Incremental Predictive Value of Conventional MRI Pelvimetry for Major Low Anterior Resection Syndrome After Rectal...

Background: Low anterior resection syndrome (LARS) is a frequent functional complication after sphincter-preserving rectal cancer surgery. Although tumor location and treatment-related factors are established determinants, the contribution of pelvic skeletal anatomy assessed by magnetic resonance imaging (MRI) pelvimetry remains uncertain. This study evaluated whether three conventional MRI-derived skeletal pelvimetry measurements provide incremental predictive information for major LARS beyond a parsimonious preoperative clinical reference model. Methods: This single-center observational cohort study included 168 patients who underwent sphincter-preserving rectal cancer surgery and completed postoperative LARS assessment. MRI pelvimetry included anteroposterior pelvic diameter, interspinous distance, and sacral curvature depth. Associations with LARS score and major LARS were evaluated using correlation analyses, pelvic-tertile comparisons, and multivariable logistic regression. Firth penalized logistic regression was prespecified as a sensitivity analysis. The primary clinical reference model included age, categorical tumor position, and neoadjuvant radiotherapy. Internal validation used stratified five-fold cross-validation repeated 100 times. Incremental model fit was assessed using a likelihood-ratio test. Results: Major LARS occurred in 48 patients (28.6%), and the median postoperative LARS score was 18 (interquartile range 9-31). LARS scores and major LARS prevalence were similar across anteroposterior pelvic diameter tertiles. In the parsimonious multivariable model, low tumor position was associated with major LARS (OR 9.51, 95% CI 2.00-45.15; p = 0.005), whereas anteroposterior pelvic diameter and interspinous distance were not associated with major LARS. The parsimonious preoperative clinical model achieved a mean repeated five-fold cross-validated AUC of 0.669, compared with 0.671 after addition of MRI pelvimetry (ΔAUC = +0.002, 95% CI -0.052 to 0.054), with no statistically detectable improvement in nested-model fit (likelihood-ratio p = 0.122). Brier scores were 0.187 for both models. Calibration, overall prediction error, and decision-curve performance were similar between models. Conclusions: In this selected retrospective cohort, the three conventional MRI-derived skeletal pelvimetry measurements evaluated did not demonstrate a clear incremental improvement in prediction of major LARS beyond the parsimonious preoperative clinical reference model.