Variable selection with multiply-imputed datasets: choosing between stacked and grouped methods

Predicting implementation of response to intervention in math using elastic net logistic regression

Introduction

The primary objective of this study was to identify variables that significantly influence the implementation of math Response to Intervention (RTI) at the school level, utilizing the ECLS-K: 2011 dataset.

Methods

Due to missing values in the original dataset, a Random Forest algorithm was employed for data imputation, generating a total of 10 imputed datasets. Elastic net logistic regression, combined with nested cross-validation, was applied to each imputed dataset, potentially resulting in 10 models with different variables. Variables for the models derived from the imputed datasets were selected using four methods, leading to four candidate models for final selection. These models were assessed based on their performance of prediction accuracy, culminating in the selection of the final model that outperformed the others.

Results and discussion

Method50 and Methodcoef emerged as the most effective, achieving a balanced accuracy of 0.852. The ultimate model selected relevant variables that effectively predicted RTI. The predictive accuracy of the final model was also demonstrated by the receiver operating characteristic (ROC) plot and the corresponding area under the curve (AUC) value, indicating its ability to accurately forecast math RTI implementation in schools for the following year.

Intention to quit or reduce e-cigarettes, cannabis, and their co-use among a school-based sample of adolescents

INTRODUCTION

Little is known about the prevalence and predictors of adolescents' intention to quit or reduce use of e-cigarettes and/or cannabis.

METHODS

Frequencies of intention to change (quit, reduce) e-cigarettes and/or cannabis use were examined among 23,915 surveyed middle and high school students with sole and co-use. Predictors of intention to change were identified via LASSO/multilevel logistic regression.

RESULTS

Among those with sole e-cigarette use (n = 543), 40.9 % intended to quit and 24.1 % intended to reduce; non-daily e-cigarette use predicted intention to quit and reduce e-cigarettes (p's < 0.03). Among those with sole cannabis use (n = 546), 10.6 % intended to quit and 25.1 % intended to reduce; absence of cannabis cravings predicted intention to reduce cannabis use (p < 0.01). Among those with co-use (n = 816), 26.2 % intended to either quit or reduce (quit/reduce) both substances, 27.5 % intended to quit/reduce e-cigarettes only, and 6.9 % intended to quit/reduce cannabis only. No predictors emerged for intention to change e-cigarette use among those with co-use (p's > 0.09), but younger age, lack of poly-tobacco use, and lack of cannabis craving predicted intention to quit/reduce cannabis use (p's < 0.04).

CONCLUSIONS

More than half of adolescents with past-month e-cigarette use, regardless of concurrent cannabis use, expressed interest in changing their use. However, only heaviness of e-cigarette use emerged as a predictor of intention to change suggesting. While fewer students expressed interest in changing their cannabis use, cannabis cravings and poly-tobacco use predicted intent to change. Overall, findings emphasize the need to tailor interventions towards adolescents engaging in more problematic substance use patterns.

Hyperarousal dynamics reveal an overnight increase boosted by insomnia

Hyperarousal is a key symptom of anxiety, stress-related disorders, and insomnia. However, it has been conceptualized in many different ways, ranging from various physiological markers (e.g. cortisol levels, high-frequency EEG activity) to personality traits, or state assessments of subjective anxiety and tension. This approach resulted in partly inconsistent evidence, complicating unified interpretations. Crucially, no previous studies addressed the likely variability of hyperarousal within and across days, nor the relationship of such variability in hyperarousal with the night-by-night variability in sleep quality characteristic of insomnia. Here, we present a novel data-driven approach to understanding dynamics of state hyperarousal in insomnia. Using ecological momentary assessment, we tracked fluctuations in a wide range of emotions across 9 days in 169 people with insomnia disorders and 38 controls without sleep problems. Exploratory factor analysis identified a hyperarousal factor, comprised of items describing tension and distress. People with insomnia scored significantly higher on this factor than controls at all timepoints. In both groups, the hyperarousal factor score peaked in the morning and waned throughout the day, pointing to a potential contributing role of sleep or other circadian processes. Importantly, the overnight increase in hyperarousal was stronger in people with in insomnia than in controls. Subsequent adaptive LASSO regression analysis revealed a stronger overnight increase in hyperarousal across nights of worse subjective sleep quality. These findings demonstrate the relationship between subjective sleep quality and overnight modulations of hyperarousal. Disorders in which hyperarousal is a predominant complaint might therefore benefit from interventions focused on improving sleep quality.

