SCORE2 Tracks WHO Cardiovascular Risk Estimates Most Closely in a Congolese HIV Cohort

People living with HIV carry an increased burden of cardiovascular risk, but the risk calculators that clinicians rely on can produce substantially different estimates when applied to the same patient. A new study from Kikwit, in the Democratic Republic of the Congo, set out to measure just how far apart those estimates diverge in a real-world African clinical setting, comparing four widely used prediction models against the World Health Organization cardiovascular risk model.

The research, published in AIDS Research and Therapy, was a retrospective cohort study of adults living with HIV who were receiving antiretroviral therapy at two HIV care centers in Kikwit. Participants were enrolled between January 1 and June 30, 2026. For each individual, ten-year cardiovascular risk estimates were calculated using the Framingham, SCORE2, ASCVD, and QRISK3 models, and each of these was compared with the estimate produced by the WHO model, which served as the reference method because of its applicability to low- and middle-income settings, including sub-Saharan Africa. The authors emphasize that the WHO model was not treated as a clinical gold standard.

Agreement was assessed with a battery of statistical measures rather than a single metric. These included Lin’s concordance correlation coefficient, Pearson’s correlation coefficient, the bias-correction factor, mean differences with 95% limits of agreement from Bland-Altman analysis, the proportion of estimates falling within plus or minus 30 percent of the WHO estimate, and the root mean squared error. Together, these measures capture not only whether two models rank patients similarly but also how closely their absolute risk percentages match.

A total of 107 people living with HIV were included in the analysis. SCORE2 showed the highest concordance with the WHO model, with a concordance correlation coefficient of 0.937 (95% CI 0.909 to 0.957), a very high correlation of 0.997, a bias-correction factor of 0.940, and a mean difference of 1.72 percentage points. Every single SCORE2 estimate fell within 30 percent of the corresponding WHO estimate, a proportion of 100.0 percent.

Framingham also performed relatively well, with a concordance correlation coefficient of 0.818 (95% CI 0.743 to 0.872) and 92.5 percent of its estimates within 30 percent of the WHO value. ASCVD showed somewhat lower concordance, at 0.807 (95% CI 0.729 to 0.864), with a negative mean difference of 2.86 percentage points, meaning it tended to produce lower risk estimates than the WHO model. Notably, only 51.4 percent of ASCVD estimates fell within the 30 percent agreement band.

QRISK3 stood out as the clear outlier. It showed the lowest concordance with the WHO model, at 0.406 (95% CI 0.235 to 0.554), a much lower correlation of 0.628, a mean difference of 5.20 percentage points, and substantially wider limits of agreement ranging from 21.57 to 11.17 percentage points. Only 9.3 percent of QRISK3 estimates fell within 30 percent of the WHO estimate, indicating that in this cohort the two calculators frequently produced markedly different risk figures for the same patient.

The study’s authors are careful about what these findings do and do not mean. No prospective cardiovascular outcomes were available in the cohort, so the predictive performance, calibration, and discrimination of the models against observed cardiovascular events were not assessed. In other words, the study describes how closely the calculators agree with one another, not which one most accurately predicts who will actually have a heart attack or stroke. The authors state explicitly that the results should not be interpreted as evidence of superior prognostic accuracy for any model.

That limitation matters for clinical interpretation. Agreement between two risk models does not establish that either is correct, and a calculator that happens to track the WHO reference closely in this Congolese cohort could still misclassify risk in ways that only longitudinal follow-up can reveal. The WHO model itself was chosen as a comparator for its suitability to low- and middle-income settings, not because it has been validated as a definitive standard in this population.

The authors call for prospective studies with longitudinal cardiovascular outcomes to assess the calibration, discrimination, clinical utility, and comparative prognostic performance of cardiovascular risk prediction models in people living with HIV in the Democratic Republic of the Congo and other sub-Saharan African settings. Until such data exist, the practical takeaway from this study is narrower but still useful: in this Kikwit cohort, SCORE2 tracked WHO estimates most closely, while QRISK3 diverged most sharply, and clinicians interpreting risk scores in similar populations should be aware that the choice of calculator can materially change the number a patient receives.

Subject of Research: Agreement between cardiovascular risk prediction models among people living with HIV in the Democratic Republic of the Congo

Article Title: Agreement between 10-year cardiovascular risk prediction models among people living with HIV in Kikwit, Democratic Republic of the Congo

Article References: Lupenzi, B. M., Nkodila, A. N., Lubura, V. B., Kintoki, E. V., Longo-Mbenza, B., Longokolo, M., & Kasiam, J. B. (2026). Agreement between 10-year cardiovascular risk prediction models among people living with HIV in Kikwit, Democratic Republic of the Congo. AIDS Research and Therapy. https://doi.org/10.1186/s12981-026-00973-3

Image Credits: AI Generated

DOI: 10.1186/s12981-026-00973-3

Keywords: HIV, cardiovascular risk, risk prediction models, SCORE2, WHO risk model, QRISK3, Framingham, ASCVD, Kikwit, Democratic Republic of the Congo, antiretroviral therapy, Bland-Altman agreement