Advanced statistical approaches in tuberculosis diagnosis and treatment outcomes in Africa: a systematic review

Tuberculosis (TB) remains a major public health challenge in Africa; however, conventional statistical approaches often fail to capture diagnostic uncertainty, spatial heterogeneity, and complex disease dynamics, limiting evidence-based decision-making. This review synthesizes the application and performance of advanced statistical and computational methods for TB diagnosis and treatment outcomes in Africa.

Following the PRISMA 2020 guidelines, we systematically searched PubMed and Scopus for peer-reviewed studies (2010-2025) conducted in Africa that applied advanced methods, including Bayesian models, machine learning (ML) algorithms, spatiotemporal analyses, time-series models, and multistate or survival frameworks. Study selection, data extraction, and quality appraisal were independently conducted by multiple reviewers using the Joanna Briggs Institute tools. Findings were synthesized narratively because of substantial methodological heterogeneity.

Twenty-seven studies from nine African countries were included. Bayesian hierarchical, geostatistical, and latent class models improved estimation of TB incidence, mortality, and diagnostic accuracy by accounting for uncertainty, imperfect reference standards, and sparse data. Spatiotemporal analyses consistently identified geographic TB and TB-HIV hotspots linked to low Bacillus Calmette-Guérin vaccine coverage, illiteracy, and urban crowding as risk factors. Time-series models quantified the significant decline in TB notifications during the COVID-19 disruptions. ML approaches, particularly random forests, outperformed traditional regression in predicting latent TB infection and treatment outcomes. Drug resistance, notably to bedaquiline and levofloxacin, showed clear spatial clustering and dynamic progression patterns.

Advanced statistical and computational methods are increasingly improving TB diagnosis, prognosis, and treatment insights in Africa by capturing complex patterns beyond classical models, but their broader impact is limited by data gaps, heterogeneity, and insufficient validation, highlighting the need for high-quality, multi-country research and stronger implementation frameworks.

CRD420251160248.