Author(s)
Katrina Bootes, BS
Elena Quinionez Del Cid, BS
Anna van Dorsten
Kees Heetderks
Nikhia Tsamasis
Nathan Barefoot, BS
Priya J Desai, BS
Tochigi Kosuke
Cristine Klatt-Cromwell, MD
Jackson Vuncannon, MD
Brian Thorp, MD, FARS
Charles Ebert, MD, FARS
Adam Kimple, MD, PhD, FARS
Brent Senior, MD, FARS
Affiliation(s)
University of North Carolina Chapel Hill; UNC Department of Otolaryngology/Head and Neck Surgery
Abstract:
Background: Survey burden is significant with frequent administration of long questionnaires such as the 22-item Sinonasal Outcome Test (SNOT-22), limiting survey accuracy and clinical utility. Prior efforts to reduce burden have produced several static short-form versions of the SNOT-22, but have not evaluated whether tailored tests, which can make accurate predictions with fewer items, can efficiently predict SNOT-22 outcomes. A decision tree model is one potential approach which requires less computational overhead than adaptive models.
Methods: A dataset of 2,005 anonymized SNOT-22 surveys from recent clinical encounters was divided into training (n=1605) and validation sets (n=400). Seven decision tree models were developed using Classification and Regression Tree (CART) and Chi-squared Automatic Interaction Detector (CHAID) methodologies. Each model was trained to predict the total SNOT-22 score and subsequently evaluated on the validation set.
Results: Among the evaluated models, CART demonstrated the best performance, predicting total SNOT-22 scores a median of 4.4 points from measured values (mean = 6.0) while only requiring eight or fewer item responses. This prediction error falls within established minimal clinically important difference guidelines (8.9-12 points) for the SNOT-22.
Conclusion: Decision tree models show strong potential for accurately estimating sinonasal symptom burden while substantially reducing patient survey burden. These findings support the feasibility of integrating low-computational, short-item decision trees into routine clinical assessment of chronic rhinosinusitis.