Predicting Risk of Lung Cancer From Medical History.

Background and objectives: Lung cancer causes 130 000 deaths annually in the United States, with treatment costs averaging $150 000 per patient and a 5-year survival rate of 20.5%. Current screening criteria rely on smoking history and age, missing other risk factors. This study aimed to identify clinical risk factors and social determinants of health (SDoH) for enhanced risk assessment using electronic health record (EHR) data.

Methods: We analyzed 410 298 patient records from the All of Us Research Program, including 9375 lung cancer cases identified through SNOMED coding. Using Logistic LASSO regression, we developed predictive models based on diagnoses grouped by body systems and their interactions.

Results: Respiratory, cardiovascular, and immune systems showed three-fold greater association with lung cancer than other systems. Brain metastasis showed the strongest association (odds ratio 5.0, 95% CI: 4.2-5.8). The final model achieved an AUC of 0.82 (95% CI: 0.80-0.84) and 78% sensitivity in validation. Patients with documented social determinants showed 2.5-fold higher risk (95% CI: 2.1-2.9).

Conclusions: EHR-based prediction models effectively identify lung cancer risk using readily available medical history data. These findings support expanding current screening criteria beyond traditional risk factors.