In our study, we found both expected and less expected comorbidities that were positively associated with MI at UC and Stanford.

First, we explored whether patients’ overall diagnosis profiles differed based on MI status. We also explored whether they differed based on demographic features, location of care, and hospital utilization to determine whether to account for these covariates in our analyses. We found significant differences in patients’ diagnoses profiles based on MI status, age, self- or provider-identified race, self- or provider-identified ethnicity, number of visits, and months in the EHR across both hospital systems. Interestingly, there were fewer significant differences between patients’ diagnoses profiles based on MI status for diagnoses first obtained before or after the 6-month cutoff at Stanford relative to UC. For Stanford, only the first UMAP component significantly differed for patients’ diagnosis profiles for conditions first obtained before the 6-month cutoff, while no significant differences were observed in patients’ diagnosis profiles for conditions first obtained after the 6-month cutoff.

We found 76 diagnoses at UC and 19 diagnoses at Stanford that were positively associated with MI before the 6-month cutoff, including 15 diagnoses that overlapped across both hospital systems. That we found relatively fewer diagnoses significantly associated with MI at Stanford is consistent with our low-dimensional embedding visualizations, which revealed fewer significant differences in patients’ diagnosis profiles based on MI status at Stanford.

A number of comorbidities found to be significantly associated with MI before the 6-month cutoff at UC and Stanford were expected, such as abnormal spermatozoa, testicular hypofunction, and varicose veins. However, we also found a number of less expected comorbidities, such as anemias, hypothyroidism, and prostate cancer.

Prior studies have explored the relationship between male infertility with each of these less expected comorbidities to varying extents. The research on anemia generally is limited, although it is known that sickle cell disease, which can lead to primary and secondary testicular failure, can infer a higher risk for male infertility25,26,27. Importantly, the current report did not identify sickle cell disease as the specific etiology. For hypothyroidism, prior research has been primarily limited to rodent studies or human studies with a small number of participants28,29,30,31,32,33,34,35. While thyroid dysfunction is a common cause of female infertility, the evaluation of the infertile male does not include such screening36. Finally, studies exploring the relationship between prostate cancer and male infertility often measure the risk of developing prostate cancer after receiving a male infertility diagnosis37,38,39,40,41. This is distinct from our finding, which suggests that patients with prostate cancer are receiving this diagnosis prior to or concomitant with their male infertility diagnosis, suggesting that it is possible that these patients are seeking treatment for their infertility, which could be the result of cancer treatment42. However, to assess whether the association between prostate cancer and the development of MI could indeed be due to cancer treatment, confounders such as surgery, radiation, and androgen deprivation would need to be controlled for and hopefully will be explored in future studies. The average age of prostate cancer diagnosis occurs in the seventh decade of life, an age beyond which reproduction is usually desired or common. Nevertheless, the age of paternity is increasing, and other groups have reported that reproduction should be increasingly considered among prostate cancer patients43,44.

We also found several circulatory system diagnoses that were positively associated with MI before the 6-month cutoff in the UC analysis, including circulatory disease, congestive heart failure, and hypertensive complications, with one study suggesting an association between hypertension and impaired semen quality45.

Our results suggest that it may be beneficial to perform larger, longitudinal studies to clarify the relationship between MI and these comorbidities, which are all treatable conditions. Treatment for hypothyroidism and anemia in particular may be promising for restoring male fertility.

Only 13 and 3 diagnoses were positively associated with MI after the 6-month cutoff (with no fixed follow-up time) in the UC and Stanford analyses, respectively. Abnormal spermatozoa and testicular hypofunction were the 2 diagnoses that overlapped between these hospital systems and were expected.

In the UC analysis, we found that arrhythmia, circulatory disease, and other forms of chronic heart disease were positively associated with MI after the 6-month cutoff. That we found circulatory system conditions is consistent with several longitudinal and cross-sectional studies that suggest an increased incidence of or association with circulatory system conditions for patients with MI7,11,46,47,48. Our study is consistent with these findings and supports testing the hypothesis that MI may be a marker of an elevated risk of developing circulatory system diseases.

