Abstract
Background
Few studies have addressed the clinical parameters' predictive power related to caries lesion associated with their progression. This study assessed the predictive validity and proposed simplified models to predict short-term caries progression using clinical parameters related to caries lesion activity status.
Methods
The occlusal surfaces of primary molars, presenting no frank cavitation, were examined according to the following clinical predictors: colour, luster, cavitation, texture, and clinical depth. After one year, children were re-evaluated using the International Caries Detection and Assessment System to assess caries lesion progression. Progression was set as the outcome to be predicted. Univariate multilevel Poisson models were fitted to test each of the independent variables (clinical features) as predictors of short-term caries progression. The multimodel inference was made based on the Akaike Information Criteria and C statistic. Afterwards, plausible interactions among some of the variables were tested in the models to evaluate the benefit of combining these variables when assessing caries lesions.
Results
205 children (750 surfaces) presented no frank cavitations at the baseline. After one year, 147 children were reassessed (70%). Finally, 128 children (733 surfaces) presented complete baseline data and had included primary teeth to be reassessed. Approximately 9% of the reassessed surfaces showed caries progression. Among the univariate models created with each one of these variables, the model containing the surface integrity as a predictor had the lowest AIC (364.5). Univariate predictive models tended to present better goodness-of-fit (AICs < 388) and discrimination (C:0.959–0.966) than those combining parameters (AIC:365–393, C:0.958–0.961). When only non-cavitated surfaces were considered, roughness compounded the model that better predicted the lesions' progression (AIC = 217.7, C:0.91).
Conclusions
Univariate model fitted considering the presence of cavitation show the best predictive goodness-of-fit and discrimination. For non-cavitated lesions, the simplest way to predict those lesions that tend to progress is by assessing enamel roughness. In general, the evaluation of other conjoint parameters seems unnecessary for all non-frankly cavitated lesions.
Land Subsidence Prediction using Recurrent Neural Networks
