Marc Matrana

Table S2. Top selected ICD-10-CM medical diagnoses across overall, gender, race, gender-race, and ADI patient groups.

Figure S4. A hierarchical clustering heatmap of 98 ICD-10-CM codes (RR >2.0) across four demographic groups.

Figure S6. A clustered heatmap of socioeconomic factors' impact on cancer diagnoses in low and high ADI.

Group barplots depicting the deviation from mean RR by gender and race with annotated effect sizes for 20 ICD-10-CM chapters across all patients. Overall, these effect sizes indicate that cancer affects women more than men and that cancer experiences differ between Black individuals and White individuals. These insights are crucial for tailoring cancer research to meet the specific needs of various groups. AA, African Americans.

Demographics of individuals with cancer (%).

Figure S2. Stacked barplots depicting the relative risk of overall cancer and 20 specific cancer types, categorized by 20 ICD-10-CM Chapters.

Supplementary Appendix: Materials and Methods

Cancer is associated with many preexisting health conditions (PHC); however, accurately quantifying these links remains challenging. Although some studies have examined these associations, large-scale analyses using diverse electronic health record (EHR) data remain limited and lack the ability to evaluate cancer risk when patients are stratified by interpersonal differences. Using a real-world EHR dataset from a large Louisiana health system comprising 8,283,236 records from 1,460,738 patients (2013–2022), we evaluated associations between PHCs and subsequent cancer diagnoses within a fixed 5-year risk window. We applied epidemiologic, statistical, and artificial intelligence methods to the full dataset and to subgroups stratified by gender, race, and area deprivation index (ADI) for overall cancer and 20 cancer types. We identified 9 chapters of International Classification of Diseases, Tenth Revision, including chapter 4 (metabolic) and chapter 14 (genitourinary), with 221 PHCs linked to increased cancer risk (relative risk >1, 95% confidence interval excluding 1, and Benjamini–Hochberg false discovery rate–adjusted P < 0.05). Key PHCs include systemic sclerosis, blood type, benign mammary dysplasia, immune mechanism disorders, disturbances of smell, lipoprotein metabolism disorders, human immunodeficiency virus disease, vitamin D deficiency, and diabetes. Chapter 12 (skin diseases) and chapter 9 (circulatory diseases) showed strong associations with 10 and 13 cancer types, respectively. Age-, gender-, race-, and ADI-specific high-risk PHCs were also identified. However, these findings should be interpreted carefully as ADI may not fully capture individual-level socioeconomic or environmental exposures, and the lack of tobacco data may introduce residual confounding.Significance:This framework offers a systematic approach to linking PHCs with cancer risk, providing valuable insights for cancer prediction, management, and prevention across diverse patient populations. This approach reveals prediagnostic disease patterns and demographic heterogeneity in cancer incidence.

A hierarchical clustering heatmap of the top 76 ICD-10-CM codes (RR >1 in >3 subtype cancers) grouped by 20 cancer types. Darker red indicates higher risk. Lung, prostate, breast, and skin cancers emerge as key early cancer indicators, strongly associated with multiple conditions.

Table S1. Standard and binomial relative risks of ICD-10-CM medical diagnosis chapters across all patients.