The finding
In 1992, 31,925 men in the Health Professionals Follow-up Study reported their average monthly ejaculation frequency. They were followed to 2010. Across 480,831 person-years, 3,839 prostate cancers were diagnosed. Men who reported 21 or more ejaculations per month had a lower incidence of prostate cancer than men who reported 4 to 7: a multivariable hazard ratio of 0.81 (95% CI 0.72 to 0.92) for frequency recalled at ages 20-29, and 0.78 (95% CI 0.69 to 0.89) for ages 40-49. Both trends were significant at p<0.0001.
That is the largest prospective observation anyone has made in this area, and it runs against the most common health claim made for semen retention. The caveats below are real, and none of them makes the finding go away.
Retaining semen is good for your prostate.
The largest prospective dataset points the other way. In 31,925 men followed for 480,831 person-years, higher reported ejaculation frequency was associated with lower prostate cancer incidence (Rider 2016). No study has ever measured a group of men practising deliberate retention, so this is not proof that retention causes harm. It does mean the one large, repeatedly analysed observation in this area runs opposite to the claim that holding it in benefits the prostate.
Rider 2016, multivariable adjusted. The 4 to 7 category is the reference and sits at 1.0 by definition. 95% CI 0.72 to 0.92 and 0.69 to 0.89. One observational cohort of US male health professionals; exposure was self-reported recall.
See the numbers
| 4-7 a month (reference) | 1hazard ratio |
|---|---|
| 21+ a month, recalled at ages 20-29 | 0.81hazard ratio |
| 21+ a month, recalled at ages 40-49 | 0.78hazard ratio |
What the observational design costs
Three limits matter, and they are the ones a careful reader should press on.
- Self-report and recall. Men were asked to remember how often they ejaculated decades earlier. That is a soft measurement of a private behaviour, and errors in it are not necessarily random.
- Who the cohort is. US male health professionals. Not demographically representative, and a group whose sexual frequency travels with partnership status, general health, medication and mood, none of which can be fully adjusted away.
- Detection bias. The obvious worry is that men with different sexual frequencies get screened at different rates, so what looks like a difference in disease could be a difference in who gets tested.
The third one is what retention forums usually reach for, and it is the one the authors actually tested. PSA testing rates and biopsy frequency were similar across frequency categories, and the association held when the analysis was restricted to a PSA-screened subgroup. That is a genuine methodological strength, and it is why the fair criticism of this study is that it is observational and about frequency, not that it is weak.
No randomised trial of ejaculation frequency exists, and none could plausibly be run. So the confounding cannot be resolved. It can only be probed, and here it was probed more carefully than most.
That study is just an artefact of PSA screening. More sex means more doctor visits means more diagnoses.
The authors anticipated this and tested it: PSA testing rates and biopsy frequency were comparable across ejaculation-frequency categories, and the inverse association persisted within the PSA-screened subgroup (Rider 2016, 31,925 men). The screening objection is reasonable in principle and does not survive contact with this particular paper.
The replication that is not one
A 2025 systematic review and dose-response meta-analysis pooled 29 studies and 315,193 participants and found higher ejaculation frequency associated with lower odds of prostate cancer, at a pooled odds ratio of 0.83 (95% CI 0.77 to 0.90). That number gets passed around as independent confirmation. It is not. The pool includes the Health Professionals Follow-up Study analyses themselves, so it partly contains the very cohort it appears to be checking.
A meta-analysis that contains the cohort it is checking is not a second opinion.
The same analysis produced a result that complicates the picture and rarely gets quoted alongside the headline: frequency of intercourse (OR 1.02), number of female partners (OR 1.01) and masturbation frequency (OR 1.12) were all non-significant. The mechanism usually offered for a protective effect, sometimes called prostatic stagnation, has never been demonstrated, and if clearing the prostate were what mattered, masturbation frequency ought to behave like ejaculation frequency. In the largest pooled analysis available it did not. That discrepancy is an open question in the literature, not a settled point in either direction.
A separate meta-analysis has independently confirmed the ejaculation-frequency finding.
Raeisvandi 2025 pooled 29 studies and 315,193 participants for an OR of 0.83, but its pool includes the Health Professionals Follow-up Study cohort analyses, so it is not independent replication of them. It also pools case-control with cohort designs, which bakes recall bias into one arm. It is a useful summary of the literature. It is not a second, separate test.
