Abstract 2202: Association of multi-morbidity with incident cancer diagnoses

Abstract Background: Multi-morbidity (MM), which is usually defined as a simple count of the number of different diseases an individual has, has previously been identified as a risk factor for some incident cancer diagnoses. This study aimed to use more sophisticated machine learning techniques to classify MM into disease clusters and to investigate their association with incident colorectal (CRC), breast (BC), lung (LC) and prostate (PrC) cancer diagnoses. Methods: Participants in UK Biobank with a hospital (HES) record and no prevalent cancer were included in the study. Self-reported (SR), HES and primary-care (GP) recorded diseases were converted to phecodes. Sex-stratified latent class analyses (LCA) was used to identify disease clusters in the SR data. Associations between clusters and selected cancer diagnoses was undertaken using Cox proportional hazards modelling, adjusting for age, ethnicity, socio-economic status, year and region of recruitment, body mass index, smoking history, alcohol consumption and menopausal hormone therapy use (women only). The analyses were validated using HES and GP data. Results: The analysis included 158, 951 women and 127, 406 men. Just over 30% (49, 061) of women and 37% (48, 110) of men had MM (≥2 diseases) at recruitment. LCA identified 10 disease clusters in women, the most common of which was related to respiratory disease, largely comprising asthma (9344; 19. 1%). Nine disease clusters were identified in men, the most common of which was related to heart disease (9358; 19. 5%), and largely comprising ischemic heart disease and hypertension. No association between disease cluster and incident CRC or BC was identified. In women, LC risk was associated with several disease clusters: respiratory (Hazard Ratio 1. 57 (95% confidence interval 1. 28 - 1. 91) p0. 001), arthritis/gut (1. 46 (1. 15 - 1. 85) p=0. 002), hyperlipidemia (1. 51 (1. 19 - 1. 92) p=0. 001, cardiac (1. 74 (1. 40 - 2. 17) p0. 001) and type 2 diabetes (T2DM) (1. 56 (1. 16 - 2. 09) p=0. 003). In men, LC was associated with disease clusters of hyperlipidemia/hypertension (1. 23 (1. 02 - 1. 48) p=0. 03), cardiac (1. 40 (1. 18 - 1. 66) p0. 001), T2DM (1. 26 (1. 02 - 1. 56) p=0. 03), respiratory (2. 74 (2. 21 - 3. 39) p0. 001) and cardiometabolic/arthritis (1. 81 (1. 08 - 3. 03) p=0. 02). T2DM (0. 80 (0. 68 - 0. 87) p0. 001) was associated with an inverse risk of PrC. Cluster distribution was similar in HES and GP data, with the association between the cardiac and the arthritis clusters replicated for LC in women, and the respiratory cluster for LC and T2DM cluster for PrC in men. Conclusion: Clusters of MM were identified in a large cancer-free population, which were replicated using different sources of healthcare data. Several disease clusters were associated with incident LC and, to a lesser extent, PrC. Further work to understand the potential causal nature of these associations is needed. Citation Format: Megan C. Conroy, Gillian K. Reeves, Naomi E. Allen. Association of multi-morbidity with incident cancer diagnoses abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 2202.