A flexible approach for variable selection in large-scale healthcare database studies with missing covariate and outcome data.

BACKGROUND: Prior work has shown that combining bootstrap imputation with tree-based machine learning variable selection methods can provide good performances achievable on fully observed data when covariate and outcome data are missing at random (MAR). This approach however is computationally expensive, especially on large-scale datasets. METHODS: We propose an inference-based method, called RR-BART, which leverages the likelihood-based Bayesian machine learning technique, Bayesian additive regression trees, and uses Rubin's rule to combine the estimates and variances of the variable importance measures on multiply imputed datasets for variable selection in the presence of MAR data. We conduct a representative simulation study to investigate the practical operating characteristics of RR-BART, and compare it with the bootstrap imputation based methods. We further demonstrate the methods via a case study of risk factors for 3-year incidence of metabolic syndrome among middle-aged women using data from the Study of Women's Health Across the Nation (SWAN). RESULTS: The simulation study suggests that even in complex conditions of nonlinearity and nonadditivity with a large percentage of missingness, RR-BART can reasonably recover both prediction and variable selection performances, achievable on the fully observed data. RR-BART provides the best performance that the bootstrap imputation based methods can achieve with the optimal selection threshold value. In addition, RR-BART demonstrates a substantially stronger ability of detecting discrete predictors. Furthermore, RR-BART offers substantial computational savings. When implemented on the SWAN data, RR-BART adds to the literature by selecting a set of predictors that had been less commonly identified as risk factors but had substantial biological justifications. CONCLUSION: The proposed variable selection method for MAR data, RR-BART, offers both computational efficiency and good operating characteristics and is utilitarian in large-scale healthcare database studies.

Comparison of statistical and machine learning models for healthcare cost data: a simulation study motivated by Oncology Care Model (OCM) data.

BACKGROUND: The Oncology Care Model (OCM) was developed as a payment model to encourage participating practices to provide better-quality care for cancer patients at a lower cost. The risk-adjustment model used in OCM is a Gamma generalized linear model (Gamma GLM) with log-link. The predicted value of expense for the episodes identified for our academic medical center (AMC), based on the model fitted to the national data, did not correlate well with our observed expense. This motivated us to fit the Gamma GLM to our AMC data and compare it with two other flexible modeling methods: Random Forest (RF) and Partially Linear Additive Quantile Regression (PLAQR). We also performed a simulation study to assess comparative performance of these methods and examined the impact of non-linearity and interaction effects, two understudied aspects in the field of cost prediction. METHODS: The simulation was designed with an outcome of cost generated from four distributions: Gamma, Weibull, Log-normal with a heteroscedastic error term, and heavy-tailed. Simulation parameters both similar to and different from OCM data were considered. The performance metrics considered were the root mean square error (RMSE), mean absolute prediction error (MAPE), and cost accuracy (CA). Bootstrap resampling was utilized to estimate the operating characteristics of the performance metrics, which were described by boxplots. RESULTS: RF attained the best performance with lowest RMSE, MAPE, and highest CA for most of the scenarios. When the models were misspecified, their performance was further differentiated. Model performance differed more for non-exponential than exponential outcome distributions. CONCLUSIONS: RF outperformed Gamma GLM and PLAQR in predicting overall and top decile costs. RF demonstrated improved prediction under various scenarios common in healthcare cost modeling. Additionally, RF did not require prespecification of outcome distribution, nonlinearity effect, or interaction terms. Therefore, RF appears to be the best tool to predict average cost. However, when the goal is to estimate extreme expenses, e.g., high cost episodes, the accuracy gained by RF versus its computational costs may need to be considered.

Clinical Outcomes and Associations With Radial to Femoral Crossover in ST-Elevation Myocardial Infarction.

