No abstract

The SNP + 45 (rs2241766) of the AdipoQ gene is not associated with T2D neither in Yi nor in Han people in southwestern China.

2006 through 2015 were used to construct a cohort of prescription opioid users (N = 1,246,642). A total of 278 features (potential predictors) were derived from baseline data, including demographics, pain and mental conditions, physical comorbidities, exposure to other medications, concomitant medication uses with opioids, and opioid using behaviors. Opioid overdoses were defined as opioid poisoning diagnoses or naloxone administration, and for persons older than 50 without a history of heart or lung disease, a respiratory depression diagnosis recorded in the emergency department. We used stratified 10-fold cross-validation to choose the classifier with the best performance. Model performance was evaluated using sensitivity, specificity, and discrimination. As an alternative, traditional approaches based on logistic regression were also explored. Results: A total of 2,274 opioid overdose cases were identified. The boosted tree classifier outperformed other learning algorithms with a sensitivity of 0.70, a specificity of 0.77, and a c-statistic of 0.77. Logistic regressions achieved similar levels of performance. While the list and rank of the top prognostic features were not the same from the two approaches, early refills, total days' supply, concomitant use of antidepressants, concomitant use of antipsychotics, and total opioid claims were defined as the most significant prognostic features by both approaches. ConClusions: Given the nature of the extremely imbalanced data, even the best classifier produces moderate performance. However, identifying key prognostic features will help identify high risk patients. The prediction tool enumerates the risks of opioid overdoses and provides a potential explicit standard for clinicians when making individual patient prescribing and dispensing decisions.