Bayesian Mendelian Randomization Reveals a Protective Effect of Later Age at First Sexual Intercourse Against Erectile Dysfunction

Genetic proxies for later first sex associated with ~37% lower ED odds in mixed-sex cohorts; male-only results lose significance once confounders are stripped out.

Journal: Medicine | Published: 2026-06-12 | Type: Journal Article | PMID: 42299590 Authors: Wu Jiani (First Hospital of Putian City, Fujian, China); Hui Xueming (Fuyao University of Science and Technology, Fuzhou, China) Funding/COI: None declared — two-author team, no external funding

Summary

Wu & Hui (2026) used Mendelian randomization (MR) — a technique that uses genetic variants as proxies for a behavioral exposure — to test whether age at first sexual intercourse (AFS) causally affects ED risk. Their strongest results come from mixed-sex GWAS datasets, which is an awkward foundation for a male-specific condition. When they ran the analysis in male-only cohorts and properly removed instrumental variables (IVs) linked to known confounders, the effect shrank and lost significance (OR = 0.680, P = .064). They recovered significance by relaxing the IV selection criteria and by applying Bayesian methods seeded with priors drawn from the same data — a circular maneuver they acknowledge but don't fully resolve.

Claims

• In mixed-sex cohorts (Groups 1 and 2), later AFS was associated with lower ED risk: OR = 0.626 (95% CI: not reported in abstract, P = 2.73 × 10⁻⁶) and OR = 0.617 (P = 3.56 × 10⁻⁵) via inverse variance weighted MR
• In male-specific cohorts without IV adjustment (Group 3): OR = 0.643 (95% CI: 0.459–0.901, P = .010); heterogeneity was present (Cochran Q P = .011) and MR-PRESSO flagged potential pleiotropy (P = .012)
• After removing IVs associated with confounders including BMI, testosterone, alcohol, depression, and diabetes (Group 4): effect dropped to OR = 0.680 (P = .064), no longer statistically significant
• Relaxing IV selection thresholds (Group 5, 51 SNPs): significance returned at OR = 0.695 (P = .016), with heterogeneity resolved
• Bayesian MR with an informed prior from Group 3 (Group 8): posterior mean θ = −0.358, 95% credible interval (−0.575, −0.136)
• Bayesian MR with an antagonistic prior favoring harm (Group 10): posterior still pulled negative, mean θ = −0.261 (CI: −0.555, 0.022 — upper bound crosses zero)
• Authors note AFS may affect ED indirectly through reduced lifetime sexual partners or improved metabolic profiles

Study Quality

This is a two-sample MR study using GWAS summary statistics, not individual patient data. MR is designed to mitigate confounding and reverse causation by exploiting random genetic variation at conception as a proxy for the exposure, but it only works if the three core assumptions hold: the genetic variants must associate with AFS, must not associate with unmeasured confounders, and must affect ED only through AFS. The pleiotropy detected in Group 3 (MR-PRESSO P = .012) suggests the third assumption is under pressure, which is why Group 4's confounder-adjusted analysis is the most credible male-specific result — and it's the one that falls short of significance.

The Bayesian MR framework (MR-HORSE, introduced by Grant & Burgess 2024) is a legitimate methodological advance for handling weak instruments. However, Groups 7 and 8 use priors derived from the same conventional MR results (Groups 4 and 3 respectively), which the authors correctly label as circular. The genuinely independent Bayesian analyses — Group 9 (uniform prior) and Group 10 (antagonistic prior) — are the cleanest tests, and Group 10's credible interval still barely clips zero at the upper bound.

Red Flags

• The most statistically compelling results (Groups 1 and 2) use mixed-sex GWAS data for a condition that only affects males — the authors flag this themselves, then still feature these results prominently
• The key male-specific analysis with proper confounder control (Group 4) is non-significant; significance is recovered by loosening IV criteria (Group 5), which the authors acknowledge may introduce weak instrument bias
• Bayesian Groups 7 and 8 use priors bootstrapped from the same dataset, creating circularity — informative priors should come from independent sources
• Only European-ancestry GWAS data used; generalizability is unknown
• ED subtypes (psychogenic, vasculogenic, neurogenic) are not distinguished — pooling them may dilute or confound any real signal
• Two authors, no funding: no peer-reviewed replication, no independent dataset validation
• The discussion's suggestion that findings "support public health policies promoting later AFS" is speculative and far outstrips what the data support

Strengths

• MR methodology is an appropriate and sophisticated choice for this question — observational studies of AFS and ED are riddled with confounding
• Ten separate analyses with varied IV selection criteria and prior specifications constitute genuine sensitivity testing
• Group 10's antagonistic prior test is methodologically honest — putting a thumb on the scale against your hypothesis and checking if the data still win
• The authors plainly acknowledge the circularity in their Bayesian priors and the failure of male-specific conventional MR to reach significance after confounder adjustment
• Sensitivity analyses (MR Egger, MR-PRESSO, Cochran Q) are applied consistently

Verdict

This paper is methodologically interesting as an early application of Bayesian MR to a behavioral-ED question, but the headline claim rests on shaky ground. The male-specific analysis — the only one that's conceptually appropriate — loses significance the moment confounders are properly excluded. What's left is either mixed-sex GWAS results (wrong population) or Bayesian analyses that borrow their priors from the same data they're meant to reinforce. The genetic signal for a causal AFS→ED pathway exists but is weak and sensitive to analytical choices. Read it as a proof-of-concept for Bayesian MR in sexual health research, not as evidence that when men first have sex determines their erectile future.