Machine learning analysis identifies key predictors of baseline lung allograft dysfunction and increased mortality in lung transplant patients.

BACKGROUND

Baseline lung allograft dysfunction (BLAD) is a phenotype of early graft dysfunction associated with impaired survival after lung transplantation. Although prior work links BLAD with several pre- and post-transplant factors, a robust set of donor and recipient predictors remains unclear.

RESEARCH QUESTION: What are important donor and recipient predictors of BLAD?

METHODS

We retrospectively reviewed 124 adults who underwent bilateral lung transplantation at a single center. Donor and recipient variables were evaluated using machine learning (gradient boosting and classification tree models), with high-importance predictors incorporated into logistic regression. Survival was assessed using Kaplan-Meier and Cox proportional hazards models.

RESULTS

BLAD occurred in 58 of 124 patients (46.7%). Our analysis identified 7 key factors associated with BLAD: Increased cold ischemia time (p = 0.04) and extremes of Body Mass Index (BMI) (p = 0.04) were associated with BLAD in a non-linear manner. Increased post-transplant Intensive Care Unit (ICU) length of stay (Odds Ratio [OR]: 1.08, 95% Confidence Interval [CI]: 1.00-1.17, p = 0.04) and donor tobacco use > 20 pack years (OR: 16.83, 95% CI: 2.34-121.2, p = 0.005) were also associated with higher odds of BLAD. Female sex (OR: 0.08, 95% CI: 0.02-0.39, p = 0.002), older age (OR: 0.93, 95% CI: 0.88-0.97, p = 0.003), and higher pre-operative Forced Expiratory Volume in 1 second (FEV₁) (OR: 0.19, 95% CI: 0.07-0.55, p = 0.002) were associated with lower odds of BLAD. The model demonstrated strong predictive performance (C-index 0.79, cross-validated). Patients with moderate to severe BLAD (Grade 2 and 3) had a significantly increased risk of death over a median follow-up period of 815 days (Hazard Ratio [HR] 3.53, 95% CI: 1.07-11.71, p = 0.04).

CONCLUSIONS

BLAD is common after bilateral lung transplantation. Moderate to severe BLAD is associated with worse survival. Machine learning identified biological variables associated with BLAD and modifiable peri-operative variables, highlighting opportunities for pre-transplant and intra-operative optimization.