BACKGROUND: Health care-associated infections due to multidrug-resistant organisms (MDROs), such as methicillin-resistant Staphylococcus aureus (MRSA) and Clostridioides difficile (CDI), place a significant burden on our health care infrastructure.
OBJECTIVE: Screening for MDROs is an important mechanism for preventing spread but is resource intensive. The objective of this study was to develop automated tools that can predict colonization or infection risk using electronic health record (EHR) data, provide useful information to aid infection control, and guide empiric antibiotic coverage.
METHODS: We retrospectively developed a machine learning model to detect MRSA colonization and infection in undifferentiated patients at the time of sample collection from hospitalized patients at the University of Virginia Hospital. We used clinical and nonclinical features derived from on-admission and throughout-stay information from the patient\'s EHR data to build the model. In addition, we used a class of features derived from contact networks in EHR data; these network features can capture patients\' contacts with providers and other patients, improving model interpretability and accuracy for predicting the outcome of surveillance tests for MRSA. Finally, we explored heterogeneous models for different patient subpopulations, for example, those admitted to an intensive care unit or emergency department or those with specific testing histories, which perform better.
RESULTS: We found that the penalized logistic regression performs better than other methods, and this model\'s performance measured in terms of its receiver operating characteristics-area under the curve score improves by nearly 11% when we use polynomial (second-degree) transformation of the features. Some significant features in predicting MDRO risk include antibiotic use, surgery, use of devices, dialysis, patient\'s comorbidity conditions, and network features. Among these, network features add the most value and improve the model\'s performance by at least 15%. The penalized logistic regression model with the same transformation of features also performs better than other models for specific patient subpopulations.
CONCLUSIONS: Our study shows that MRSA risk prediction can be conducted quite effectively by machine learning methods using clinical and nonclinical features derived from EHR data. Network features are the most predictive and provide significant improvement over prior methods. Furthermore, heterogeneous prediction models for different patient subpopulations enhance the model\'s performance.

The National Mental Health Survey of India (NMHS) was a ground-breaking nationwide study that harnessed a uniform, standardized methodology blending quantitative and qualitative approaches. Covering data from 12 states across diverse regions, its mission was to gauge the prevalence of psychiatric disorders, bridge treatment gaps, explore service utilization, and gauge the socioeconomic repercussions of these conditions. This initiative provided pivotal insights into the intricate landscape of mental health in India. One of the analyses planned for NMHS data was to undertake a logistic regression analysis with an aim to unravel how various sociodemographic factors influence the presence or absence of specific psychiatric disorders. Within this pursuit, two substantial challenges loomed. The first pertained to data separation, a complication that could perturb parameter estimation. The second challenge stemmed from the existence of disorders with lower prevalence rates, which resulted in datasets of limited density, potentially undermining the statistical reliability of our analysis. In response to these data-driven hurdles, NMHS recognized the critical necessity for an alternative to conventional logistic regression, one that could adeptly navigate these complexities, ensuring robust and dependable insights from the collected data. Traditional logistic regression, a widely prevalent method for modeling binary outcomes, has its limitations, especially when faced with limited datasets and rare outcomes. Here, the problem of \"complete separation\" can lead to convergence failure in traditional logistic regression estimations, a conundrum frequently encountered when handling binary variables. Firth\'s penalized logistic regression emerges as a potent solution to these challenges, effectively mitigating analytical biases rooted in small sample sizes, rare events, and complete separation. This article endeavors to illuminate the superior efficacy of Firth\'s method in managing small datasets within scientific research and advocates for its more widespread application. We provide a succinct introduction to Firth\'s method, emphasizing its distinct advantages over alternative analytical approaches and underscoring its application to data from the NMHS 2015-2016, particularly for disorders with lower prevalence.

Cancer classification and gene selection are important applications in DNA microarray gene expression data analysis. Since DNA microarray data suffers from the high-dimensionality problem, automatic gene selection methods are used to enhance the classification performance of expert classifier systems. In this paper, a new penalized logistic regression method that performs simultaneous gene coefficient estimation and variable selection in DNA microarray data is discussed. The method employs prior information about the gene coefficients to improve the classification accuracy of the underlying model. The coordinate descent algorithm with screening rules is given to obtain the gene coefficient estimates of the proposed method efficiently. The performance of the method is examined on five high-dimensional cancer classification datasets using the area under the curve, the number of selected genes, misclassification rate and F-score measures. The real data analysis results indicate that the proposed method achieves a good cancer classification performance with a small misclassification rate, large area under the curve and F-score by trading off some sparsity level of the underlying model. Hence, the proposed method can be seen as a reliable penalized logistic regression method in the scope of high-dimensional cancer classification.

