Leveraging electronic medical records reveals comorbidities significantly associated with male infertility

Male infertility (MI) accounts for at least 30% of infertility etiology for infertile couples. While there are a few known MI risk factors (RFs) such as advanced age, the full breadth of potential MI RFs have not been explored. Here, we use electronic medical records (EMRs) across 5 medical centers to implement a data-driven case-control study identifying MI-associated comorbidities to uncover conditions that may be explored as potential MI RFs in future studies. We identified MI patients by selecting for patients with at least 1 MI-related diagnosis. Age- and location- matched control patients were identified using 1:2 propensity score matching for our primary analysis. We also performed 4 sensitivity analyses by identifying different matched controls that were also matched on race and ethnicity (R&E), area deprivation index (ADI), and / or hospital utilization, specifically number of visits and years in the EMR. All diagnoses, mapped to Phecode-corresponding phenotypes (phewascatalog.org), were compared between MI and control patients, with significantly different diagnoses identified using chi-squared tests (or Fisher’s exact tests for very low counts) with a Bonferroni-corrected p < 0.05. We identified 6,691 MI patients (age: mean 43.90, SD 8.77; number of visits: mean 43.60, SD 76.96; number of years in the EMR: mean 3.80, SD 3.46) and 13,382 matched controls (age: mean 43.90, SD 8.77; number of visits: mean 20.21, SD 40.96; number of years in the EMR: mean 2.22, SD 2.95) in the primary analysis. We compared 1,689 phenotypes in the primary analysis, with 282 phenotypes significant for MI patients, including expected conditions like testicular hypofunction (OR = 21.5, p < 0.001), varicose veins (OR = 20.8, p < 0.001), and obesity (OR = 1.7, p < 0.001), as well as less established associations like disorders of iron metabolism (OR = 4.7, p-value < 0.001), disorders involving the immune mechanism (OR = 2.1, p < 0.001), irritable bowel syndrome (OR = 2.5, p = 1.0e-14), and hyperlipidemia (OR = 1.8, p < 0.001). The primary analysis, the analysis matching on R&E, and the analysis matching on R&E and ADI, had the most overlapping phenotypes (n = 197). All 5 analyses had 33 phenotypes in common, most of which are expected, such as erectile dysfunction and other disorders of male genital organs, although circulatory disease and lipoma were among the less established associations found. Leveraging EMR data, we identified previously known and potentially novel MI-associated comorbidities. Matching on hospital utilization filtered for phenotypes most strongly associated with MI, which may be due to matching with less healthy control patients. MI patients tend to utilize healthcare more relative to controls when not matching on hospital utilization. Future directions include exploring potential differences in medications to assess whether any could be potential RFs for MI, validating our findings in other medical centers, and assessing the temporal relationship between these phenotypes and MI.