Radial access during primary percutaneous coronary intervention is associated with reduced mortality and major bleeding compared with femoral access and is the recommended access site. Nevertheless, failure to secure radial access may necessitate crossover to femoral access. This study aimed to identify the associations with crossover from radial to femoral access in all comers with ST-elevation myocardial infarction and to compare the clinical outcomes with those patients who did not require crossover. From 2016 to 2021, a total of 1,202 patients presented to our institute with ST-elevation myocardial infarction. Associations, clinical outcomes, and independent predictors of crossover from radial to femoral access were identified. From 1,202 patients, radial access was used in 1,138 patients (94.7%) and crossover to femoral access occurred in 64 patients (5.3%). Patients who required crossover to femoral access had higher rates of access site complications and longer length of stay in the hospital. Inpatient mortality was higher in the group requiring a crossover. This study identified 3 independent predictors of crossover from radial to femoral access in primary percutaneous coronary intervention: cardiogenic shock, cardiac arrest before arrival at the catheterization laboratory, and previous coronary artery bypass grafting. Biochemical infarct size and peak creatinine was also found to be higher in those requiring crossover. In conclusion, crossover in this study portended an increased rate of access site complications, greatly prolonged length of stay, and a significantly higher risk of death.

Predictors of radiologists' perceived risk of malpractice lawsuits in breast imaging.

OBJECTIVE: The shortage of radiologists in breast imaging may be related to heightened malpractice lawsuit concerns. Our objective was to examine radiologists' reported experiences and perceptions of future lawsuit risk and explore personal and professional factors that may be associated with elevated perceptions of risk. MATERIALS AND METHODS: Radiologists who routinely interpret mammography examinations in diverse regions of the United States (Washington, Colorado, and New Hampshire) completed a mailed survey in 2002 and 2006, including questions on demographics, practice characteristics, and medicolegal experience and perceptions as well as a validated scale measuring reactions to uncertainty in clinical situations. A survey assessing the radiologists' work facilities was also completed in 2002. RESULTS: Participation by eligible radiologists was 77% (139 of 181) in 2002 and 71% (84 of 118) in 2006. The percentage of radiologists reporting malpractice claims related to mammography in the previous 5 years was 8% on the 2002 survey and 10% on the 2006 survey. Radiologists' mean estimate of the probability of being sued for malpractice in the next 5 years (41% in 2002 and 35% in 2006) was markedly higher than the actual reported risk. Radiologists' age, sex, clinical experience, and workload were not associated with a higher perceived risk of being sued. Radiologists who reported higher perceived risk of lawsuits were more likely to have experienced a prior malpractice claim, to report knowing colleagues with prior lawsuits, and to score higher on a scale measuring anxiety caused by uncertainty in clinical situations. Radiologists working at facilities that did not use double reading reported higher perceived risk, but the difference was not statistically significant. CONCLUSION: Radiologists working in breast imaging substantially overestimate their risk of a future malpractice lawsuit. Radiologists with higher risk perceptions show more negative reactions to uncertainty in a clinical setting. Understanding that their actual risk of malpractice lawsuits may be substantially lower than anticipated may help reduce radiologists' fears and alleviate the manpower shortage in mammography. Programs to address the shortage of breast imagers could be targeted toward radiologists with heightened malpractice lawsuit concerns.

How does anger coping style affect glycemic control in diabetes patients?

BACKGROUND: Although various forms of anger have been found to influence the psychological and physical health in many chronic illness populations, little is known about the effects of anger in diabetes patients. PURPOSE: Associations between anger coping style, diabetes-related psychological distress, and glycosylated hemoglobin (HbA1c) were examined in 100 diabetes patients. METHOD: Participants completed the Problem Areas in Diabetes and Coping Styles questionnaires, and had HbA1c assessments at study entry (Time 1 = T1), six months (T2), and 12 months after T1 (T3). RESULTS: Linear regression analyses revealed T1 anger coping associated with T3 HbA1c (beta = .22, p < .05), but T1 HbA1c did not associate with T3 anger coping (beta = .13, p = NS). After controlling for significant covariates (of gender, age, education, type and duration of diabetes), regression analyses revealed that T2 diabetes-related psychological distress partially mediated this association. CONCLUSION: These results suggested that higher levels of anger coping may promote poorer glycemic control in diabetes patients by provoking greater diabetes-related distress. Areas of future research on this topic are discussed.

The role of resilience on psychological adjustment and physical health in patients with diabetes.