For finite samples with binary outcomes penalized logistic regression such as ridge logistic regression has the potential of achieving smaller mean squared errors (MSE) of coefficients and predictions than maximum likelihood estimation. There is evidence, however, that ridge logistic regression can result in highly variable calibration slopes in small or sparse data situations.
In this paper, we elaborate this issue further by performing a comprehensive simulation study, investigating the performance of ridge logistic regression in terms of coefficients and predictions and comparing it to Firth\'s correction that has been shown to perform well in low-dimensional settings. In addition to tuned ridge regression where the penalty strength is estimated from the data by minimizing some measure of the out-of-sample prediction error or information criterion, we also considered ridge regression with pre-specified degree of shrinkage. We included \'oracle\' models in the simulation study in which the complexity parameter was chosen based on the true event probabilities (prediction oracle) or regression coefficients (explanation oracle) to demonstrate the capability of ridge regression if truth was known.
Performance of ridge regression strongly depends on the choice of complexity parameter. As shown in our simulation and illustrated by a data example, values optimized in small or sparse datasets are negatively correlated with optimal values and suffer from substantial variability which translates into large MSE of coefficients and large variability of calibration slopes. In contrast, in our simulations pre-specifying the degree of shrinkage prior to fitting led to accurate coefficients and predictions even in non-ideal settings such as encountered in the context of rare outcomes or sparse predictors.
Applying tuned ridge regression in small or sparse datasets is problematic as it results in unstable coefficients and predictions. In contrast, determining the degree of shrinkage according to some meaningful prior assumptions about true effects has the potential to reduce bias and stabilize the estimates.

Ramp metering relieves traffic congestion, reduces delay, and maintains the capacity flow on freeways. Due to its operational mechanism, ramp metering can also improve freeway safety. While the operational benefits of ramp metering have extensively been quantified, research on its safety effects is sparse. This study focused on evaluating the effects of ramp metering on the safety performance of the freeway mainline. It developed a crash risk prediction model for segments downstream of the entrance ramps when ramp metering is activated. The study was based on a corridor with system-wide ramp metering along I-95 in Miami, Florida. Real-time traffic, crash, and ramp metering operations data collected from 2016 to 2018 were used in the analysis. The study adopted a matched crash and non-crash case approach to evaluate the crash risk when ramp meters were activated and deactivated. A penalized logistic regression model was developed using a bootstrap resampling technique to estimate the effects of ramp metering activation and select important variables that could predict crash risk when ramp meters were activated. Results indicated that ramp metering improves safety along the freeway corridor by reducing the crash risk downstream of the entrance ramps. During ramp metering activation, the crash risk on segments downstream of the entrance ramps 5 min later can be predicted using the difference in the average lane speeds between upstream and downstream detectors, the average traffic volume in the lanes at the downstream and upstream detectors, and the coefficient of variation of speed between lanes in the upstream detectors. Also, the coefficient of variation of occupancy downstream could predict the crash risk 15 min later. The study results could be used by transportation agencies when evaluating the deployment of ramp meters. Moreover, the developed crash risk prediction model could be used in real-time to help agencies identify the increased crash risk and provide appropriate warning information to the upstream traffic.