Multiple metal exposures associate with higher amyotrophic lateral sclerosis risk and mortality independent of genetic risk and correlate to self-reported exposures: a case-control study

BACKGROUND

The pathogenesis of amyotrophic lateral sclerosis (ALS) involves both genetic and environmental factors. This study investigates associations between metal measures in plasma and urine, ALS risk and survival and exposure sources.

METHODS

Participants with and without ALS from Michigan provided plasma and urine samples for metal measurement via inductively coupled plasma mass spectrometry. ORs and HRs for each metal were computed using risk and survival models. Environmental risk scores (ERS) were created to evaluate the association between exposure mixtures and ALS risk and survival and exposure source. ALS (ALS-PGS) and metal (metal-PGS) polygenic risk scores were constructed from an independent genome-wide association study and relevant literature-selected single-nucleotide polymorphisms.

RESULTS

Plasma and urine samples from 454 ALS and 294 control participants were analysed. Elevated levels of individual metals, including copper, selenium and zinc, significantly associated with ALS risk and survival. ERS representing metal mixtures strongly associated with ALS risk (plasma, OR=2.95, CI=2.38-3.62, p<0.001; urine, OR=3.10, CI=2.43-3.97, p<0.001) and poorer ALS survival (plasma, HR=1.37, CI=1.20-1.58, p<0.001; urine, HR=1.44, CI=1.23-1.67, p<0.001). Addition of the ALS-PGS or metal-PGS did not alter the significance of metals with ALS risk and survival. Occupations with high potential of metal exposure associated with elevated ERS. Additionally, occupational and non-occupational metal exposures were associated with measured plasma and urine metals.

CONCLUSION

Metals in plasma and urine associated with increased ALS risk and reduced survival, independent of genetic risk, and correlated with occupational and non-occupational metal exposures. These data underscore the significance of metal exposure in ALS risk and progression.

Factors associated with lower quarter performance-based balance and strength tests: a cross-sectional analysis from the project baseline health study

Objectives

Physical performance tests are predictive of mortality and may screen for certain health conditions (e.g., sarcopenia); however, their diagnostic and/or prognostic value has primarily been studied in age-limited or disease-specific cohorts. Our objective was to identify the most salient characteristics associated with three lower quarter balance and strength tests in a cohort of community-dwelling adults.

Methods

We applied a stacked elastic net approach on detailed data on sociodemographic, health and health-related behaviors, and biomarker data from the first visit of the Project Baseline Health Study (N = 2,502) to determine which variables were most associated with three physical performance measures: single-legged balance test (SLBT), sitting-rising test (SRT), and 30-second chair-stand test (30CST). Analyses were stratified by age (<65 and ≥65).

Results

Female sex, Black or African American race, lower educational attainment, and health conditions such as non-alcoholic fatty liver disease and cardiovascular conditions (e.g., hypertension) were consistently associated with worse performance across all three tests. Several other health conditions were associated with either better or worse test performance, depending on age group and test. C-reactive protein was the only laboratory value associated with performance across age and test groups with some consistency.

Conclusions

Our results highlighted previously identified and several novel salient factors associated with performance on the SLBT, SRT, and 30CST. These tests could represent affordable, noninvasive biomarkers of prevalent and/or future disease in adult individuals; future research should validate these findings.

Clinical Trial Registration

ClinicalTrials.gov, identifier NCT03154346, registered on May 15, 2017.

Agreement between 30-day and 90-day modified Rankin Scale score and utility-weighted modified Rankin Scale score in acute intracerebral hemorrhage: An analysis of ATACH-2 trial data

The relationship between 30- and 90-day modified Rankin Scale (mRS) scores in intracerebral hemorrhage (ICH) patients was evaluated. This post hoc cohort analysis of the ATACH-2 trial included patients with acute ICH who were alive at 30 days and who had mRS scores reported at 30 and 90 days. The mRS score was then converted to a utility (EuroQol-5 Dimension-3 Level [EQ-5D-3L])-weighted mRS score. After adjustment of 30-day mRS score for key covariates using multivariable ordinal regression, the relationship between 30-day and observed 90-day functional outcome was assessed via absolute difference in the utility-weighted version. Of the 1000 trial subjects, 898 met inclusion criteria. This low-moderate severity ICH cohort had a median baseline GCS score of 15 and median hematoma volume of 9.7 mL. Observed 30-day mRS had the largest association with observed 90-day values (χ2 = 302.9, p < 0.0001). Patients generally either maintained the same mRS scores between 30 and 90 days (48 %) or experienced a 1-point (32 %) or 2-point (10 %) improvement by 90 days. The mean ± standard deviation (SD) EQ-5D-3L at 90 days was 0.67 ± 0.26. Following adjustment, the mean absolute difference between predicted and observed utility-weighted 90-day mRS scores was 0.006 ± 0.13 points and less than the estimated minimal clinically important difference of 0.13 points. The difference in average utility-weighted mRS scores at 30 and 90 days was not clinically relevant, suggesting 30-day score may be a reasonable proxy for 90-day values in patients with ICH when 90-day values are not available.