OBJECTIVE: This study used a longitudinal design to investigate the buffering role of resilience on worsening HbA(1c) and self-care behaviours in the face of rising diabetes-related distress. METHOD: A total of 111 patients with diabetes completed surveys and had their glycosylated haemoglobin (HbA(1c)) assessed at baseline and at 1-year follow-up. Resilience was defined by a factor score of self-esteem, self-efficacy, self-mastery and optimism. Diabetes-related distress and self-care behaviours were also assessed. RESULTS: Baseline resilience, diabetes-related distress and their interaction predicted physical health (HbA(1c)) at 1 year. Patients with low, moderate and high resilience were identified. Those with low or moderate resilience levels showed a strong association between rising distress and worsening HbA(1c) across time (r=.57, .56, respectively). However, those with high resilience scores did not show the same associations (r=.08). Low resilience was also associated with fewer self-care behaviours when faced with increasing distress (r=-.55). These correlation coefficients remained significant after controlling for starting-points. CONCLUSION: In patients with diabetes, resilience resources predicted future HbA(1c) and buffered worsening HbA(1c) and self-care behaviours in the face of rising distress levels.

Reactions to uncertainty and the accuracy of diagnostic mammography.

BACKGROUND: Reactions to uncertainty in clinical medicine can affect decision making. OBJECTIVE: To assess the extent to which radiologists' reactions to uncertainty influence diagnostic mammography interpretation. DESIGN: Cross-sectional responses to a mailed survey assessed reactions to uncertainty using a well-validated instrument. Responses were linked to radiologists' diagnostic mammography interpretive performance obtained from three regional mammography registries. PARTICIPANTS: One hundred thirty-two radiologists from New Hampshire, Colorado, and Washington. MEASUREMENT: Mean scores and either standard errors or confidence intervals were used to assess physicians' reactions to uncertainty. Multivariable logistic regression models were fit via generalized estimating equations to assess the impact of uncertainty on diagnostic mammography interpretive performance while adjusting for potential confounders. RESULTS: When examining radiologists' interpretation of additional diagnostic mammograms (those after screening mammograms that detected abnormalities), a 5-point increase in the reactions to uncertainty score was associated with a 17% higher odds of having a positive mammogram given cancer was diagnosed during follow-up (sensitivity), a 6% lower odds of a negative mammogram given no cancer (specificity), a 4% lower odds (not significant) of a cancer diagnosis given a positive mammogram (positive predictive value [PPV]), and a 5% higher odds of having a positive mammogram (abnormal interpretation). CONCLUSION: Mammograms interpreted by radiologists who have more discomfort with uncertainty have higher likelihood of being recalled.

Psychophysiological mediators of caregiver stress and differential cognitive decline.

The authors examined relationships between chronic stress and cognitive decline and whether such relationships were mediated by psychophysiological factors. Ninety-six caregivers of spouses with Alzheimer's disease (AD) were compared with 95 similar noncaregiver spouses. All were free of diabetes. Although the groups started similarly, over 2 years caregivers declined by a small but significant amount (1 raw score point and 4 percentile points, each p<.05) on Shipley Vocabulary. In contrast, noncaregivers did not change. Higher hostile attribution (beta=-.09; p<.05) and metabolic risk (beta=-.10; p<.05) in caregivers mediated the cognitive decline. This is the first study of cognitive decline and mediators in caregivers. This work has implications for caregiver and care-recipient health and for research on cognition, psychophysiology, diabetes, and AD.

The confidence in diabetes self-care scale: psychometric properties of a new measure of diabetes-specific self-efficacy in Dutch and US patients with type 1 diabetes.

