A statistical model for tuberculosis diagnosis to guide multi-test strategies and clinical decision-making

Objectives The diagnosis of tuberculosis (TB) remains complex, requiring multiple microbiological tests - microscopy, PCR, and culture - across various clinical specimen types. A central challenge is determining the optimal combination of test, specimen, and strategy for a given patient. We developed a predictive scoring system that integrates patient characteristics (e.g., age and sex) with test performance data to guide personalized, data-driven diagnostic decisions in multi-test settings. Methods We retrospectively analyzed data from 4,179 patients evaluated for respiratory TB between 2008 and 2018. Variables included age, sex, specimen type and diagnostic test results: smear microscopy, Xpert MTB/RIF, in-house real-time PCR for Mycobacterium tuberculosis complex (MTB), and mycobacterial culture. A multivariate logistic regression model was used to estimate test-specific performance between subgroups. These outputs informed a Hidden Markov Model (HMM) to simulate sequential testing strategies and calculate diagnostic probabilities. The entropy analysis quantified the information gained with each successive test. Results Multivariate logistic regression identified that age significantly influences TB prevalence, peaking in the 21–40 age group. Culture showed the highest sensitivity, followed by PCR, and microscopy performed the poorest. HMM-based simulations revealed that sequential testing improves diagnostic yield, but with diminishing returns. Entropy analysis confirmed that most diagnostic uncertainty is resolved by the first few tests, while subsequent tests contribute marginal additional value. Conclusions Mathematical modeling, combining HMM and entropy analysis, offers a framework for optimizing the diagnostic pathways for TB. Our approach enables test selection to be tailored to patient characteristics, potentially improving diagnostic efficiency and reducing unnecessary tests. Future work should integrate broader clinical and epidemiological variables as well as cost-effectiveness analysis to be able to inform diagnostic stewardship strategies for TB and other complex syndromes.