Limits of questionnaire-based prediction of oral health-related quality of life in non-smokers.

backgroundOral health-related quality of life (OHRQoL) reflects the perceived impact of oral health on daily functioning, psychosocial well-being, and social interaction. While clinical symptoms strongly influence OHRQoL, it is unclear how well OHRQoL can be estimated using only non-clinical information, particularly in non-smoking populations. This study evaluated whether sociodemographic, behavioral, and self-perceived health variables can predict OHRQoL in non-smokers.

methodsThis cross-sectional study recruited 130 adults. Participants completed a questionnaire capturing demographics, behavioral habits, and self-rated health. OHRQoL was measured using the OHIP-14. Bayesian regression was used to examine associations between these variables and OHIP-14 scores. Additional analyses were performed to identify the variables contributing most to predictions and examine predictions at the individual level. Predictions were compared with 14 machine learning (ML) models to evaluate whether algorithmic complexity could overcome model limitations.

resultsThe Bayesian model explained 30% of variance in OHIP-14 scores (Bayesian R² = 0.30, 95% CrI: 0.20-0.39). Educational level, gender, and self-rated oral health were the strongest predictors. Simulated changes in self-rated oral health and dental attendance produced the largest reductions in predicted population-level OHIP-14 scores. The model performed reasonably well for individuals with mild-to-moderate impairment but systematically underestimated individuals with severe impairment (OHIP-14 > 30). ML methods did not substantially improve identification of severe cases.

conclusionsNon-clinical questionnaire information can provide moderate estimation of OHRQoL, particularly among individuals with mild-to-moderate impairment, but has limited ability to identify severe burden. These findings suggest that digital or questionnaire-based screening should incorporate a small number of symptom- or clinically relevant questions when the goal is to detect individuals with severe OHRQoL impairment. However, given the modest sample size, convenience sampling, and absence of cross-validation and external validation, the generalizability and robustness of these findings require confirmation in larger studies with independent external validation.