Comparison of model feature importance statistics to identify covariates that contribute most to model accuracy in prediction of insomnia.

IMPORTANCE: Sleep is critical to a person's physical and mental health and there is a need to create high performing machine learning models and critically understand how models rank covariates. OBJECTIVE: The study aimed to compare how different model metrics rank the importance of various covariates. DESIGN, SETTING, AND PARTICIPANTS: A cross-sectional cohort study was conducted retrospectively using the National Health and Nutrition Examination Survey (NHANES), which is publicly available. METHODS: This study employed univariate logistic models to filter out strong, independent covariates associated with sleep disorder outcome, which were then used in machine-learning models, of which, the most optimal was chosen. The machine-learning model was used to rank model covariates based on gain, cover, and frequency to identify risk factors for sleep disorder and feature importance was evaluated using both univariable and multivariable t-statistics. A correlation matrix was created to determine the similarity of the importance of variables ranked by different model metrics. RESULTS: The XGBoost model had the highest mean AUROC of 0.865 (SD = 0.010) with Accuracy of 0.762 (SD = 0.019), F1 of 0.875 (SD = 0.766), Sensitivity of 0.768 (SD = 0.023), Specificity of 0.782 (SD = 0.025), Positive Predictive Value of 0.806 (SD = 0.025), and Negative Predictive Value of 0.737 (SD = 0.034). The model metrics from the machine learning of gain and cover were strongly positively correlated with one another (r > 0.70). Model metrics from the multivariable model and univariable model were weakly negatively correlated with machine learning model metrics (R between -0.3 and 0). CONCLUSION: The ranking of important variables associated with sleep disorder in this cohort from the machine learning models were not related to those from regression models.

Perimacular Atrophy Following Voretigene Neparvovec-Rzyl Treatment in the Setting of Previous Contralateral Eye Treatment With a Different Viral Vector.

Purpose: To report on cases of unilateral perimacular atrophy after treatment with voretigene neparvovec-rzyl, in the setting of previous contralateral eye treatment with a different viral vector. Design: Single-center, retrospective chart review. Methods: In this case series, four patients between the ages of six and 11 years old with RPE65-related retinopathy were treated unilaterally with rAAV2-CB-hRPE65 as part of a gene augmentation clinical trial (NCT00749957). Six to 10 years later the contralateral eyes were treated with the Food and Drug Administration-approved drug, voretigene neparvovec-rzyl. Best-corrected visual acuity (BCVA), fundus photos, ocular coherence tomography, two-color dark-adapted perimetry, full field stimulus threshold testing (FST), and location of subretinal bleb and chorioretinal atrophy were evaluated. Results: Three out of four patients showed unilateral perimacular atrophy after treatment with voretigene, ranging from five to 22 months after treatment. Areas of robust visual field improvement were followed by areas of chorioretinal atrophy. Despite perimacular changes, BCVA, FST, and subjective improvements in vision and nyctalopia were maintained. Perimacular atrophy was not observed in the first eye treated with the previous viral vector. Conclusions: We observed areas of robust visual field improvement followed by perimacular atrophy in voretigene treated eyes, as compared to the initially treated contralateral eyes. Translational Relevance: Caution is advised when using two different viral vectors between eyes in gene therapy. This may become an important issue in the future with increasing gene therapy clinical trials for inherited retinal dystrophies.

Genomic loss of GPR108 disrupts AAV transduction in birds.

The G protein-coupled receptor 108 (GPR108) gene encodes a protein factor identified as critical for adeno-associated virus (AAV) entry into mammalian cells, but whether it is universally involved in AAV transduction is unknown. Remarkably, we have discovered that GPR108 is absent in the genomes of birds and in most other sauropsids, providing a likely explanation for the overall lower AAV transduction efficacy of common AAV serotypes in birds compared to mammals. Importantly, transgenic expression of human GPR108 and manipulation of related glycan binding sites in the viral capsid significantly boost AAV transduction in zebra finch cells. These findings contribute to a more in depth understanding of the mechanisms and evolution of AAV transduction, with potential implications for the design of efficient tools for gene manipulation in experimental animal models, and a range of gene therapy applications in humans.

