The following is a summary of “Predictive accuracy of machine learning for the risk of death in HIV patients: a systematic review and meta-analysis,” published in the May 2024 issue of Infectious Disease by Li et al.
Predicting how long someone with HIV will live is difficult, but machine learning offers promise despite variations in illness and factors to consider.
Researchers conducted a retrospective study to evaluate the use of machine learning for predicting HIV-related deaths early in the disease course.
They searched PubMed, Cochrane, Embase, and Web of Science databases (November 25, 2023). The original studies utilized the Predictive Model Bias Risk Assessment Tool (PROBAST) to assess bias risk. Subgroup analysis during the meta-analysis was based on survival and non-survival models. Meta-regression was employed to investigate the impact of death time on the model’s predictive value for related deaths in HIV-related.
The results showed 24 pieces of literature involving data from 401,389 individuals diagnosed with HIV. Among them, 23 articles focused on deaths during long-term follow-ups outside hospital settings. The machine learning models used included COX regression for survival and non-survival models. In the training set, the c-index for predicting deaths among people with HIV (PWH) using predictive models was 0.83 (95% CI: 0.75–0.91), while in the validation set, it was 0.81 (95% CI: 0.78–0.85). The meta-regression analysis indicated that neither follow-up time nor the occurrence of death events significantly impacted the performance of the machine learning models.
Investigators concluded that machine learning showed promise for developing general predictions of deaths in HIV, but more studies from various centers are needed for confirmation.
Source: bmcinfectdis.biomedcentral.com/articles/10.1186/s12879-024-09368-z