Comments on “Predicting delayed graft function after adult kidney transplantation procedure through donor-recipient demographic-anthropometric characteristics analysis: Extensive systematic review and meta-analysis of observational studies”

Dear Editor, We read with interest the article by Warli et al,1 “Predicting delayed graft function after adult kidney transplantation procedure through donor-recipient demographic-anthropometric characteristics analysis.” The authors tackle a clinically important question and synthesize a substantial body of observational evidence. We offer several methodological comments to aid interpretation. The review has important strengths. The protocol was registered in PROSPERO, and a PRISMA-style flow diagram was presented, which enhances transparency. The search strategy targeted contemporary adult cohorts, and random-effects models were used, which is appropriate in the presence of between-study variability. Focusing on simple demographic and anthropometric variables is clinically pragmatic because these characteristics are routinely available at the time of transplantation. Our main concern relates to the framing of the study as identifying “predicting factors.” Contemporary guidance on prediction research, such as the TRIPOD Statement, stresses that prediction studies should develop or validate multivariable models and report performance measures such as discrimination and calibration2. In contrast, Warli et al1 pooled univariable mean differences and odds ratios comparing delayed graft function with immediate graft function groups. No multivariable model was developed or validated, and potential confounding was not addressed. Guidance for systematic reviews of prognostic factors emphasizes that adjusted estimates are preferred when assessing independent prognostic value3. The results, therefore, provide useful evidence of association but should not be interpreted as directly yielding clinical prediction rules. Second, statistical heterogeneity is substantial, with I² statistics often above 60%. As Higgins et al4 have highlighted, random-effects models accommodate but do not explain inconsistency between studies. Likely sources here include varying definitions of delayed graft function, differences in donor type, region and calendar period. Methodological guidance recommends prespecified subgroup analyses or meta-regression to explore such heterogeneity and to avoid over-generalising summary effects3. In this review, studies using different definitions of delayed graft function were pooled into a single outcome, which may introduce outcome misclassification and mask clinically relevant differences. Third, the deliberate exclusion of registry-based studies, on the grounds of their database-only design and larger size, may introduce selection bias. Registries frequently provide the most representative data on current practice; omitting them may limit external validity. In addition, restricting extraction to unadjusted group-level data means that residual confounding is likely, especially where demographic variables correlate with established determinants of delayed graft function such as cold ischemia time and donor quality. Finally, we note a minor reporting issue. The abstract states that 50 studies were included, whereas the results section refers to 49 studies and 20,497 transplant procedures. Consistent reporting of denominators is important for reproducibility and for understanding the precision of pooled estimates. Warli et al1 provide a helpful overview of associations between donor and recipient characteristics and delayed graft function after kidney transplantation. More cautious use of prediction terminology, fuller investigation of between-study heterogeneity and a greater emphasis on adjusted prognostic estimates in future research would further strengthen the methodological rigour and clinical utility of work in this area.