Problem
The relationship between chronic diseases and new cancer incidence remains unclear. We studied the association between existing long-term conditions (LTCs) and lung cancer incidence in a general population sample using different data science methods and explored the validity of a lung cancer incidence prediction model, using information on existing LTCs.
Approach
Data: We used UK Biobank with a sample size of N=502536, aged between 37-73 years. The sample size used for analysis N=488085, after excluding missing values, participants with previous lung cancer and those lost to follow-up. A comprehensive phenotype description was obtained from participants at recruitment; including information on demographics, lifestyle including smoking consumption, family history of lung cancer, previous history of cancer (other than lung), and N=42 health conditions which included chronic diseases and a history of previous lung infections. Record linkage with the cancer registry was used to capture lung cancer incidence. Statistical Models: Five different variable selection methods (Automated likelihood-ratio- F test-based backward selection, Akaike information criterion backward selection, Lasso regression, Elastic net regression and Xtreme gradient boosting) were used to reduce data dimension and select candidate variables from a list of demographic/ lifestyle/health conditions as lung cancer risk factors. Cox's proportional hazards regression was used to build a final lung cancer risk prediction model.
Findings
At the end of seven year median follow-up, lung cancer incidence was 2541 (0.5%) of total sample. In total, N=15 variables including five LTCs were chosen as significant predictors using variable selection methods. Four out of five LTCs were found to have a significant association with higher lung cancer risk-Chronic Obstructive Pulmonary Disease (COPD) (Hazard Ratio-HR 1.92; 95% Confidence intervals-CI 1.67-2.20) , Coronary Heart Disease-(CHD) (HR 1.24; 95% CI 1.10-1.41), Peripheral Vascular Disease-(PVD) (HR 1.83; 95% CI 1.25-2.68), Dementia (HR 6.90; 95% CI 2.59-18.43), while Hypertension did not have a significant association (HR 1.09; 95% CI 0.99-1.18), in the fully adjusted model. In mediation analysis, smoking variables could explain only a small proportion of the observed association between chronic diseases and lung cancer association (15% for CHD, 6.2% for PVD, 1.4% for Dementia).
Consequences
• Four chronic diseases, COPD, CHD, PVD and Dementia, were found to have a significant association with the risk of lung cancer incidence, in addition to ten previously validated demographic and lifestyle risk factors.• Associations were robust in mediation analysis with smoking variables. • Association between vascular diseases and lung cancer incidence needs further research. This association may have a role in lung cancer screening as well as lung cancer symptom assessment in primary care.