Development and Internal Validation of a Clinical-Epidemiological Model for Active Pulmonary Tuberculosis Among Patients Already Evaluated for Suspected Disease in a Low-Incidence Hospital.

BACKGROUND

In low-incidence hospital settings, airborne isolation capacity is limited. We developed a clinical-epidemiological model to estimate active pulmonary tuberculosis (PTB) probability among adults already selected for PTB evaluation, using information available before definitive diagnostic testing.

METHODS

We retrospectively analyzed consecutive adults who underwent microbiological testing for suspected PTB and were managed with airborne precautions at a university hospital in Rome from January 2016 through December 2021 (n=682 complete cases). A LASSO-penalized logistic regression model was developed. Discrimination was assessed by 10-fold cross-validation; calibration and decision-curve analysis were also evaluated. Probability thresholds were explored to describe low-, intermediate-, and high-risk strata; they were not prespecified clinical safety thresholds.

RESULTS

Ten predictors were retained. The cross-validated AUC was 0.78. At the lower threshold (0.1247), sensitivity was 98.4% and 48/682 patients (7.0%) were classified as low risk; 4/48 had active PTB (observed residual risk 8.3%). At the upper threshold (0.834), specificity was 99.5%; 17/682 (2.5%) were classified as high risk, including 15 PTB cases. Most patients (617/682; 90.5%) were intermediate risk.

CONCLUSIONS

The model showed moderate discrimination during internal validation and identified small groups at the extremes of predicted risk within a preselected suspected-PTB cohort. These findings do not establish that airborne precautions can safely be withheld or discontinued. Independent external validation, clinically justified decision thresholds, and prospective impact and safety assessment are required before clinical implementation.