Body Mass Index and Caries: Machine Learning and Statistical Analytics of the Dental, Oral, Medical Epidemiological (DOME) Nationwide Big Data Study

The objectives of the research were to analyze the association between Body Mass Index (BMI) and dental caries using novel approaches of both statistical and machine learning (ML) models while adjusting for cardiovascular risk factors and metabolic syndrome (MetS) components, consequences, and related conditions. This research is a data-driven analysis of the Dental, Oral, Medical Epidemiological (DOME) big data repository, that integrates comprehensive socio-demographic, medical, and dental databases of a nationwide sample of dental attendees to military dental clinics for 1 year aged 18−50 years. Obesity categories were defined according to the World Health Organization (WHO): under-weight: BMI < 18.5 kg/m2, normal weight: BMI 18.5 to 24.9 kg/m2, overweight: BMI 25 to 29.9 kg/m2, and obesity: BMI ≥ 30 kg/m2. General linear models were used with the mean number of decayed teeth as the dependent variable across BMI categories, adjusted for (1) socio-demographics, (2) health-related habits, and (3) each of the diseases comprising the MetS definition MetS and long-term sequelae as well as associated illnesses, such as hypertension, diabetes, hyperlipidemia, cardiovascular disease, obstructive sleep apnea (OSA) and non-alcoholic fatty liver disease (NAFLD). After the statistical analysis, we run the XGBoost machine learning algorithm on the same set of clinical features to explore the features’ importance according to the dichotomous target variable of decayed teeth as well as the obesity category. The study included 66,790 subjects with a mean age of 22.8 ± 7.1. The mean BMI score was 24.2 ± 4.3 kg/m2. The distribution of BMI categories: underweight (3113 subjects, 4.7%), normal weight (38,924 subjects, 59.2%), overweight (16,966, 25.8%), and obesity (6736, 10.2%). Compared to normal weight (2.02 ± 2.79), the number of decayed teeth was statistically significantly higher in subjects with obesity [2.40 ± 3.00; OR = 1.46 (1.35−1.57)], underweight [2.36 ± 3.04; OR = 1.40 (1.26−1.56)] and overweight [2.08 ± 2.76, OR = 1.05 (1.01−1.11)]. Following adjustment, the associations persisted for obesity [OR = 1.56 (1.39−1.76)] and underweight [OR = 1.29 (1.16−1.45)], but not for overweight [OR = 1.11 (1.05−1.17)]. Features important according to the XGBoost model were socioeconomic status, teeth brushing, birth country, and sweetened beverage consumption, which are well-known risk factors of caries. Among those variables was also our main theory independent variable: BMI categories. We also performed clinical features importance based on XGBoost with obesity set as the target variable and received an AUC of 0.702, and accuracy of 0.896, which are considered excellent discrimination, and the major features that are increasing the risk of obesity there were: hypertension, NAFLD, SES, smoking, teeth brushing, age as well as our main theory dependent variable: caries as a dichotomized variable (Yes/no). The study demonstrates a positive association between underweight and obesity BMI categories and caries, independent of the socio-demographic, health-related practices, and other systemic conditions related to MetS that were studied. Better allocation of resources is recommended, focusing on populations underweight and obese in need of dental care.

Frailty in motion: Amnestic mild cognitive impairment and Alzheimer's disease cohorts display heterogeneity in multimorbidity classification and longitudinal transitions

BackgroundData-driven examination of multiple morbidities and deficits are informative for clinical and research applications in aging and dementia. Resulting profiles may change longitudinally according to dynamic alterations in extent, duration, and pattern of risk accumulation. Do such frailty-related changes include not only progression but also stability and reversion?ObjectiveWith cognitively impaired and dementia cohorts, we employed data-driven analytics to (a) detect the extent of heterogeneity in frailty-related multimorbidity and deficit burden subgroups and (b) identify key person characteristics predicting differential transition patterns.MethodsWe assembled baseline and 2-year follow-up data from the National Alzheimer's Coordinating Center for amnestic mild cognitive impairment (aMCI) and Alzheimer's disease (AD) cohorts. We applied factor analyses to 43 multimorbidity and deficit indicators. Latent Transition Analysis (LTA) was applied to the resulting domains in order to detect subgroups differing in transition patterns for multimorbidity and deficit burden. We characterized heterogeneity in change patterns by evaluating key person characteristics as differential predictors.ResultsFactor analyses revealed five domains at two time points. LTA showed that two latent burden subgroups at Time 1 (Low, Moderate) differentiated into an additional two subgroups at Time 2 (adding Mild, Severe). Transition analyses detected heterogeneous changes, including progression, stability, and reversion. Baseline classifications and transitions varied according to clinical cohort, global cognition, sex, age, and education.ConclusionsHeterogeneous frailty-related subgroup transitions can be (a) detected in aging adults living with aMCI and AD, (b) characterized as not only progression but also stability and reversion, and (c) predicted by precision characteristics.