Other diseases of the kidney and ureters were also positively associated with MI after the 6-month cutoff in the UC analysis. To our knowledge, it is unknown whether MI leads to an increased incidence of renal-related conditions. One possible interpretation of our finding is that these patients may have had subclinical renal conditions that further developed sometime after their MI diagnosis. Future studies can be done to ascertain whether there is an increased incidence of renal conditions for patients with MI. Additionally, future studies can measure lab values that indicate kidney function before and after a MI diagnosis in patients who then develop a renal-related condition to ascertain whether kidney function may have been compromised before being diagnosed with MI.

Other potentially interesting diagnoses positively associated with MI in the UC analysis after the 6-month cutoff include H. pylori infection and osteoporosis; osteoporosis was also significantly associated with MI before the 6-month cutoff in UC. The few studies investigating the relationship of MI with these and other related conditions are inconclusive and conflicting49,50,51,52,53,54,55. Our findings indicate that larger studies clarifying the relationship between MI and these conditions may be beneficial.

Notably, Cox proportional hazards models revealed that MI patients had a higher risk of receiving 11 of the 13 diagnoses found to be significantly associated with MI after the 6-month cutoff at UC relative to patients with a vasectomy-related record. This demonstrates that we can use findings from the logistic regression association analyses in longitudinal studies to further clarify the relationship between MI and associated comorbidities.

When we compared the effect sizes of diagnoses from the primary analysis before and after the 6-month cutoff, we found that they were significantly correlated in both hospital systems, suggesting that there are a number of conditions that may be associated with MI regardless of cutoff time. Interestingly, however, disease category representation may depend on when they were first diagnosed relative to MI. Genitourinary diseases, including renal conditions, tend to be more associated with MI before the 6-month cutoff, while circulatory system diseases tend to be more associated with MI after the 6-month cutoff. Moreover, endocrine/metabolic conditions seem to be associated with MI regardless of cutoff time. This suggests that it is possible that different disease categories may be more likely to be risk factors for or adverse health outcomes of MI. Future hypothesis-driven, longitudinal studies utilizing EHR could clarify the relationship between the manifestation of certain conditions and the development of MI. For example, a retrospective cohort study can test whether patients with renal conditions are more likely to develop MI.

We also compared the effect sizes of diagnoses significantly associated with MI in the primary analysis at UC and/or Stanford. We found that they were significantly correlated before and after the 6-month cutoff, with the correlation being higher before the 6-month cutoff. That we found a higher correlation before the 6-month cutoff may be consistent with our temporally-stratified association analysis findings and low-dimensional embedding visualizations, where we found more diagnoses significantly associated with MI at Stanford before the 6-month cutoff as well as more differences in patients’ diagnosis profiles based on MI status before the 6-month cutoff.

There are several limitations to consider in our study. First, the logistic regression study design and its lack of accounting for time limits the interpretation of findings, in particular the directionality of associations. Thus, longitudinal analyses are required to assess directionality. To address this, we used Cox proportional hazards models to further assess the relationship between male infertility and the 13 diagnoses found to be associated with male infertility for UC patients after the 6-month cutoff. Second, some of the differences observed in the UMAP of diagnosis profiles between MI patients (cases) and vasectomy patients (controls) might be due to different screening practices for MI diagnoses and vasectomy procedures. Evaluations for both MI and vasectomy, however, mandate a thorough acquisition of medical history as well as a physical examination to evaluate for comorbidities. While some conditions may be related to a specific screening practice, such as identifying varicoceles during an MI evaluation, many others, such as hypothyroidism, would be unlikely to be diagnosed during an MI evaluation. Third, there could be differences between MI patients and vasectomy patients that are causally associated with the observed outcomes but are not due to MI. While our sensitivity analyses try to account for this, they likely do not account for all potential bias, such as potential differences in the health-seeking behaviors of MI patients and vasectomy patients. Moreover, there is inherent uncertainty about when patients first developed MI; we only know when patients were first diagnosed with the condition at UC or Stanford. Generally, we cannot know for certain when patients first developed any particular medical condition using EHR data. We are also unable to tell whether they were first diagnosed outside of UC or Stanford due to the legal limitations placed on systematically sharing personal health information across institutions; this potential missingness is an inherent limitation of studies utilizing EHR data. As a result, we cannot know for certain whether the comorbidities we identified manifested before, concurrent with, or after the development of MI.

Another limitation is that we did not perform a power analysis; thus, some of our findings may be false negatives. We also did not estimate false positives from correlated diagnoses. Thus, some of our findings may be false positives, which we attempted to mitigate by identifying significant male infertility associated diagnoses not only in the primary analysis, but also in the SDoH and hospital utilization sensitivity analyses across both UC and Stanford.

