A Cephalometric Logistic Model to Predict Altered Jaw-posture in Apnoeics

Over time, certain patients wearing mandibular advancement devices to manage obstructive sleep apnea were observed to develop irreversible alterations in jaw posture, with associated tooth damage. A retrospective pilot study (N=34) was conducted to see whether cephalometric predictors could be found that would be capable of explaining changes in those who had been affected. Logistic regression analysis identified gonial angle (GA) and maxillary-mandibular plane angle (MMPA) as vertical cephalometric predictors; while pterygoid advancement proportion (PtAP) provided a horizontal predictor. Two distinct patterns of jaw reposturing were discerned, each being associated with either a posterior or intermediate open bite pattern. By measuring the cephalometric parameters for each subject and plugging these into the two logistic formulae, the specific open bite types that had occurred in affected individuals within the pilot-study were correctly predicted. Objectives: It was hypothesised that applying these same predictor formulae to a different set of individuals in a follow-up study would correctly predict those at risk of developing altered jaw posture. Methods: 17 patients were prospectively analyzed relative to the two predictor formulae. Decision matrix analysis was undertaken using Fisher's exact test. Results: The logistic models' predictive abilities returned sensitivity and specificity scores for intermediate open bite cases of 80% and 92% respectively (p=0.01**); however, with sensitivity of 50%, the more destructive posterior open bite group was less well predicted (p=0.121). Consequently, in accordance with predictive model enhancement strategies, study-data were re-analyzed using a redefined product term. This produced an improved model (p<0.001***) that offered precise predictive ability. Conclusion: Pooling study-data and pilot-data (N=51) correctly predicted all 7 subjects who had developed posterior open bite and 44 who hadn't. Both predictor models promise to provide a statistically robust basis for clinical application by providing means of alerting patients to possible changes in jaw posture, before they commence treatment.