Predicting drooling severity after surgery in youth with neurodevelopmental disabilities: development of a...

Predicting outcomes of surgery for drooling could support candidate selection and shape expectations for patients and caregivers. This study describes the development of a prognostic prediction model for the expected drooling Visual Analog Scale (VAS) at 32-weeks post-surgery in youth with a neurodevelopmental disability.

A multivariable linear model was developed with data from a prospectively gathered pseudonymized database of drooling patients from the Saliva Control Clinic of a Dutch tertiary care center. The population consisted of drooling patients with a non-progressive neurodevelopmental disability who underwent surgery for drooling between January 2000 and March 2023. The outcome measure for the model was the VAS at 32-week follow-up. The baseline VAS was incorporated in the model to take into account the relative change compared to baseline. Internal validation was performed using bootstrapping.

The model was developed based on 227 patients (37.4% female, median age at surgery = 14.0). The final parsimonious model included 15 variables, resulting in an apparent pooled R² = 0.26 and an R²-adjusted= 0.21. Internal validation resulted in a calibration slope of 0.86 with an optimism-corrected R² of 0.16. Several parameters contributed significantly to predicting the outcome after surgery. These included age, VAS at baseline, type of surgery, type of neurodevelopmental disability, ambulatory status, use of laxatives or prokinetic agents, severe dysphagia, and abnormal oral suction and propulsion capacity.

This study provides a proof of principle that a multivariable linear prediction model can estimate postoperative anterior drooling severity in youth with neurodevelopmental disabilities undergoing surgery, using routinely available clinical characteristics. The model has not undergone external validation and is therefore not intended for clinical use in its current form. Rather, these findings demonstrate the potential of individualized outcome prediction in this population and support multicenter collaboration for external validation and further refinement. If validated, such a model could eventually support clinicians and caregivers in setting realistic expectations and assessing whether the anticipated benefit of an already selected surgical procedure is sufficient to proceed with surgery.