Date of Award


Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)



First Advisor

Eileen Stuart-Shor

Second Advisor

Suzanne Leveille

Third Advisor

Philimon Gona


Background: In recent years, non-Laboratory based (non-LB) risk assessment algorithms have been developed to facilitate absolute cardiovascular disease (CVD) risk assessment in resource constrained primary care settings. The non-LB Framingham algorithm, which substitutes body mass index (BMI) for lipids, has the best discrimination and calibration among the published algorithms, but its external validity and cost-effectiveness have not been determined.

Purpose: External validation and comparative effectiveness analysis of the non-LB versus laboratory based (LB) Framingham algorithm in a racially diverse population, and simulated cost-effectiveness analysis focusing on a black sample.

Methods: Secondary data analysis was performed using the Atherosclerosis Risk in Communities (ARIC) dataset. Cox regression models including the non-LB and LB Framingham covariates were developed. Model discrimination was assessed using the C statistic, calibration using the goodness-of-fit test, and equivalence of regression coefficients using the z-test. Algorithms based on the models were developed and their performance assessed using the area under receiver operating characteristic curve (AUROC), and agreement using kappa statistics. Analyses using simulated incremental cost-effectiveness ratios (ICER) were focused on the black sample. IRB approval was obtained. Data were analyzed using Stata© software version 14.

Results: Among 11,601 individuals (mean age 53.9 ± 5.7 years, 55% female, 24% black), the non-LB versus LB models performed as follows: C statistic (0.75 vs 0.76 for women, & 0.67 vs 0.68 for men); goodness-of-fit (14.2 vs 10.5 for women, & 25.8 vs 21.8 for men) respectively. In the black sample, regression coefficients of all covariates were similar to those generated in Framingham (z = ±1.96). The two algorithms based on the models had a kappa statistic of 0.76. When used to stratify risk in the entire ARIC sample, the non-LB and LB Framingham algorithms had AUROC of 0.706 vs 0.710 respectively. Prevention program guided by the non-LB Framingham dominated those guided by individual risk factors and LB Framingham algorithm.

Conclusions: These results demonstrate the validity and cost-effectiveness of the non-LB Framingham algorithm. This approach could provide a valuable and efficient alternative to the traditional LB approaches in the ongoing efforts to address the high burden of CVD in underserved communities.