Additionally, we selected patients with any vasectomy record, including those who underwent vasectomy reversal or who had follow-up tests after their vasectomy. Thus, for some vasectomy patients, some diagnoses before the 6-month cutoff may have occurred 6 months after their vasectomy. Even with this caveat, the vast majority of UC patients in our study (> 98%) have either a known vasectomy date or recorded post-vasectomy semen analysis, which usually occurs around 3 months after a vasectomy56. Moreover, vasectomy patients were chosen for their presumptive fertility, as previous studies have shown that 90% of patients seeking a vasectomy are fertile12. Thus, that the precise date of vasectomy may be unknown may not have a meaningful impact on our findings.

Many UC associations were not significant in the Stanford analysis, which was particularly the case after the 6-month cutoff. This could be due to several reasons. First, there were far fewer vasectomy patients in the Stanford analysis, which likely reduced statistical power for the association analyses. Additionally, there may be multiple differences in the overall patient population at UC and Stanford, which may limit how many significant findings overlap between the two hospital systems. It is also possible that the extract, transform, and load (ETL) process underlying the transfer of patient records from the hospitals’ EHR to de-identified or limited data set databases systematically differ between the two hospital systems, which may lead to differences in diagnosis representation. Nonetheless, it is promising that the effect sizes of overlapping diagnoses were significantly correlated between UC and Stanford, even if the diagnoses themselves were not necessarily significantly associated in both institutions. Future studies can perform association analyses with larger numbers of case and control patients to explore whether similar comorbidities are significantly associated with MI.

Another important consideration is potentially reduced sensitivity for identifying MI patients in the EHR. The specific codes used for patients may vary by provider and the vagaries of insurance coverage. Thus, administrative differences amongst physicians may affect how patients’ conditions are represented in the EHR. This is indeed an inherent limitation for any EHR study that uses administrative/insurance claims to study disease57.

Moreover, the sensitivity and specificity of our criteria for male infertility and vasectomy are inherently incalculable in the UC and Stanford EHR due to a lack of accessible ground truth. That is, we do not know how many patients truly have male infertility or have had a vasectomy. In general, having broad inclusion criteria would maximize sensitivity while potentially compromising specificity, while more stringent inclusion criteria generally will increase specificity while decreasing sensitivity. For example, a 2017 review demonstrated that of the patients who have undergone a semen analysis and have been diagnosed with at least one of five male infertility ICD-9 codes (606.0, 606.1, 606.8, 606.9, and V26.21), 92.4% were determined to have abnormal semen quality58. If the patients had at least three distinct infertility codes, this rises to 99.8%58.

We also found that adjusting for hospital utilization substantially affected the number of significant comorbidities positively associated with MI, particularly for diagnoses first obtained after the 6-month cutoff. This may be due to increased healthcare utilization for patients with MI relative to post-vasectomy patients, leading to more opportunities to be diagnosed with a particular condition. This underscores the importance of accounting for potential differences in healthcare utilization when performing EHR-based studies.

Another limitation is that our patient cohort may not be representative of the overall population experiencing MI. Male infertility counseling is done relatively less often compared to female infertility counseling, which may in part be due to social stigma59. This suggests that there may be patients experiencing MI who are not seeking infertility care. Moreover, it is possible that those seeking infertility treatment may be disproportionately more likely to afford expensive fertility treatments relative to the general population. Finally, there are likely individuals who are unaware they have MI, especially if they are not pursuing family planning. Thus, our study of MI may have limited generalizability.

There have also been studies suggesting that environmental exposures may play a role in MI and declining sperm count60. However, environmental exposures are not systematically captured in either of the EHR databases utilized in this study. Future work can ascertain the relationship between MI, comorbidities, and environmental exposures by taking into account where patients live.

Overall, our data-driven approach revealed expected and less expected comorbidities associated with MI across UC and Stanford. Our temporally-stratified association analyses point to conditions that we hypothesize may be risk factors for or adverse outcomes of MI. Our findings can serve as a useful starting point for clarifying the relationship between these conditions and MI in future longitudinal studies. Ultimately, we hope that this study will help uncover new etiologies underlying currently idiopathic MI as well as actionable insights for prevention and treatment.