Prediction ability of Oral Mortality Estimator Score in the Elderly

Objective: Number of teeth, dental occlusion, and periodontal health have been reported to be predictive of mortality. We created Oral Mortality Estimator Score (OMES) in a cohort of elderly nursing home patients (mean age=79) utilizing baseline data (N=370) and tested its ability to predict mortality over a median follow-up of 5 years. Methods: All oral health variables in our dataset were evaluated by bivariate logistic regression analyses with mortality as the outcome and retained those variables that reached a predetermined p-value for potential inclusion in the final model. Candidate variables were then re-assessed by backward elimination with manual appraisal for confounding and co-linearity at each step and reached at a final model consisting of 3 variables, i.e., low number of remaining teeth, high salivary urea levels and high salivary candida levels. Each variable was weighted by its likelihood ratio and summed to create Oral Mortality Estimator Score (OMES). We assessed the test characteristics of OMES including sensitivity, specificity, predictive value positive (PPV+) and predictive value negative (PPV-). Results: The optimal sensitivity of OMES was 0.63, specificity 0.52, PPV+ 0.49, and PPV- was 0.66 respectively. The overall performance of OMES was comparable to age which is the single most powerful predictor of all-cause mortality with sensitivity 0.66, specificity 0.55, PPV+ 0.52, PPV- 0.69. C-reactive protein (CRP) >= 3 mg/L displayed sensitivity 0.52, specificity 0.30, PPV+ 0.52 and PPV- 0.39. These results are presented in Table 1. Conclusion: OMES demonstrated comparable ability to age in predicting 5-year mortality in this elderly population. Further testing in other cohort is needed to validate the prediction ability of OMES.