Thoracoscore Risk Bands Substantially Underpredict In-Hospital Mortality After Thoracic Surgery: An External Validation Study in a Tertiary Referral Cohort

Abstract

Background: Thoracoscore is widely used to estimate preoperative mortality after general thoracic surgery, but its performance in Southeast Asian referral practice is uncertain.

Objective: To evaluate the discrimination and calibration of the Thoracoscore risk-band table used routinely at an Indonesian tertiary centre.

Methods: This retrospective cohort included 106 consecutive adults who underwent thoracic surgery at Dr Mohammad Hoesin General Hospital, Palembang, from January to December 2025. Seven of nine Thoracoscore domains were retrievable; therefore, the published points-to-risk band table, rather than the original logistic equation, was validated. Discrimination was assessed by permutation testing, calibration by observed-to-expected (O:E) ratios and within-band tests, and independent predictors by Firth-penalised logistic regression.

Result: In-hospital mortality was 26.4% (28/106; 95% CI 19.0–35.5). Discrimination was absent (AUC 0.572, 95% CI 0.457–0.687; permutation p = 0.249). Calibration failed in every risk band: 28 deaths occurred compared with 2.88 expected (O:E 9.71, 95% CI 6.45–14.03), with a minimum O:E of 4.89 under the most favourable assignment of the two unrecorded domains. Malignancy was the only independent predictor (adjusted OR 4.12, 95% CI 1.51–13.10), but it did not remain significant after multiplicity adjustment (BH p = 0.061).

Conclusion: The published Thoracoscore risk-band probabilities should not be used for individual consent at this centre. Recalibration reduces calibration error, but discrimination remains inadequate for reliable individual prediction.