Application of a transparent artificial intelligence algorithm for US adults in the obese category of weight.

OBJECTIVE AND AIMS: Identification of associations between the obese category of weight in the general US population will continue to advance our understanding of the condition and allow clinicians, providers, communities, families, and individuals make more informed decisions. This study aims to improve the prediction of the obese category of weight and investigate its relationships with factors, ultimately contributing to healthier lifestyle choices and timely management of obesity. METHODS: Questionnaires that included demographic, dietary, exercise and health information from the US National Health and Nutrition Examination Survey (NHANES 2017-2020) were utilized with BMI 30 or higher defined as obesity. A machine learning model, XGBoost predicted the obese category of weight and Shapely Additive Explanations (SHAP) visualized the various covariates and their feature importance. Model statistics including Area under the receiver operator curve (AUROC), sensitivity, specificity, positive predictive value, negative predictive value and feature properties such as gain, cover, and frequency were measured. SHAP explanations were created for transparent and interpretable analysis. RESULTS: There were 6,146 adults (age > 18) that were included in the study with average age 58.39 (SD = 12.94) and 3122 (51%) females. The machine learning model had an Area under the receiver operator curve of 0.8295. The top four covariates include waist circumference (gain = 0.185), GGT (gain = 0.101), platelet count (gain = 0.059), AST (gain = 0.057), weight (gain = 0.049), HDL cholesterol (gain = 0.032), and ferritin (gain = 0.034). CONCLUSION: In conclusion, the utilization of machine learning models proves to be highly effective in accurately predicting the obese category of weight. By considering various factors such as demographic information, laboratory results, physical examination findings, and lifestyle factors, these models successfully identify crucial risk factors associated with the obese category of weight.

The contribution of low Apgar scores in identifying neonates with short-term morbidities in a large single center cohort.

OBJECTIVE: To evaluate the association and utility of low 1- and 5-min Apgar scores to identify short-term morbidities in a large newborn cohort. METHODS: 15,542 infants >22 weeks gestation from a single center were included. Clinical data and low Apgar scores were analyzed for significance to ten short-term outcomes and were used to construct Receiver Operating Characteristic Curves and the AUC calculated for ten outcomes. RESULTS: A low Apgar score related to all (1-min) or most (5-min) outcomes by univariate and multivariate logistic regression analysis. Including any of the 4 low Apgar scores only improved the clinical factor AUC by 0.9% ± 2.7% (±SD) and was significant in just 5 of the 40 score/outcome scenarios. CONCLUSION: The contribution of a low Apgar score for identifying risk of short-term morbidity does not appear to be clinically significant.

Stochastic modeling of obesity status in United States adults using Markov Chains: A nationally representative analysis of population health data from 2017-2020.

Importance: The prevalence of obesity among United States adults has increased from 34.9% in 2013-2014 to 42.8% in 2017-2018. Developing methods to model the increase of obesity over-time is a necessity to know how to accurately quantify its cost and to develop solutions to combat this national public health emergency. Methods: A cross-sectional cohort study using the publicly available National Health and Nutrition Examination Survey (NHANES 2017-2020) was conducted in individuals who completed the weight questionnaire and had accurate data for both weight at the time of survey and weight 10 years ago. To model the dynamics of obesity, a Markov transition state matrix was created, which allowed for the analysis of weight transitions over time. Bootstrap simulation was incorporated to account for uncertainty and generate multiple simulated datasets, providing a more robust estimation of the prevalence and trends in obesity within the cohort. Results: Of the 6146 individuals who met the inclusion criteria, 3024 (49%) individuals were male and 3122 (51%) were female. There were 2252 (37%) White individuals, 1257 (20%) Hispanic individuals, 1636 (37%) Black individuals, and 739 (12%) Asian individuals. The average BMI was 30.16 (SD = 7.15), the average weight was 83.67 kilos (SD = 22.04), and the average weight change was a 3.27 kg (SD = 14.97) increase in body weight. A total of 2411 (39%) individuals lost weight, and 3735 (61%) individuals gained weight. 87 (1%) individuals were underweight (BMI <18.5), 2058 (33%) were normal weight (18.5 ≤ BMI <25), 1376 (22%) were overweight (25 ≤ BMI <30) and 2625 (43%) were in the obese category (BMI >30). Conclusion: United States adults are at risk of transitioning from normal weight to the overweight or obese category. Markov modeling combined with bootstrap simulations can accurately model long-term weight status.

