-year CVD death as the outcome of interest, for individuals with complete data. For men (Figure 3a), the performances in risk discrimination, as assessed by AUC (i.e., area under the ROC curve) were 0.776, 0.774, 0.778, and 0.780 for the Framingham CVD and multistage (75%, 50%, and 25% of adults receiving laboratory-based risk assessments) risk scores, respectively, with a p-value for the differences compared to the Framingham score of 0.71, 0.74, and 0.57. For women (Figure 3b), the corresponding AUC results were 0.834, 0.827, 0.819, 0.812, with p-values for the differences compared to the Framingham score of 0.15, 0.14, and 0.06.