The effect of tuberculosis on mortality in HIV positive people: a meta-analysis

Abstract

Background: Tuberculosis is a leading cause of death in people living with HIV (PLWH). We conducted a meta analysis to assess the effect of tuberculosis on mortality in people living with HIV.

Methods: Meta-analysis of cohort studies assessing the effect of tuberculosis on mortality in PLWH. To identify eligible studies we systematically searched electronic databases (until December 2008), performed manual searches of citations from relevant articles, and reviewed conference proceedings. Multivariate hazard ratios (HR) of mortality in PLWH with and without tuberculosis, estimated in individual cohort studies, were pooled using random effect weighting according to "Der Simonian Laird method" if the p-value of the heterogeneity test was <0.05.

Results: Fifteen cohort studies were systematically retrieved. Pooled overall analysis of these 15 studies estimating the effect of tuberculosis on mortality in PLWH showed a Hazard Ratio (HR) of 1.8 (95% confidence interval (CI): 1.4-2.3). Subanalysis of 8 studies in which the cohort was not exposed to highly active antiretroviral therapy (HAART) showed an HR of 2.6 (95% CI: 1.8-3.6). Subanalysis of 6 studies showed that tuberculosis did not show an effect on mortality in PLWH exposed to HAART: HR 1.1 (95% CI: 0.9-1.3).

Conclusion: These results provide an indication of the magnitude of benefit to an individual that could have been expected if tuberculosis had been prevented. It emphasizes the need for additional studies assessing the effect of preventing tuberculosis or early diagnosis and treatment of tuberculosis in PLWH on reducing mortality. Furthermore, the results of the subgroup analyses in cohorts largely exposed to HAART provide additional support to WHO's revised guidelines, which include promoting the initiation of HAART for PLWH co-infected with tuberculosis. The causal effect of tuberculosis on mortality in PLWH exposed to HAART needs to be further evaluated once the results of more cohort studies become available.

Figure 1. Flow diagram of papers accepted and rejected during selection procedure.

Figure 2. Funnelplot of 15 studies estimating the effect of tuberculosis on either all cause mortality or AIDS-related mortality in people living with HIV.

Points indicate log hazard ratios (x-axis) from 15 studies (contributing to 18 separate hazard ratio's) assessing the effect of “prevalent”, “incident”, and “prevalent+incident” TB on AIDS-related/all cause mortality in HIV positive individuals. Precision is calculated as 1/standard error. A lower precision indicates a more accurate estimate.

Figure 3. Studies assessing the effect of “prevalent”, “incident” and “prevalent”+“incident” TB on AIDS-related/all cause mortality in HIV positive individuals.

Study ID on the Y-axis refers to first author and publication year; % weight refers to influence of each study on overall estimate (weights are from random effect analyses); for each study the central diamond indicates multivariate hazard ratio, line represents 95% confidence interval (CI), and the grey square reflects the study's weight in the pooling; overall estimate refers to pooled estimate of hazard ratio after mathematical combination of all studies; the X-axis indicates the scale and the direction of the effect of tuberculosis on mortality in HIV positive individuals. I-squared denotes the extent of heterogeneity in study outcomes, with a (hypothetical) value of 100% meaning considerable heterogeneity and 0% meaning no heterogeneity between studies.

Figure 4. Studies assessing the effect of “prevalent”, “incident” and “prevalent”+“incident” TB on all cause mortality in HIV positive individuals with baseline CD4 cell count ≤200 cells/µL.

Study ID on the Y-axis refers to first author and publication year; % weight refers to influence of each study on overall estimate (weights are from random effect analyses); for each study the central diamond indicates multivariate hazard ratio, line represents 95% confidence interval (CI), and the grey square reflects the study's weight in the pooling; overall estimate refers to pooled estimate of hazard ratio after mathematical combination of all studies; the X-axis indicates the scale and the direction of the effect of tuberculosis on mortality in HIV positive individuals.

Figure 5. Studies assessing the effect of “prevalent”, “incident” and “prevalent”+“incident” TB on all cause mortality in HIV positive individuals before HAART era (<1996) or ≤10% of cohort has reported use of HAART.

Study ID on the Y-axis refers to first author and publication year; % weight refers to influence of each study on overall estimate (weights are from fixed effect analyses); for each study the central diamond indicates multivariate hazard ratio, line represents 95% confidence interval (CI), and the grey square reflects the study's weight in the pooling; overall estimate refers to pooled estimate of hazard ratio after mathematical combination of all studies; the X-axis indicates the scale and the direction of the effect of tuberculosis on mortality in HIV positive individuals.

Figure 6. Studies assessing the effect of “prevalent”, “incident” and “prevalent”+“incident” TB on all cause mortality in HIV positive individuals during HAART era and ≥50% of cohort has reported use of HAART.

Study ID on the Y-axis refers to first author and publication year; % weight refers to influence of each study on overall estimate (weights are from fixed effect analyses); for each study the central diamond indicates multivariate hazard ratio, line represents 95% confidence interval (CI), and the grey square reflects the study's weight in the pooling; overall estimate refers to pooled estimate of hazard ratio after mathematical combination of all studies; the X-axis indicates the scale and the direction of the effect of tuberculosis on mortality in HIV positive individuals.