Search

Mediation analysis allows the examination of effects of a third variable in the pathway between an exposure and an outcome. The general multiple mediation analysis method, proposed by Yu et al, improves traditional methods (eg, estimation of natural and controlled direct effects) to enable consideration of multiple mediators/confounders simultaneously and the use of linear and nonlinear predictive models for estimating mediation/confounding effects. In this paper, we extend the method for time-to-event outcomes and apply the method to explore the racial disparity in breast cancer survivals. Breast cancer is the most common cancer and the second leading cause of cancer death among women of all races. Despite improvement of survival rates of breast cancer in the US, a significant difference between white and black women remains. Previous studies have found that more advanced and aggressive tumors and less than optimal treatment may explain the lower survival rates for black women as compared to white women. Due to limitations of current analytic methods and the lack of comprehensive data sets, researchers have not been able to differentiate the relative effect each factor contributes to the overall racial disparity. We use the CDC-funded Patterns of Care study to examine the determinants of racial disparities in breast cancer survival using a novel multiple mediation analysis. Using the proposed method, we applied the Cox hazard model and multiple additive regression trees as predictive models and found that all racial disparity in survival among Louisiana breast cancer patients were explained by factors included in the study. [ABSTRACT FROM AUTHOR]

Drug-drug interactions (DDIs) are a common cause of adverse drug events (ADEs). The electronic medical record (EMR) database and the FDA's adverse event reporting system (FAERS) database are the major data sources for mining and testing the ADE associated DDI signals. Most DDI data mining methods focus on pair-wise drug interactions, and methods to detect high-dimensional DDIs in medical databases are lacking. In this paper, we propose 2 novel mixture drug-count response models for detecting high-dimensional drug combinations that induce myopathy. The "count" indicates the number of drugs in a combination. One model is called fixed probability mixture drug-count response model with a maximum risk threshold (FMDRM-MRT). The other model is called count-dependent probability mixture drug-count response model with a maximum risk threshold (CMDRM-MRT), in which the mixture probability is count dependent. Compared with the previous mixture drug-count response model (MDRM) developed by our group, these 2 new models show a better likelihood in detecting high-dimensional drug combinatory effects on myopathy. CMDRM-MRT identified and validated (54; 374; 637; 442; 131) 2-way to 6-way drug interactions, respectively, which induce myopathy in both EMR and FAERS databases. We further demonstrate FAERS data capture much higher maximum myopathy risk than EMR data do. The consistency of 2 mixture models' parameters and local false discovery rate estimates are evaluated through statistical simulation studies. [ABSTRACT FROM AUTHOR]

Aim: To develop and validate models based on logistic regression and artificial intelligence for prognostic prediction of molar survival in periodontally affected patients. Materials and Methods: Clinical and radiographic data from four different centres across four continents (two in Europe, one in the United States, and one in China) including 515 patients and 3157 molars were collected and used to train and test different types of machine‐learning algorithms for their prognostic ability of molar loss over 10 years. The following models were trained: logistic regression, support vector machine, K‐nearest neighbours, decision tree, random forest, artificial neural network, gradient boosting, and naive Bayes. In addition, different models were aggregated by means of the ensembled stacking method. The primary outcome of the study was related to the prediction of overall molar loss (MLO) in patients after active periodontal treatment. Results: The general performance in the external validation settings (aggregating three cohorts) revealed that the ensembled model, which combined neural network and logistic regression, showed the best performance among the different models for the prediction of MLO with an area under the curve (AUC) = 0.726. The neural network model showed the best AUC of 0.724 for the prediction of periodontitis‐related molar loss. In addition, the ensembled model showed the best calibration performance. Conclusions: Through a multi‐centre collaboration, both prognostic models for the prediction of molar loss were developed and externally validated. The ensembled model showed the best performance in terms of both discrimination and validation, and it is made freely available to clinicians for widespread use in clinical practice. [ABSTRACT FROM AUTHOR]

