Linear Regression Model for Predicting Dressing-Change Pain in Diabetic Foot Ulcers: Development, Internal Validation, and Personalized Analgesic Care.
Liping L, Junhong W, et al. • Journal of visualized experiments : JoVE • 2026
A linear regression model using ulcer area, ulcer duration, duration of diabetes, and SAS score predicted dressing-change pain in diabetic foot ulcers, and a model-based personalized analgesic regimen improved pain, adherence, and wound healing outcomes compared to conventional care.
Key Findings
Results
Multivariable linear regression identified four predictors of dressing-change pain in diabetic foot ulcers: ulcer area, ulcer duration, duration of diabetes, and Self-Rating Anxiety Scale (SAS) score.
The study enrolled 232 patients with diabetic foot ulcers (DFUs) across two stages.
Stage 1 included 160 patients divided into a training set (n = 113) and an internal validation set (n = 47).
Pain was measured as a continuous outcome using the Visual Analog Scale (VAS).
The four predictors were identified through multivariable linear regression analysis.
Results
The linear regression model achieved an adjusted R² of 0.458 and an RMSE of 1.719 in the training set.
Adjusted R² of 0.458 indicates the model explained approximately 45.8% of the variance in dressing-change VAS scores.
Training set RMSE was 1.719.
These metrics were derived from the training set of 113 patients.
Results
In internal validation, the model demonstrated useful discrimination for identifying high dressing-change pain with an AUC of 0.913.
Validation set included 47 patients.
Validation-set RMSE was 2.058.
Calibration intercept was 1.414 and calibration slope was 0.721, indicating some miscalibration.
AUC for identifying high pain was 0.913, indicating strong discriminative ability.
Results
The intervention group receiving model-based personalized analgesic care had significantly lower dressing-change VAS scores than the control group after 4 weeks.
Stage 2 included 72 cases assigned to a control group or intervention group.
After 4 weeks, intervention group VAS score was 3.83 ± 1.61 versus 4.94 ± 1.91 in the control group (P = 0.010).
The intervention group received personalized analgesic regimens based on model-derived risk stratification.
The control group received conventional analgesic care.
Results
The intervention group demonstrated significantly higher overall treatment adherence compared to the control group.
Overall adherence rate was 86.11% in the intervention group versus 63.89% in the control group.
The difference in adherence was statistically significant (P = 0.029).
Adherence was assessed over a 4-week follow-up period.
Results
The intervention group had a significantly higher ulcer area reduction rate than the control group.
Ulcer area reduction rate was 41.84% ± 11.01% in the intervention group versus 26.22% ± 7.55% in the control group.
The difference was statistically significant (P < 0.001).
This outcome was measured after 4 weeks of intervention.
Results
Wound-healing time was significantly shorter in the intervention group compared to the control group.
Median wound-healing time was 32.0 days (IQR 29.0–35.0) in the intervention group.
Median wound-healing time was 38.0 days (IQR 34.0–43.5) in the control group.
The difference was statistically significant (P < 0.001).
Conclusions
The authors note that external validation and further refinement of model calibration are required before routine clinical implementation.
The calibration slope of 0.721 and intercept of 1.414 indicate the model has calibration limitations.
Only internal validation was performed in this study.
The authors explicitly state that external validation is needed prior to clinical adoption.
What This Means
This research suggests that pain experienced by diabetic foot ulcer patients during wound dressing changes can be predicted using a statistical model that incorporates four factors: the size of the ulcer, how long the ulcer has been present, how long the patient has had diabetes, and the patient's anxiety level as measured by a standard questionnaire. The model was developed in a group of 113 patients and tested in a separate group of 47 patients, where it showed strong ability to identify patients likely to experience high levels of pain, though its accuracy in predicting the precise pain score was moderate.
In a second part of the study, 72 patients were divided into two groups. One group received standard pain management, while the other received a personalized pain management plan based on their predicted pain level from the model. After four weeks, the group receiving personalized care reported meaningfully lower pain during dressing changes, were more likely to follow their treatment plans, had greater reductions in wound size, and healed faster compared to those receiving standard care.
This research suggests that using a simple prediction model to tailor pain management for diabetic foot ulcer patients could lead to better pain control and faster wound healing. However, the authors caution that the model still needs to be tested in different patient populations and that its accuracy in predicting exact pain scores needs improvement before it can be widely adopted in clinical practice.