Analysis of factors associated with comorbid anxiety and depression in patients with functional dyspepsia: a...

To explore factors associated with comorbid anxiety and depression (CAD) in patients with functional dyspepsia (FD) and to compare the performance of logistic regression models (LRM) and decision tree models (DTM) in risk identification and assessment.

A cross‑sectional design was used. A total of 1,227 FD patients were recruited from a tertiary hospital in Xinjiang between January 2023 and August 2024. Assessments included the Generalized Anxiety Disorder‑7 scale (GAD‑7), Patient Health Questionnaire‑9 (PHQ‑9), Gastrointestinal Symptom Rating Scale (GSRS), Multidimensional Health Locus of Control Scale (MHLC), and the Simplified Nepean Dyspepsia Index (SF‑NDI). Candidate variables were selected based on clinical knowledge and a causal framework (informed by directed acyclic graph principles). Binary logistic regression and a decision tree model using the CRT algorithm were used to analyse factors associated with CAD. To ensure fair comparison, the data were randomly split into a training set (70%) and a test set (30%), and both models were developed on the same training set and evaluated on the same test set. The decision tree was subjected to 10‑fold cross‑validation, and the logistic regression was assessed using the Hosmer‑Lemeshow goodness‑of‑fit test. A sensitivity analysis was performed using a higher cut‑off (≥ 10 on both GAD‑7 and PHQ‑9). ROC curves were plotted to compare the discriminative ability of the two models.

The detection rate of CAD was higher in female FD patients (33.9%) than in males (26.3%). Logistic regression showed that a higher score for powerful others health locus of control (PHLC) (OR = 1.022, 95% CI: 1.001-1.043, P = 0.038), more severe gastrointestinal symptoms (OR = 1.050, 95% CI: 1.033-1.068, P < 0.001), and a higher SF‑NDI score (OR = 1.029, 95% CI: 1.012-1.048, P = 0.002) were associated with a higher likelihood of CAD; male sex, older age and higher education level were protective factors. In the decision tree model, SF‑NDI was the most important factor (normalised importance = 100%), followed by GSRS (68%), sex (52%), age (47%), education level (39%), chance health locus of control (CHLC, 33%) and PHLC (30%). On the test set, the LRM achieved an AUC of 0.710 (95% CI: 0.679-0.741) with a sensitivity of 74.1% and specificity of 58.1%; the DTM achieved an AUC of 0.764 (95% CI: 0.736-0.792) with a sensitivity of 71.4% and specificity of 66.6%. The mean AUC from 10‑fold cross‑validation was 0.708 ± 0.012 for the LRM and 0.759 ± 0.015 for the DTM. The Hosmer‑Lemeshow test gave P = 0.292, indicating good calibration of the logistic regression. In the sensitivity analysis using a cut‑off of ≥ 10, the direction and magnitude of effects for core variables remained largely unchanged (GSRS: OR = 1.047, P < 0.001; SF‑NDI: OR = 1.026, P = 0.003; PHLC: OR = 1.020, P = 0.048). Sex‑stratified analysis showed that SF‑NDI, GSRS, CHLC and age were the main factors in males, whereas PHLC and education level were additionally influential in females.

CAD in FD patients is clearly associated with sex, age, education level, health locus of control orientation, severity of gastrointestinal symptoms, and disease‑related quality of life. Logistic regression and decision tree models each have advantages in identifying associated factors and assessing risk; using them jointly can help clinicians identify high‑risk individuals early and develop personalised, integrated physical and psychological interventions.