The joinpoint regression model (JRM) is used to describe trend changes in many applications and relies on the detection of joinpoints (changepoints). However, the existing joinpoint detection methods, namely, the grid search (GS)-based methods, are computationally demanding, and hence, the maximum number of computable joinpoints is limited. Herein, we developed a genetic algorithm-based joinpoint (GAJP) model in which an explicitly decoupled computing procedure for optimization and regression is used to embed a binary genetic algorithm into the JRM for optimal joinpoint detection. The combinations of joinpoints were represented as binary chromosomes, and genetic operations were performed to determine the optimum solution by minimizing the fitness function, the Bayesian information criterion (BIC) and BIC3 . The accuracy and computational performance of the GAJP model were evaluated via intensive simulation studies and compared with those of the GS-based methods using BIC, BIC3 , and permutation test. The proposed method showed an outstanding computational efficiency in detecting multiple joinpoints. Finally, the suitability of the GAJP model for the analysis of cancer incidence trends was demonstrated by applying this model to data on the incidence of colorectal cancer in the United States from 1975 to 2016 from the National Cancer Institute's Surveillance, Epidemiology, and End Results program. Thus, the GAJP model was concluded to be practically feasible to detect multiple joinpoints up to the number of grids without requirement to preassign the number of joinpoints and be easily extendable to cancer trend analysis utilizing large datasets. [ABSTRACT FROM AUTHOR]

Objective: Develop and validate a surveillance model to identify outpatient surgical adverse events (AEs) based on previously developed electronic triggers.Data Sources: Veterans Health Administration's Corporate Data Warehouse.Study Design: Six surgical AE triggers, including postoperative emergency room visits and hospitalizations, were applied to FY2012-2014 outpatient surgeries (n = 744,355). We randomly sampled trigger-flagged and unflagged cases for nurse chart review to document AEs and measured positive predictive value (PPV) for triggers. Next, we used chart review data to iteratively estimate multilevel logistic regression models to predict the probability of an AE, starting with the six triggers and adding in patient, procedure, and facility characteristics to improve model fit. We validated the final model by applying the coefficients to FY2015 outpatient surgery data (n = 256,690) and reviewing charts for cases at high and moderate probability of an AE.Principal Findings: Of 1,730 FY2012-2014 reviewed surgeries, 350 had an AE (20 percent). The final surveillance model c-statistic was 0.81. In FY2015 surgeries with >0.8 predicted probability of an AE (n = 405, 0.15 percent), PPV was 85 percent; in surgeries with a 0.4-0.5 predicted probability of an AE, PPV was 38 percent.Conclusions: The surveillance model performed well, accurately identifying outpatient surgeries with a high probability of an AE. [ABSTRACT FROM AUTHOR]

Objective: Electronic clinical quality measures (eCQMs) rely on computer algorithms to extract data from electronic health records (EHRs). On behalf of the American College of Rheumatology (ACR), we sought to develop and test eCQMs for rheumatoid arthritis (RA).Methods: Drawing from published ACR guidelines, a working group developed candidate RA process measures and subsequently assessed face validity through an interdisciplinary panel of health care stakeholders. A public comment period followed. Measures that passed these levels of review were electronically specified using the quality data model, which provides standard nomenclature for data elements (category, datatype, and value sets) obtained through an EHR. For each eCQM, 3 clinical sites using different EHR systems tested the scientific feasibility and validity of measures. Measures appropriate for accountability were presented for national endorsement.Results: Expert panel validity ratings were high for all measures (median 8-9 of 9). Health system performance on the eCQMs was 53.6% for RA disease activity assessment, 69.1% for functional status assessment, 93.1% for disease-modifying antirheumatic drug (DMARD) use, and 72.8% for tuberculosis screening. Kappa statistics, which evaluated whether the eCQM validly captured data obtained from manual EHR chart review, demonstrated moderate to substantial agreement (0.54 for functional status assessment, 0.73 for tuberculosis screening, 0.84 for disease activity, and 0.85 for DMARD use).Conclusion: Four eCQMs for RA have achieved national endorsement and are recommended for use in federal quality reporting programs. Implementation and further refinement of these measures is ongoing in the ACR's registry, the Rheumatology Informatics System for Effectiveness (RISE). [ABSTRACT FROM AUTHOR]