Background Syndemics (synergistic epidemics) have traditionally been studied using a mixed methods approach in which qualitative analysis precedes quantitative analysis. Typically, scholars first develop hypotheses about interactions between two or more health problems and the social conditions in which they exist via qualitative fieldwork. Subsequently, the synergistic interaction may be validated quantitatively. However, qualitative studies are time-consuming, opportunistic and may offer limited guidance on probabilities between diseases and social conditions. We propose an alternative method to detect potential syndemics at scale using machine learning and big data that could guide subsequent detailed qualitative investigation.

Methods Records of 99,847 cancer patients from the ethically approved Northern Ireland Cancer Registry were linked with 1.1 million diagnostic hospital records which provided information on comorbidities between 2009–2019. We used an association rule mining algorithm, ECLAT, to find most frequent clusters of diagnoses for each cancer site and calculated Synergy Factor (SF) for each cluster. We compared comorbidity patterns of cancer patients from different social groups, defined using the Northern Ireland Multiple Deprivation Index. The analysis is implemented in Python.

Results We detected significant indications of synergies within several disease clusters. Essential hypertension & disorders of lipoprotein metabolism showed synergy among patients with breast cancer (SF=1.2), lymphoma (SF=1.3), melanoma (SF=1.4), leukaemia (SF=1.5), oesophageal cancer (SF=1.4), brain and central nervous system cancer (SF=1.6). Essential hypertension & chronic ischaemic heart disease showed synergy among patients with lymphoma (SF=1.3), head and neck cancer (SF=1.4), melanoma (SF=1.8), leukaemia (SF=1.4). Essential hypertension & diverticular disease of intestine showed synergy among patients with colorectal cancer (SF=1.2), kidney cancer (SF=1.3), gallbladder and other biliary cancer (SF=1.7). Disorders of lipoprotein metabolism & chronic ischaemic heart disease showed synergy among prostate cancer patients (SF=1.2). For liver cancer patients dyads of hypertension & gastritis/duodenitis (SF=1.5), type 2 diabetes mellitus & gastritis/duodenitis (SF=1.7), type 2 diabetes mellitus & disorders of lipoprotein metabolism (SF=1.4) showed synergy. SF varied across deprivation indices.

Conclusion The proposed method provides a proof of principle for detection of potential syndemics relationships at scale using readily available electronic health records. Although qualitative studies are still necessary to understand the interaction between health problems and social conditions, this method enables efforts to identify such work to be targeted more efficiently. Findings, in this case, warrant further research on synergistic relationships between cancer, cardiovascular and metabolic diseases and in areas beyond cancer.