Metal mixtures associate with higher amyotrophic lateral sclerosis risk and mortality independent of genetic risk and correlate to self-reported exposures: a case-control study

Background

The pathogenesis of amyotrophic lateral sclerosis (ALS) involves both genetic and environmental factors. This study investigates associations between metal measures in plasma and urine, ALS risk and survival, and exposure sources.

Methods

Participants with and without ALS from Michigan provided plasma and urine samples for metal measurement via inductively coupled plasma mass spectrometry. Odds and hazard ratios for each metal were computed using risk and survival models. Environmental risk scores (ERS) were created to evaluate the association between exposure mixtures and ALS risk and survival and exposure source. ALS (ALS-PGS) and metal (metal-PGS) polygenic risk scores were constructed from an independent genome-wide association study and relevant literature-selected SNPs.

Results

Plasma and urine samples from 454 ALS and 294 control participants were analyzed. Elevated levels of individual metals, including copper, selenium, and zinc, significantly associated with ALS risk and survival. ERS representing metal mixtures strongly associated with ALS risk (plasma, OR=2.95, CI=2.38-3.62, p<0.001; urine, OR=3.10, CI=2.43-3.97, p<0.001) and poorer ALS survival (plasma, HR=1.42, CI=1.24-1.63, p<0.001; urine, HR=1.52, CI=1.31-1.76, p<0.001). Addition of the ALS-PGS or metal-PGS did not alter the significance of metals with ALS risk and survival. Occupations with high potential of metal exposure associated with elevated ERS. Additionally, occupational and non-occupational metal exposures associated with measured plasma and urine metals.

Conclusion

Metals in plasma and urine associated with increased ALS risk and reduced survival, independent of genetic risk, and correlated with occupational and non-occupational metal exposures. These data underscore the significance of metal exposure in ALS risk and progression.

Environmental risk scores of persistent organic pollutants associate with higher ALS risk and shorter survival in a new Michigan case/control cohort

BACKGROUND

Amyotrophic lateral sclerosis (ALS) is a fatal, progressive neurogenerative disease caused by combined genetic susceptibilities and environmental exposures. Identifying and validating these exposures are of paramount importance to modify disease risk. We previously reported that persistent organic pollutants (POPs) associate with ALS risk and survival and aimed to replicate these findings in a new cohort.

METHOD

Participants with and without ALS recruited in Michigan provided plasma samples for POPs analysis by isotope dilution with triple quadrupole mass spectrometry. ORs for risk models and hazard ratios for survival models were calculated for individual POPs. POP mixtures were represented by environmental risk scores (ERS), a summation of total exposures, to evaluate the association with risk (ERSrisk) and survival (ERSsurvival).

RESULTS

Samples from 164 ALS and 105 control participants were analysed. Several individual POPs significantly associated with ALS, including 8 of 22 polychlorinated biphenyls and 7 of 10 organochlorine pesticides (OCPs). ALS risk was most strongly represented by the mixture effects of OCPs alpha-hexachlorocyclohexane, hexachlorobenzene, trans-nonachlor and cis-nonachlor and an interquartile increase in ERSrisk enhanced ALS risk 2.58 times (p<0.001). ALS survival was represented by the combined mixture of all POPs and an interquartile increase in ERSsurvival enhanced ALS mortality rate 1.65 times (p=0.008).

CONCLUSIONS

These data continue to support POPs as important factors for ALS risk and progression and replicate findings in a new cohort. The assessments of POPs in non-Michigan ALS cohorts are encouraged to better understand the global effect and the need for targeted disease risk reduction strategies.

