A Discussion on Methodological Considerations and the Interpretation of Findings in a Nursing Burnout Cohort Analysis

We recently read the article by Paixão et al.(da Paixão et al. 2026), titled “Burnout Syndrome Predictors in Nursing Professionals During and After the COVID-19 Pandemic,” published in the Journal of Clinical Nursing. The study used a multicenter design and multivariable regression analysis to examine the long-term impact of the pandemic on nursing professionals. However, we want to point out a fundamental flaw in the core data analysis methodology. This flaw directly calls into question its primary conclusion that burnout prevalence remained stable, which may be a statistical artefact rather than a reflection of reality. Our primary concern is that the authors treated the longitudinal data as if it were cross-sectional. The authors state they treated observations from two time points as independent rather than as repeated measures (section 2.7) due to participant loss. This approach is a misapplication of a prospective cohort design, whose principal strength lies in analysing intra-individual changes over time. The initial sample was 844 participants, but only 163 were successfully matched after 2 years, representing an attrition rate of 80.7%. Although the authors acknowledge a “healthy-worker effect” bias in the limitations, its actual impact may be even more severe. The final sample is very likely a “survivor” cohort, who may be more psychologically resilient, in better health, or in less adverse work situations. Conversely, nurses experiencing the most severe burnout, those most likely to resign or take extended leave, are precisely the ones most likely to be lost from the cohort. By treating these two disparate groups as independent cross-sectional samples, the study's central finding that burnout prevalence “remained stable” (9.2% vs. 7.4%) becomes highly suspect. This apparent stability is more plausibly an artefact of severe selection bias rather than a reflection of reality in the target population. Moreover, the regression results revealed a clinically unreasonable finding, leading us to question the reliability of the model. In table 7 of the article, the authors report that Medical leave is a significant negative predictor of Low professional accomplishment (β = −2.069, p = 0.011). This implies that nurses who took sick leave experienced higher professional accomplishment than those who did not. This conclusion defies clinical logic and professional experience (Ahola et al. 2008). Professional accomplishment stems from a sense of efficacy, contribution, and value realisation at work. This counterintuitive result is most likely a statistical artefact attributable to the small sample size and the scarcity of outcome events: only 12 burnout cases in 2022. Regression with multiple variables on a small sample is highly unstable, prone to producing exaggerated or directionally incorrect associations. The extremely wide confidence interval for Medical leave in the logistic regression (95% CI: 1.25–28.85, table 3) is clear evidence of this instability. Collecting follow-up data amidst a long health crisis is an arduous task, and the resultant missing data present a universal dilemma for researchers. In light of these constraints, their analytical choice is understandable. Yet because such hard-won data are scarce, we must analyse them rigorously to maximise their value. Therefore, we gently suggest that employing analytical techniques designed for longitudinal data, such as mixed-effects models, could offer complementary insights by modelling individual trajectories, even in the presence of missing data. In conclusion, the study by Paixão et al. provides valuable data and insights. We believe that addressing these methodological issues will not only strengthen the validity of the conclusions drawn from this important cohort study, but also enhance the practical value of its findings. We thank the authors for their research in this vital area of burnout research and look forward to further analyses of this valuable longitudinal dataset. Xingxia Peng: writing – original draft preparation. Nuoyi Zhang: formal analysis. Wangying Jiang: investigation. Tiantian Zhang: writing – review and editing. All authors have read and agreed to the published version of the manuscript. The authors have nothing to report. The authors have nothing to report. The authors declare no conflicts of interest. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.