Where the evidence disagrees with itself
The precursor analysis of the same cohort, published in 2004 with 8 years of follow-up, covered 29,342 men and 222,426 person-years. It found a relative risk of 0.68 (95% CI 0.53 to 0.86) for 21 or more ejaculations per month versus 4 to 7 at ages 40-49, and 0.67 (95% CI 0.51 to 0.89) averaged across a lifetime. Most individual frequency categories were unrelated to risk. Because it analyses the same men earlier, it is not an independent replication of the 2016 result. For advanced prostate cancer there was no statistically significant association, but there were only 147 advanced cases in the whole study. That is a question the analysis was not powered to answer, not an answer it gave.
The case-control literature disagrees with the cohorts, and it disagrees in an interesting place. A UK study of 431 cases and 409 controls, all diagnosed at age 60 or younger, found the opposite direction in early adulthood: sexual activity in the twenties, and masturbation in the twenties and thirties, were associated with higher risk, with little effect from activity in the forties and possible protection only in the fifties. The differences were attributed mainly to masturbation rather than intercourse.
That design asks men who already have a cancer diagnosis to recall their sexual frequency decade by decade, which is precisely the recall bias the cohort analyses were built to avoid, and it is small by the standards of this literature. It is a real inconsistency in the evidence base, not a refutation of the cohorts. A 2024 narrative review of eleven studies concluded much the same thing from the other end: some studies show an inverse correlation, others conflict, and no consensus exists.
Ejaculating 21 or more times a month cuts your prostate cancer risk by about 20 percent.
The inverse association is real and sizeable, and it appeared in both analyses of the Health Professionals cohort. The causal reading is not established, and the effect is narrower than the headline: in Rider 2016 the associations were driven by low-risk disease, and the case-control literature (Dimitropoulou 2009, 431 cases and 409 controls) found the opposite direction for activity in the twenties and thirties. Consistent inverse association in one large cohort analysed twice, inconsistent in case-control work, no demonstrated causation.
What the finding does not reach
This is incidence, not mortality. Rider's associations were concentrated in low-risk, screen-detected disease, in a cohort followed through an era of heavy PSA testing. Health media compressed the result into a claim about dying of prostate cancer. No one has shown that ejaculation frequency changes those odds either way, and whether it does may not be answerable with observational data at all.
That is a limit on the finding's reach, not a reason to set it aside. The association with diagnosis is there, it held up under the obvious objection, and it points away from the prostate claim that retention content makes.
Ejaculating more often means you are less likely to die of prostate cancer.
Nobody has shown this. The cohort findings are about diagnosis, and in Rider 2016 the association was driven by low-risk disease. Leitzmann 2004 had 147 advanced cases, too few to resolve the question either way. No relationship between ejaculation frequency and prostate cancer mortality has been established in either direction.
What this means if you are retaining
Nothing here shows that abstinence is dangerous, and nothing here has tested it. What the evidence says is narrower and still uncomfortable: the best available observational data on the adjacent exposure run opposite to the prostate benefit that retention content advertises, and there is no counterweight, because nobody has ever followed a defined group of men practising deliberate retention and measured a single urological endpoint. Not symptom scores, not ejaculatory function, not pelvic floor tone, not semen parameters, over 30 or 60 or 90 days. The most basic descriptive study does not exist.
There is no counterweight, because nobody has ever studied a group of men who were deliberately retaining.
So the honest reading is not alarm. It is that a benefit claim is being made where the nearest available data point the other way, and where the population the claim is about has never been measured. That is a reason to hold the claim loosely, and to drop it from your reasons for practising if it was one of them.
None of this substitutes for clinical care. Persistent urinary symptoms, blood in urine or semen, ongoing pelvic or testicular pain, or a family history that concerns you are all reasons to see a doctor. Prostate risk is not something a habit tracker can screen for, and no journalling practice should delay that appointment.
Why an average cannot answer your question
A hazard ratio of 0.78 is a statement about how two large groups of men differed across half a million person-years. It describes no individual in either group. It cannot be run on you, it was never measured in anyone doing what you are doing, and the confounding that produced it is not resolvable at population scale by any study anyone is likely to fund. Even if it were causal, it would tell you about population incidence and not about how a given week feels, or what changes in your sleep, mood, libido or training.
That gap does not close with more reading. It closes, partially and only for you, with a record.