Multi-omics regulatory network inference in the presence of missing data

A key problem in systems biology is the discovery of regulatory mechanisms that drive phenotypic behaviour of complex biological systems in the form of multi-level networks. Modern multi-omics profiling techniques probe these fundamental regulatory networks but are often hampered by experimental restrictions leading to missing data or partially measured omics types for subsets of individuals due to cost restrictions. In such scenarios, in which missing data is present, classical computational approaches to infer regulatory networks are limited. In recent years, approaches have been proposed to infer sparse regression models in the presence of missing information. Nevertheless, these methods have not been adopted for regulatory network inference yet. In this study, we integrated regression-based methods that can handle missingness into KiMONo, a Knowledge guided Multi-Omics Network inference approach, and benchmarked their performance on commonly encountered missing data scenarios in single- and multi-omics studies. Overall, two-step approaches that explicitly handle missingness performed best for a wide range of random- and block-missingness scenarios on imbalanced omics-layers dimensions, while methods implicitly handling missingness performed best on balanced omics-layers dimensions. Our results show that robust multi-omics network inference in the presence of missing data with KiMONo is feasible and thus allows users to leverage available multi-omics data to its full extent.

Derivation of a Multivariable Psoriatic Arthritis Risk Estimation Tool (PRESTO): A Step Towards Prevention

OBJECTIVE

A simple, scalable tool that identifies psoriasis patients at high risk for developing psoriatic arthritis (PsA) could improve early diagnosis. We aimed to develop a risk prediction model for the development of PsA and to assess its performance among patients with psoriasis.

METHODS

We analyzed data from a prospective cohort of psoriasis patients without PsA at enrollment. Participants were assessed annually by a rheumatologist for the development of PsA. Information about their demographics, psoriasis characteristics, comorbidities, medications, and musculoskeletal symptoms was used to develop prediction models for PsA. Penalized binary regression models were used for variable selection while adjusting for psoriasis duration. Risks of developing PsA over 1- and 5-year time periods were estimated. Model performance was assessed by the area under the curve (AUC) and calibration plots.

RESULTS

Among 635 psoriasis patients, 51 and 71 developed PsA during the 1-year and 5-year follow-up periods, respectively. The risk of developing PsA within 1 year was associated with younger age, male sex, family history of psoriasis, back stiffness, nail pitting, joint stiffness, use of biologic medications, patient global health, and pain severity (AUC 72.3). The risk of developing PsA within 5 years was associated with morning stiffness, psoriatic nail lesion, psoriasis severity, fatigue, pain, and use of systemic nonbiologic medication or phototherapy (AUC 74.9). Calibration plots showed reasonable agreement between predicted and observed probabilities.

CONCLUSIONS

The development of PsA within clinically meaningful time frames can be predicted with reasonable accuracy for psoriasis patients using readily available clinical variables.

Immune Cell Infiltration Analysis Based on Bioinformatics Reveals Novel Biomarkers of Coronary Artery Disease

Background

Coronary artery disease (CAD) is a multifactorial immune disease, but research into the specific immune mechanism is still needed. The present study aimed to identify novel immune-related markers of CAD.

Methods

Three CAD-related datasets (GSE12288, GSE98583, GSE113079) were downloaded from the Gene Expression Integrated Database. Gene ontology annotation, Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis and weighted gene co-expression network analysis were performed on the common significantly differentially expressed genes (DEGs) of these three data sets, and the most relevant module genes for CAD obtained. The immune cell infiltration of module genes was evaluated with the CIBERSORT algorithm, and characteristic genes accompanied by their diagnostic effectiveness were screened by the machine-learning algorithm least absolute shrinkage and selection operator (LASSO) regression analysis. The expression levels of characteristic genes were evaluated in the peripheral blood mononuclear cells of CAD patients and healthy controls for verification.

Results

A total of 204 upregulated and 339 downregulated DEGs were identified, which were mainly enriched in the following pathways: "Apoptosis", "Th17 cell differentiation", "Th1 and Th2 cell differentiation", "Glycerolipid metabolism", and "Fat digestion and absorption". Five characteristic genes, LMAN1L, DOK4, CHFR, CEL and CCDC28A, were identified by LASSO analysis, and the results of the immune cell infiltration analysis indicated that the proportion of immune infiltrating cells (activated CD8 T cells and CD56 DIM natural killer cells) in the CAD group was lower than that in the control group. The expressions of CHFR, CEL and CCDC28A in the peripheral blood of the healthy controls and CAD patients were significantly different.

Conclusion

We identified CHFR, CEL and CCDC28A as potential biomarkers related to immune infiltration in CAD based on public data sets and clinical samples. This finding will contribute to providing a potential target for early noninvasive diagnosis and immunotherapy of CAD.