OBJECTIVE: To examine psychometric properties of the Confidence in Diabetes Self-Care (CIDS) scale, a newly developed instrument assessing diabetes-specific self-efficacy in Dutch and U.S. patients with type 1 diabetes. RESEARCH DESIGN AND METHODS: Reliability and validity of the CIDS scale were evaluated in Dutch (n = 151) and U.S. (n = 190) outpatients with type 1 diabetes. In addition to the CIDS scale, assessment included HbA(1c), emotional distress, fear of hypoglycemia, self-esteem, anxiety, depression, and self-care behavior. The Dutch sample completed additional measures on perceived burden and importance of self-care. Test-retest reliability was established in a second Dutch sample (n = 62). RESULTS: Internal consistency (Cronbach's alpha = 0.86 for Dutch patients and 0.90 U.S. patients) and test-retest reliability (Spearman's r = 0.85, P < 0.0001) of the CIDS scale were high. Exploratory factor analysis showed one strong general factor. Spearman's correlations between the CIDS scale and other measures were moderate and in the expected directions, and high HbA(1c) levels were associated with low CIDS scores in the U.S. sample only. Low CIDS scores were positively associated with self-care but not with glycemic control in the original samples. CIDS scores in the U.S. and Dutch samples did not show any statistically significant differences. U.S. men had higher CIDS scores than U.S. women. CONCLUSIONS: The CIDS scale is a reliable and valid measure of diabetes-specific self-efficacy for use in patients with type 1 diabetes. High psychometric similarity allows for cross-cultural comparisons.

Type 1 and Type 2 diabetes in Asian and Pacific Islander U.S. youth: the SEARCH for Diabetes in Youth Study.

OBJECTIVE: Given limited reports on diabetes among U.S. Asian and Pacific Islander youth, we describe the clinical characteristics, incidence, and prevalence of diabetes among Asian, Pacific Islander, and mixed Asian-Pacific Islander youth. RESEARCH DESIGN AND METHODS: Data were collected from 245 Asian, Pacific Islander, and Asian-Pacific Islander participants in the SEARCH for Diabetes in Youth Study, a population-based study of diabetes in youth (aged <20 years). Additionally, we estimated the incidence and prevalence of type 1 and type 2 diabetes for Asian, Pacific Islander, and Asian-Pacific Islander youth combined. RESULTS: Most participants with type 2 diabetes were obese (range Asian 71% to Pacific Islander 100%) with mean BMI >33 kg/m(2). In those with type 1 diabetes, Pacific Islanders were more likely to be obese, with a mean BMI of 26 vs. 20 kg/m(2) for Asian and Asian-Pacific Islander youth (P < 0.0001). The incidence of type 1 diabetes for youth aged 0-9 years was 6.4 per 100,000 person-years and 7.4 per 100,000 person-years for youth aged 10-19 years. The incidence of type 2 diabetes was 12.1 per 100,000 person-years for youth aged 10-19 years. CONCLUSIONS: While Asian and Asian-Pacific Islanders with type 1 and type 2 diabetes had lower mean BMIs than Pacific Islanders, all Asian, Pacific Islander, and Asian-Pacific Islanders with type 2 diabetes had mean BMIs above adult ethnicity-specific definitions of obesity. While the majority of Asian, Pacific Islander, and Asian-Pacific Islander youth had type 1 diabetes, older Asian, Pacific Islander, and Asian-Pacific Islander youth (aged 10-19 years) have an incidence of type 2 diabetes almost double that of type 1 diabetes. Public health efforts to prevent type 2 diabetes and obesity in Asian, Pacific Islander, and Asian-Pacific Islander adolescents are needed.

Stress-resilience, illness, and coping: a person-focused investigation of young women athletes.

Coping correlates of resilience, defined as resistance to illness in the face of exposure to high numbers of negative life events, were studied in a sample of 404 young women high school athletes. Negative life events and coping strategies were assessed preseason, and daily illness data were collected during the course of the season. Among athletes with high levels of exposure to negative life events, resilient (no illness time loss) and nonresilient (upper third of time loss distribution) groups were compared on 6 scales of the Ways of Coping Checklist. Coping profiles of the groups differed significantly, with resilient athletes favoring Problem-focused Coping and Seeking Social Support, and nonresilient athletes reporting greater use of Avoidance and Blaming Others. Correlations among Problem-focused Coping, Seeking Social Support, and Minimize Threat were higher in the resilient group. Results suggest that certain coping strategies may contribute to illness-resistance in the face of high life stress.