Lung cancer in primary care: Development & validation of a prediction algorithm

Lung cancer (LC) is among the most common and life-threatening cancer worldwide. Early identification in primary care through robust predictive models can enhance early referral and improve outcomes. This cohort study included 3,454,735 patients aged ≥ 30 years from an Italian federated network of general practitioners' (aggregated) data, spanning 2002–2021. We developed and validated a multivariable prediction algorithm for 5-year LC risk. Risk factors included demographics, lifestyle exposures, comorbidities, and symptoms. A Cox proportional hazard model was estimated and performance assessed using pseudo-R 2 , AUC, and calibration metrics. Model overfitting was assessed via bootstrap methodology. Risk thresholds were established with the goal of developing a decision support tool. Incidence rate of LC was 1.5/1000 person-years in development and validation cohorts. Multivariate analysis showed smoking (HR=14.75), COPD (HR=2.3), and advanced age (HR=1.29 for ≥80) as strong predictors as well as female sex (HR=2.46). Obesity (HR=0.74), diabetes (HR=0.91), and family history of LC (HR=0.48) showed inverse associations. The final model achieved pseudo-R 2 of 0.609, AUC of 0.822, and calibration slope of 1.06 (p value 0.65). Bootstrap analysis confirmed model accuracy. The 5-year predicted risk of LC was 0.73%. Based on risk, low (<0.11%), intermediate (0.11–0.89%), and high (≥0.9%) risk groups were identified. This LC risk model is a viable predictive tool for primary care. Though future refinements should sustain inclusion of other risk factors, the tool can support timely referral and resource prioritization. Clinical decision support systems using this algorithm might therefore enhance early LC detection. • A lung cancer risk prediction model was developed and validated using Italian primary care data. • The model achieved good performance in predicting lung cancer with a pseudo-R 2 of 0.609 and AUC of 0.822. • Smoking, COPD, advanced age, and female sex were strong predictors of lung cancer risk. • The model can support timely referral and resource prioritization in primary care settings. • Integration of the model into clinical decision support systems may enhance early lung cancer detection.