How to apply variable selection machine learning algorithms with multiply imputed data: A missing discussion

Psychological researchers often use standard linear regression to identify relevant predictors of an outcome of interest, but challenges emerge with incomplete data and growing numbers of candidate predictors. Regularization methods like the LASSO can reduce the risk of overfitting, increase model interpretability, and improve prediction in future samples; however, handling missing data when using regularization-based variable selection methods is complicated. Using listwise deletion or an ad hoc imputation strategy to deal with missing data when using regularization methods can lead to loss of precision, substantial bias, and a reduction in predictive ability. In this tutorial, we describe three approaches for fitting a LASSO when using multiple imputation to handle missing data and illustrate how to implement these approaches in practice with an applied example. We discuss implications of each approach and describe additional research that would help solidify recommendations for best practices. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Key Factors Associated With Pulmonary Sequelae in the Follow-Up of Critically Ill COVID-19 Patients

INTRODUCTION

Critical COVID-19 survivors have a high risk of respiratory sequelae. Therefore, we aimed to identify key factors associated with altered lung function and CT scan abnormalities at a follow-up visit in a cohort of critical COVID-19 survivors.

METHODS

Multicenter ambispective observational study in 52 Spanish intensive care units. Up to 1327 PCR-confirmed critical COVID-19 patients had sociodemographic, anthropometric, comorbidity and lifestyle characteristics collected at hospital admission; clinical and biological parameters throughout hospital stay; and, lung function and CT scan at a follow-up visit.

RESULTS

The median [p₂₅-p₇₅] time from discharge to follow-up was 3.57 [2.77-4.92] months. Median age was 60 [53-67] years, 27.8% women. The mean (SD) percentage of predicted diffusing lung capacity for carbon monoxide (D_LCO) at follow-up was 72.02 (18.33)% predicted, with 66% of patients having D_LCO<80% and 24% having D_LCO<60%. CT scan showed persistent pulmonary infiltrates, fibrotic lesions, and emphysema in 33%, 25% and 6% of patients, respectively. Key variables associated with D_LCO<60% were chronic lung disease (CLD) (OR: 1.86 (1.18-2.92)), duration of invasive mechanical ventilation (IMV) (OR: 1.56 (1.37-1.77)), age (OR [per-1-SD] (95%CI): 1.39 (1.18-1.63)), urea (OR: 1.16 (0.97-1.39)) and estimated glomerular filtration rate at ICU admission (OR: 0.88 (0.73-1.06)). Bacterial pneumonia (1.62 (1.11-2.35)) and duration of ventilation (NIMV (1.23 (1.06-1.42), IMV (1.21 (1.01-1.45)) and prone positioning (1.17 (0.98-1.39)) were associated with fibrotic lesions.

CONCLUSION

Age and CLD, reflecting patients' baseline vulnerability, and markers of COVID-19 severity, such as duration of IMV and renal failure, were key factors associated with impaired D_LCO and CT abnormalities.

Machine learning-based scoring system to predict in-hospital outcomes in patients hospitalized with COVID-19

BACKGROUND

The evolution of patients hospitalized with coronavirus disease 2019 (COVID-19) is still hard to predict, even after several months of dealing with the pandemic.

AIMS

To develop and validate a score to predict outcomes in patients hospitalized with COVID-19.

METHODS

All consecutive adults hospitalized for COVID-19 from February to April 2020 were included in a nationwide observational study. Primary composite outcome was transfer to an intensive care unit from an emergency department or conventional ward, or in-hospital death. A score that estimates the risk of experiencing the primary outcome was constructed from a derivation cohort using stacked LASSO (Least Absolute Shrinkage and Selection Operator), and was tested in a validation cohort.

RESULTS

Among 2873 patients analysed (57.9% men; 66.6±17.0 years), the primary outcome occurred in 838 (29.2%) patients: 551 (19.2%) were transferred to an intensive care unit; and 287 (10.0%) died in-hospital without transfer to an intensive care unit. Using stacked LASSO, we identified 11 variables independently associated with the primary outcome in multivariable analysis in the derivation cohort (n=2313), including demographics (sex), triage vitals (body temperature, dyspnoea, respiratory rate, fraction of inspired oxygen, blood oxygen saturation) and biological variables (pH, platelets, C-reactive protein, aspartate aminotransferase, estimated glomerular filtration rate). The Critical COVID-19 France (CCF) risk score was then developed, and displayed accurate calibration and discrimination in the derivation cohort, with C-statistics of 0.78 (95% confidence interval 0.75-0.80). The CCF risk score performed significantly better (i.e. higher C-statistics) than the usual critical care risk scores.

CONCLUSIONS

The CCF risk score was built using data collected routinely at hospital admission to predict outcomes in patients with COVID-19. This score holds promise to improve early triage of patients and allocation of healthcare resources.