Sex and disease differences in patients’ diagnostic experiences in autoimmune diseases: an international cross-sectional study

The study surveyed a large number of people and used reasonable methods for an online cross-sectional design. But two problems are fundamental.

First, no outcome is validated against any external source. "Misdiagnosis" is what the patient believes, captured by a single yes/no question. A clinician who mentioned stress as a contributing factor, a clinician who correctly diagnosed coexisting anxiety, and a clinician who dismissed autoimmune symptoms as psychiatric all produce the same "yes" answer. The study cannot distinguish among these clinically different situations.

Second, the adjusted models leave out the confounders most directly relevant to the central finding. Women in the sample reported anxiety at 24% versus 16% for men, and fibromyalgia at 14% versus 5%. A woman who has both developing autoimmune disease and genuine anxiety may receive a correct anxiety diagnosis that she later reinterprets as a misdiagnosis of her autoimmune symptoms. This would produce a sex difference in reported misdiagnosis driven entirely by differences in actual comorbidity, with no clinician bias required. The data to test this were collected. They were not used.

The headline comparisons on diagnostic delay and visit counts are never adjusted for disease type, despite large differences in disease distribution between the sexes. Forty-four percent of the sample dropped out before the misdiagnosis questions, and the paper does not report whether dropout rates differed by sex. The survey was distributed through patient charities and advocacy channels, with no way to calculate response rates or assess how much self-selection shaped the sample.

The statistical methods are well-chosen for the data — Bayesian ordinal regression for ordered categories, Firth penalized logistic regression for sparse cells. The Bayesian models use diffuse priors to let the data dominate. Odds ratios are reported with confidence intervals.

The failure is in model specification, not technique. The adjusted models control for disease group, age, continent, and household finances. They do not control for comorbid anxiety (8 percentage points higher in women), depression, or fibromyalgia (nearly three times more common in women). These are the variables most likely to explain why women would report more psychiatric misdiagnosis even without any difference in clinician behavior. The data were available. The analysis was not run.

The diagnostic delay and visit-count comparisons — two of the paper's five core claims — are presented without any adjustment for disease type. Women were concentrated in thyroid disease, lupus, and Sjögren's disease, all of which have vaguer biomarkers and characteristically longer diagnostic paths. Men were concentrated in rheumatoid arthritis and vasculitis, which tend to present more recognizably. The reported sex difference in delay could reflect differences in what diseases men and women have, not differences in how clinicians handle them.

The Cramér's V for the sex-delay association is 0.06 — very weak. In a sample of over 6,000, even trivial associations produce small p-values. The paper reports this statistic but frames the association as meaningful without addressing its small magnitude.

The paper measures what patients believe happened during their diagnostic journey. The abstract and interpretation present the findings as evidence about what clinicians did.

The abstract states that autoimmune disease had been "attributed to psychiatric and psychosomatic causes" at specific rates, without the qualifier "perceived." The interpretation recommends that clinicians "consider autoimmune disease at an early stage" and that "sex biases in medicine should be examined, challenged and mitigated." These recommendations follow only if the survey data reflect actual clinical misattribution rather than retrospective reinterpretation, correctly diagnosed comorbid conditions, or differential diagnoses that were reasonable given what was known at the point of care.

The paper acknowledges the perception-versus-reality issue in its limitations section, noting that "clinicians may have differing interpretations in some cases" and that "some reported misdiagnoses may reflect initial differential diagnoses." But these qualifications do not appear in the abstract. The adjustment for disease group shows that disease mix alone does not explain the sex difference — a legitimate finding. But the model excludes actual psychiatric comorbidity. The residual sex effect after adjusting for disease group, age, continent, and finances cannot be interpreted as evidence of bias when anxiety, depression, and fibromyalgia are omitted.

The paper moves from "women report more perceived misdiagnosis" to "women receive more misdiagnosis" without establishing that the second follows from the first.

Women in this sample had higher rates of anxiety, depression, and fibromyalgia. A clinician who diagnoses anxiety in a patient who has anxiety is not misdiagnosing, even if the patient later develops autoimmune disease and reinterprets the anxiety diagnosis as an error. This alternative explanation accounts for the observed sex difference without invoking clinician bias, and it was testable with the study's data. The test was not run.

The paper's argument that recall bias "is unlikely to differ systematically by sex" is stated without evidence. Women in the sample reported longer diagnostic journeys, meaning they recalled events over longer time windows with more clinical encounters. Women also had higher rates of anxiety, which is associated with more ruminative processing of negative healthcare experiences. The recall task itself differs between the groups even if neither group is fabricating memories.

The causal claim — that the disparity "may reflect both biological differences and systemic biases in healthcare" — is interpretive commentary presented as a finding. The study collected no data on clinician behavior, reasoning, or disease severity at presentation. It cannot distinguish clinician bias from differences in disease presentation, health-seeking behavior, comorbidity, or recall.

The study has a registered protocol, disclosed funding and conflicts, obtained ethics approval, and includes a substantive limitations section. It reports a null finding (no sex difference in physical-disease misdiagnosis) alongside the positive findings.

No data, survey instrument, or analysis code is shared. No sample size justification was performed for the primary analysis. The 44% attrition on the misdiagnosis outcome is reported only as a total, without sex-specific breakdown. The paper collected comorbidity data that bear directly on the central finding, reported them in the descriptive table, and excluded them from every model without explanation.

The paper's descriptive findings stand on its survey data and do not depend on citations. The interpretive framework — that the sex difference reflects systemic bias — draws on a narrow set of sources, six from the lead author's research program. These self-citations are relevant, but no independent replication of the interpretive claims is cited.

The argument about "historical biases in medicine" cites studies documenting contemporary sex differences in diagnosis, then attributes these to historical bias without evidence for that causal link. A citation for medical-record errors (Weiner 2020) is used to argue for the validity of patient recall — a stretch, since errors in clinical notes do not establish that patient retrospective accounts are more accurate. Literature on sex differences in health-seeking behavior and on the clinical validity of functional somatic diagnoses — both directly relevant — is absent.

The study was funded by five UK patient charities, recruited through those charities' networks, and reached conclusions consistent with their advocacy position: that patients, especially women, face dismissive care. This is not the same pressure as pharmaceutical industry funding, but it is a directional interest that aligned with every analytical choice in the paper.

The decision not to adjust for psychiatric comorbidities is the most concrete concern. Including anxiety, depression, and fibromyalgia in the models could have attenuated or eliminated the sex difference, producing a result less useful to the funding charities. The data existed. The paper adjusts for disease group, age, continent, and finances — variables that do not threaten the core finding — but not for the variables that most directly could.