# Development of a clinical algorithm-based scoring system to diagnose smear-negative pulmonary tuberculosis in Sabah, Malaysia using the modified Delphi method

## TL;DR

This study developed a clinical scoring system to help diagnose smear-negative pulmonary tuberculosis in Malaysia using expert input and statistical validation.

## Contribution

A novel clinical algorithm-based scoring system for diagnosing smear-negative TB in resource-limited settings.

## Key findings

- The algorithm achieved an area under the ROC curve of 0.88 with a cut-off score of 19.5.
- It showed 86.2% sensitivity and 77.4% specificity in differentiating likely TB from unlikely TB cases.

## Abstract

Tuberculosis (TB) remains a major global health threat, particularly in resource-constrained settings where delayed diagnosis of smear-negative pulmonary TB (SNPTB) is common due to limited access to rapid molecular diagnostics. We aimed to develop a clinical algorithm-based scoring system to aid the diagnosis of SNPTB among symptomatic patients in Sabah, Malaysia.

We conducted a modified Delphi process between January and June 2024 involving three rounds of expert consultation via email to identify key clinical parameters for diagnosing SNPTB, followed by a consensus meeting to finalise the parameters and assign weightings. We then applied the algorithm to a data set of 60 symptomatic smear-negative individuals, of whom 29 were confirmed to be TB and 31 not TB based on culture. We calculated the sensitivity, specificity, positive predictive value (PPV), and negative predictive values (NPV) of the algorithm to obtain a cut-off score for ‘likely TB’ vs. ‘unlikely TB’.

Of 27 invited experts, 23 (85.2%) consented to participate in the Delphi process and contributed to the final consensus. Fifty-four parameters were identified in round 1, reduced to 26 in round 2 and 23 in round 3. Following the consensus meeting, we incorporated 21 weighted parameters (scores 1–10) into the final algorithm. The clinical algorithm achieved an area under the receiver operating characteristic curve of 0.88. A cut-off score of 19.5 differentiated ‘likely TB’ from ‘unlikely TB’, yielding a sensitivity of 86.2%, specificity of 77.4%, PPV of 78.1%, and NPV of 85.7%.

This diagnostic clinical algorithm could help doctors practicing in resource-constrained settings to diagnose SNPTB. A next step for research would be the prospective validation of the algorithm.