Adolescent onset of depression is associated with long-lasting negative consequences. Identifying adolescents at risk for developing depression would enable the monitoring of risk factors and the development of early intervention strategies. Using machine learning to combine several risk factors from multiple modalities might allow prediction of depression onset at the individual level.
A subsample of a multisite longitudinal study in adolescents, the IMAGEN study, was used to predict future (subthreshold) major depressive disorder onset in healthy adolescents. Based on 2-year and 5-year follow-up data, participants were grouped into the following: 1) those developing a diagnosis of major depressive disorder or subthreshold major depressive disorder and 2) healthy control subjects. Baseline measurements of 145 variables from different modalities (clinical, cognitive, environmental, and structural magnetic resonance imaging) at age 14 years were used as input to penalized logistic regression (with different levels of penalization) to predict depression onset in a training dataset (n = 407). The features contributing the highest to the prediction were validated in an independent hold-out sample (three independent IMAGEN sites; n = 137).
The area under the receiver operating characteristic curve for predicting depression onset ranged between 0.70 and 0.72 in the training dataset. Baseline severity of depressive symptoms, female sex, neuroticism, stressful life events, and surface area of the supramarginal gyrus contributed most to the predictive model and predicted onset of depression, with an area under the receiver operating characteristic curve between 0.68 and 0.72 in the independent validation sample.
This study showed that depression onset in adolescents can be predicted based on a combination multimodal data of clinical characteristics, life events, personality traits, and brain structure variables.

Analysis approaches for single compositional data are well established; however, effective analysis strategies for paired compositional data remain to be investigated. The current project was motivated by studies of age-related hearing loss (presbyacusis), where subjects are classified into four audiometric phenotypes that need to be ranked within these phenotypes based on their paired compositional data. We address this challenge by formulating this problem as a classification problem and integrating a penalized multinomial logistic regression model with compositional data analysis approaches. We utilize Elastic Net for a penalty function, while considering average, absolute difference, and perturbation operators for compositional data. We applied the proposed approach to the presbyacusis study of 532 subjects with probabilities that each ear of a subject belongs to each of four presbyacusis subtypes. We further investigated the ranking of presbyacusis subjects using the proposed approach based on previous literature. The data analysis results indicate that the proposed approach is effective for ranking subjects based on paired compositional data.

This study evaluated prediction performance of three different machine learning (ML) techniques in predicting opioid misuse among U.S. adolescents. Data were drawn from the 2015-2017 National Survey on Drug Use and Health (N = 41,579 adolescents, ages 12-17 years) and analyzed in 2019. Prediction models were developed using three ML algorithms, including artificial neural networks, distributed random forest, and gradient boosting machine. The performance of the ML prediction models was compared with performance of the penalized logistic regression. The area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC) were used as metrics of prediction performance. We used the AUPRC as the primary measure of prediction performance given that it is considered more informative for assessing binary classifiers on imbalanced outcome variable than AUROC. The overall rate of opioid misuse among U.S. adolescents was 3.7% (n = 1521). Prediction performance was similar across the four models (AUROC values range from 0.809 to 0.815). In terms of the AUPRC, the distributed random forest showed the best performance in prediction (0.172) followed by penalized logistic regression (0.162), gradient boosting machine (0.160), and artificial neural networks (0.157). Findings suggest that machine learning techniques can be a promising technique especially in the prediction of outcomes with rare cases (i.e., when the binary outcome variable is heavily lopsided) such as adolescent opioid misuse.

A high-dimensional quantitative structure-activity relationship (QSAR) classification model typically contains a large number of irrelevant and redundant descriptors. In this paper, a new design of descriptor selection for the QSAR classification model estimation method is proposed by adding a new weight inside L1-norm. The experimental results of classifying the anti-hepatitis C virus activity of thiourea derivatives demonstrate that the proposed descriptor selection method in the QSAR classification model performs effectively and competitively compared with other existing penalized methods in terms of classification performance on both the training and the testing datasets. Moreover, it is noteworthy that the results obtained in terms of stability test and applicability domain provide a robust QSAR classification model. It is evident from the results that the developed QSAR classification model could conceivably be employed for further high-dimensional QSAR classification studies.

Discovering important genes that account for the phenotype of interest has long been a challenge in genome-wide expression analysis. Analyses such as gene set enrichment analysis (GSEA) that incorporate pathway information have become widespread in hypothesis testing, but pathway-based approaches have been largely absent from regression methods due to the challenges of dealing with overlapping pathways and the resulting lack of available software. The R package grpreg is widely used to fit group lasso and other group-penalized regression models; in this study, we develop an extension, grpregOverlap, to allow for overlapping group structure using a latent variable approach. We compare this approach to the ordinary lasso and to GSEA using both simulated and real data. We find that incorporation of prior pathway information can substantially improve the accuracy of gene expression classifiers, and we shed light on several ways in which hypothesis-testing approaches such as GSEA differ from regression approaches with respect to the analysis of pathway data.