ABSTRACT
Background@#We used machine learning and population-based data for analyzing the determinants of sarcopenia in adult women and developing its decision support systems for various subgroups. @*Methods@#All data was acquired from the Korea National Health and Nutrition Examination Survey, and women 18 years and older were included in this research. The variables were selected based on female characteristics and the ability to be acquired in a survey format, and were ranked by importance using Random Forest. From this ranking, four main variables were selected, age, menopause age, menarche age and number of pregnancy. A decision supporting system was constructed based on a tree randomly selected from Random Forest. @*Results@#We defined sarcopenia as -2SD below the appendicular skeletal mass (ASM) index reference of 0.5136, and 89.87% (n = 8,610) were found non-sarcopenic and 10.13% (n = 971) were found sarcopenic. The subjects were divided into 6 groups based on menopausal status and BMI. The obese postmenopausal women had the highest number of sarcopenia, whereas the non-obese premenopausal women had the least number of sarcopenic subjects. In non-obese premenopausal women, which was considered to be at the lowest risk for sarcopenia, the most determining variable was the menarche age, followed by age and number of pregnancies. In obese and postmenopausal women, which was considered to be at the highest risk for sarcopenia, the most influential factor was the menopausal age, followed by age and menarche age. @*Conclusions@#We identified the major determinants of sarcopenia using machine learning and population-based data. This study demonstrated the strengths of the random forest as an effective decision support system for each stratified subgroup to find its own optimal cut-off points for the major variables of sarcopenia.
ABSTRACT
PURPOSE: Selenium plays a major role in the intracellular antioxidant system. The aim of this study was to determine whether a low serum selenium level is associated with poor neurological outcome for victims of cardiac arrest. METHODS: We enrolled consecutive patients who were admitted to the emergency intensive care unit (ICU) of a tertiary referral center for post-resuscitation care after cardiac arrest from May 2008 to April 2010. Data were collected with respect to demographic information, variables of cardiac arrest and resuscitation, and application of therapeutic hypothermia. We examined neurologic findings and measured serum selenium level at admission to the ICU. In addition, we also calculated severity scores. The Glasgow-Pittsburgh cerebral performance categories (CPCs) were used for evaluation of neurological outcome. According to the six-month CPCs, patients were divided into two groups: the good prognosis (CPC 1-2) group and the poor prognosis (CPC 3-5) group. We then compared data between the two groups. RESULTS: Among 52 enrolled patients, 17 were classified as the good prognosis group and 35 as the poor prognosis group. Glasgow coma scale (odds ratio [OR]=0.343, 95% confidence intervals [CI], 0.124-0.947, p=0.039), intact pupilary reflex (OR=0.045, 95% CI, 0.004-0.561, p=0.016), and serum selenium level (OR=0.959, 95% CI, 0.921-0.999, p=0.045) showed an independent association with poor neurological outcome for victims of cardiac arrest. CONCLUSION: Low serum selenium level showed an association with poor neurological outcome for victims of cardiac arrest.