Performance of clinicians in identifying tuberculosis as cause of death using verbal autopsy questionnaires in Siaya County, Kenya

The burden of diseases related to mortality is often derived from verbal autopsy (VA) interviews. This study assessed performance of clinicians in identifying tuberculosis among deceased persons aged above one year, who died of respiratory related symptoms in a rural Kenyan setting with a high TB/HIV prevalence. The study was done within a Demographic Surveillance Site in Siaya County in 2012- 2013. Post mortems (PM) were done at a public referral hospital. Culture and GeneXpert tests complemented the PM results. VA questionnaires were reviewed by independent clinicians. Level of agreement among the clinicians in identifying TB deaths as well as their performance against PM results was calculated using Kappa statistics. Sensitivity and specificity tests were measured. Among 156 eligible deaths, 78 (50%) were enrolled and underwent PM. The PM tests detected 11 TB deaths. Among these 11 deaths 3 (27%) (Sensitivity: 27% [95% CI 10%, 57%]) had concordant results with the clinical VA interpretation. The level of agreement among clinicians regarding the presence of TB with an unweighted Cohen’s Kappa was 0.56 (95%CI 0.34 - 0.78). Among the 65 patients without TB by PM, clinicians identified 8 (specificity: 72.3% [95%CI 65%, 86%]). The results indicated a moderate level of agreement among the clinicians regarding the presence of TB at death. The level of agreement between clinician and PM results was significantly low. Use of the current WHO VA questionnaires by clinicians to identify TB deaths is not robust enough to provide accurate estimates of TB mortality. Although the VA may be reliable, it is not valid for diagnosis of TB in a setting with high HIV prevalence.