A machine learning investigation has evaluated predictive factors associated with conservative treatment failure in patients experiencing both obstructed defecation syndrome and type 2 diabetes mellitus. Obstructed defecation syndrome can be complicated by systemic diabetic neuropathy, creating clinical challenges for standard non-surgical management. In this study, investigators analyzed clinical, metabolic, and pelvic dynamic features to determine which disease characteristics most strongly predict when conservative therapy will fail.
The study analyzed a balanced cohort of 513 comorbid patients to train and compare six distinct machine learning algorithms. The authors sought to investigate the underlying mechanisms of these comorbid conditions while creating models capable of predicting conservative treatment outcomes. Across the evaluated approaches, the Random Forest model demonstrated a comparable global discriminative capacity, achieving an area under the curve of 0.814.
To understand the model’s inner decision-making structure, the authors interpreted the Random Forest model using SHapley Additive exPlanations, commonly referred to as SHAP analysis. This interpretability method enabled the evaluation of feature importance, non-linear associations, and variable interactions. The analysis identified glycated hemoglobin, duration of type 2 diabetes mellitus, and pudendal nerve terminal motor latency as the most important predictive variables.
Beyond glycemic and neurological markers, pelvic floor measures also contributed to the algorithm’s discriminative performance. The SHAP analysis revealed that the Wexner score and maximum voluntary squeeze pressure ranked directly behind the metabolic and neuropathic indicators in overall feature importance. These findings show that both systemic disease progression and local functional deficits factor into treatment outcomes.
The interaction analysis further identified a potential interactive association between poor glycemic control and localized nerve impairment. When evaluated together, this combination was associated with a higher estimated risk of conservative treatment failure on the multiplicative scale. The interaction suggests that systemic metabolic dysfunction and localized pelvic denervation may compound risk beyond their individual contributions.
The authors concluded that machine learning algorithms are effective tools for identifying high-risk patients with comorbid obstructed defecation syndrome and type 2 diabetes. Predictive performance in the cohort appeared to be driven primarily by systemic metabolic-neuropathic decline rather than being solely related to anatomical changes. While the model-derived exploratory thresholds provide quantitative and hypothesis-generating insights, independent external validation remains necessary before applying these models to individual clinical decision-making.
Conducted by investigators at the Affiliated Hospital of Panzhihua University and published in the Clinical Diabetes section of Frontiers in Endocrinology, the study addresses the clinical intersection of pelvic floor dysfunction and endocrine disease.
To assess outcomes comprehensively, the researchers trained six distinct machine learning algorithms using clinical, metabolic, and pelvic dynamic features gathered from a balanced cohort of 513 comorbid patients. Utilizing SHapley Additive exPlanations allowed the team to uncover nonlinear associations and examine how individual variables interact to influence failure rates during conservative interventions.
These computational insights emphasize that risk in comorbid individuals accumulates on a multiplicative scale, particularly when poor glycemic control converges with localized pudendal nerve impairment. However, the authors explicitly outlined the boundaries of their conclusions. The quantitative thresholds identified by the algorithms remain exploratory and hypothesis-generating. Consequently, independent external validation across separate patient populations is mandatory before these predictive models can be safely translated into routine clinical care or individual patient management.
Subject of Research: Machine learning prediction of conservative treatment failure in obstructed defecation syndrome and type 2 diabetes
Article Title: Interactive association of diabetic somatic neuropathy and pelvic floor dynamics with conservative treatment failure in obstructed defecation syndrome: a machine learning study
Article References: Fu, W., Xue, Y., Wang, F., & Xiao, X. (2026). Interactive association of diabetic somatic neuropathy and pelvic floor dynamics with conservative treatment failure in obstructed defecation syndrome: a machine learning study. Frontiers in Endocrinology, 17, Article 1976282. https://doi.org/10.3389/fendo.2026.1976282
Image Credits: AI Generated
DOI: 10.3389/fendo.2026.1976282
Keywords: obstructed defecation syndrome, type 2 diabetes mellitus, diabetic somatic neuropathy, machine learning, Random Forest, SHAP analysis, HbA1c, pudendal nerve terminal motor latency, Wexner score, pelvic floor dynamics, conservative treatment failure, maximum voluntary squeeze pressure, glycemic control, clinical diabetes, metabolic neuropathy, pelvic dynamics
News Source: Teresa Odom. (November 19, 2026). Machine Learning Links Diabetic Neuropathy to Obstructed Defecation Therapy Failure. Scienmag.



