Abstract

LOGISTIC MODELING OF TUBERCULOSIS TREATMENT OUTCOMES FROM NON-DISJOINT EVENTS

Agada, P. O.,1* Egahi, M.,2 Igbabul, S. A.3

---

In Multinomial Logistic Regression, result interpretation can be difficult if an event that gives rise to at least one of the treatment outcomes is a merger of outcomes of two other non-disjoint events. This is obvious as the non-disjoint nature of the underlying events would not yield distinct and independent outcomes of the dependent variable. This work proposes a way of handling challenges of this nature at the primary level of data collection and when the study data is secondary with dependent variable outcomes overlaps. This work demonstrates how to remove these outcomes overlaps for clearer result interpretation using a dataset from a published research. The aforementioned research is on the evaluation of tuberculosis treatment outcome of TB/HIV Co- Infection. Our study made two assertions for removing outcome overlaps leading to the fitting of two Binary Logistic Regression Models one for each. Major results of the study include the fact that, if all successfully treated TB patients are cured or not, HIV status remains a predictor of TB treatment outcome with decrease in the odds for patients who test positive to HIV relative to those who test negative. Additional result shows that if all the successfully treated TB patients are cured, then, sputum test result for TB at baseline and the sex of patient become immaterial in predicting TB treatment outcomes. Rather, the fact that patients are on treatment support and on ART become significant predictors, with odd ratios of 1.644 and 0.759 respectively. Furthermore, if all the successfully treated TB patients resisted treatment (not cured), the predictors of TB treatment outcome excludes the fact that patients are on treatment support and on ART but include their sex and sputum test result at baseline, with odd ratios of 0.645 and 54.938 respectively.

PDF