Welcome to the TB Death Prediction Calculator — Version 2.0

This TB death prediction calculator, developed by ICMR-NIE Chennai, estimates the risk of TB-related death at diagnosis for adults with drug-sensitive TB notified from public health facilities. Version 2.0 is based on data from 21,894 adult TB patients triaged under the Tamil Nadu Kasanoi Erappila Thittam (#TNKET) programme between July and December 2023, using triage variables from the Severe TB Web Application (TB SeWA) and baseline characteristics from NTEP's Ni-kshay.

The calculator provides predicted probabilities of early TB-related death (within 2 months of diagnosis) and overall TB-related death (within 12 months of diagnosis) for an adult (≥15 years) with TB (not known to be drug-resistant at diagnosis), based on variables available at the time of diagnosis. The focus is on predicting death at TB diagnosis itself, as the majority of TB deaths are early, and appropriate action regarding inpatient care for severely ill people with TB needs to be taken at diagnosis.

Four models are available in the calculator. Model 1 uses baseline characteristics routinely captured in Ni-kshay (age, sex, TB site, previous treatment history, HIV status, bank account availability, diagnostic test used, facility type, and district). Model 2 is based solely on the five TN-KET triage variables (BMI, pedal edema, respiratory rate, oxygen saturation, and ability to stand without support), along with facility type and district; its predictive performance is similar to that of Model 1. Model 3 combines the five TN-KET triage variables with Ni-kshay variables that are readily capturable at diagnosis (age, sex, TB site, and previous treatment history) and is the recommended model for routine programmatic use. Model 4 includes all Model 3 variables plus HIV status, bank account availability, and diagnostic test used, and offers the highest predictive accuracy.

The AUC for predicting early TB-related death ranged from 0.715 (Model 1) to 0.783 (Model 4), with the recommended Model 3 achieving 0.767. An appropriate model can be used at TB diagnosis based on the context and availability of variables.