Sex differences in the association between marital status and the risk of cardiovascular, cancer, and all-cause mortality: a systematic review and meta-analysis of 7,881,040 individuals

Pooling observational studies to ask whether a correlation is larger for men is appropriate, and death was measured well, mostly from national registries. The problems are structural. Marital status was recorded once at baseline, then treated as fixed over follow-up periods lasting up to 29 years. People marry, divorce, and are widowed during those decades; the baseline label increasingly diverges from reality. One Danish registry study contributes roughly 82% of all participants, so the pooled estimates largely reflect a single country. The 21 studies adjust for different sets of confounders — one controls only for age, another for over 20 variables including depression and disability — yet their estimates are pooled as though they measure the same thing. The central confound, that healthier people are more likely to marry, is uncontrolled in every included study. If this selection pressure differs by sex — and it almost certainly does, since never-married men in many of these populations are far more negatively selected than never-married women — the entire sex difference could be an artifact of who gets married rather than what marriage does.

The core machinery — random-effects meta-analysis with a ratio-of-ratios method to compare sexes — is standard and competently executed. The analysis runs dozens of separate tests (five mortality outcomes, four marital-status categories, multiple subgroups) with no correction for multiple comparisons. At that volume, some results crossing the conventional threshold are expected by chance. The claim that never-married men have 31% higher stroke mortality than never-married women rests on a confidence interval extending all the way to no difference. The abstract presents this as a finding rather than a suggestion. The paper claims the results hold up based on leave-one-out sensitivity analysis, but with the Danish study contributing 82% of participants, removing it is the one test that matters — and its specific effect is not reported. The paper's bias tests detected problems in three subgroup analyses; claiming the results hold up while reporting detected bias is contradictory. Causal language is used throughout the discussion for observational associations.

Death is measured reliably. Marital status is not. A one-time checkbox recorded years or decades before death does not measure the sustained experience of being married or unmarried. The paper's framing requires that this snapshot captures something meaningful across the full follow-up, and it does not. The paper interprets a larger correlation in men as evidence that men "benefit more from marriage," but this requires that confounding is identical for both sexes. It is not: the reasons a man remains unmarried at age 55 in Bangladesh differ from the reasons a woman does. If sicker men are more strongly filtered out of marriage than sicker women, the sex difference is an artifact of composition, not a causal finding. The mechanistic claims — stress hormones, social support, health behaviors — have no corresponding measurement in any of the 21 studies and are imported from unrelated research.

The paper's central move — treating an observational correlation as evidence that marriage causally protects men more — drives all of its conclusions and is unsupported by its design. The most important alternative explanation, that the sex difference reflects who gets married rather than what marriage does, receives one sentence. If you replaced the causal narrative with the selection narrative, every number in the paper would remain identical but the conclusion would reverse. The discussion constructs a four-step biological story (unmarried → stress → hormonal response → sex-differential mortality) where no part of the story is actually tested. The social support claim rests on a statistic from a consumer health newsletter about American men, applied across 15 countries including Bangladesh and Japan. A borderline stroke result consistent with no effect is presented as definitive in both the abstract and conclusion.

The paper follows standard reporting guidelines (MOOSE, PRISMA), discloses government funding, and reports no conflicts. Search strategies and inclusion criteria are documented. The limitations section acknowledges residual confounding, inability to determine causation, unexamined partnership types, and detected publication bias. It omits the most consequential threats: the dominance of a single massive study, the one-time measurement of an exposure that changes over decades, and the drastic variation in confounders across studies. The study was not pre-registered, no analysis code was shared, and no power analysis assessed whether the number of included studies was sufficient for the subgroup analyses reported.

The meta-analytic pooling cites its methodological sources correctly. The discussion section's biological mechanism argument does not. Four citations underpinning the claim that testosterone drives atherosclerosis and immunosuppression do not say what the paper implies: Wang 2012 studied testosterone's effect on muscle protein in obese premenopausal women, Engström 2006 examined occupation and marital status as cardiovascular risk factors, Evans and Goldsmith 2000 measured testosterone's effect on antibody production in male house sparrows, and Pinheiro 2015 studied cardiac dysfunction in elderly women with heart attacks. None support claims about testosterone driving atherosclerosis in men. A key statistic — that 66% of men rely on their wives as their primary social support — comes from a consumer health newsletter (Harvard Men's Health Watch 2010) with no peer-reviewed source. Manzoli 2007 explicitly found that the protective effect of marriage "did not vary by gender" in elderly populations. The paper cites Manzoli for general support but omits this contradictory finding on the paper's central question.

Government-funded academic work with no commercial stakeholder. The authors disclose no competing interests. The interpretive overreach is a scientific quality issue, not an independence issue.