Covariate dependent Markov chains constructed with gradient boost modeling can effectively generate long-term predictions of obesity trends.

IMPORTANCE: The prevalence of obesity among United States adults has increased from 30.5% in 1999 to 41.9% in 2020. However, despite the recognition of long-term weight gain as an important public health issue, there is a paucity of studies studying the long-term weight gain and building models for long-term projection. METHODS: A retrospective, cross-sectional cohort study using the publicly available National Health and Nutrition Examination Survey (NHANES 2017-2020) was conducted in patients who completed the weight questionnaire and had accurate data for both weight at time of survey and weight ten years ago. Multistate gradient boost modeling classifiers were used to generate covariate dependent transition matrices and Markov chains were utilized for multistate modeling. RESULTS: Of the 6146 patients that met the inclusion criteria, 3024 (49%) of patients were male and 3122 (51%) of patients were female. There were 2252 (37%) White patients, 1257 (20%) Hispanic patients, 1636 (37%) Black patients, and 739 (12%) Asian patients. The average BMI was 30.16 (SD = 7.15), the average weight was 83.67 kilos (SD = 22.04), and the average weight change was a 3.27 kg (SD = 14.97) increase in body weight (Fig. 1). A total of 2411 (39%) patients lost weight, and 3735 (61%) patients gained weight (Table 1). We observed that 87 (1%) of patients were underweight (BMI < 18.5), 2058 (33%) were normal weight (18.5 ≤ BMI < 25), 1376 (22%) were overweight (25 ≤ BMI < 30) and 2625 (43%) were obese (BMI > 30). From analysis of the transitions between normal/underweight, overweight, and obese, we observed that after 10 years, of the patients who were underweight, 65% stayed underweight, 32% became normal weight, 2% became overweight, and 2% became obese. After 10 years, of the patients who were normal weight, 3% became underweight, 78% stayed normal weight, 17% became overweight, and 2% became obese. Of the patients who were overweight, 71% stayed overweight, 0% became underweight, 14% became normal weight, and 15% became obese. Of the patients who were obese, 84% stayed obese, 0% became underweight, 1% became normal weight, and 14% became overweight. CONCLUSIONS: United States adults are at risk of transitioning from normal weight to becoming overweight or obese. Covariate dependent Markov chains constructed with gradient boost modeling can effectively generate long-term predictions.

Shapely additive values can effectively visualize pertinent covariates in machine learning when predicting hypertension.

Machine learning methods are widely used within the medical field to enhance prediction. However, little is known about the reliability and efficacy of these models to predict long-term medical outcomes such as blood pressure using lifestyle factors, such as diet. The authors assessed whether machine-learning techniques could accurately predict hypertension risk using nutritional information. A cross-sectional study using data from the National Health and Nutrition Examination Survey (NHANES) between January 2017 and March 2020. XGBoost was used as the machine-learning model of choice in this study due to its increased performance relative to other common methods within medical studies. Model prediction metrics (e.g., AUROC, Balanced Accuracy) were used to measure overall model efficacy, covariate Gain statistics (percentage each covariate contributes to the overall prediction) and SHapely Additive exPlanations (SHAP, method to visualize each covariate) were used to provide explanations to machine-learning output and increase the transparency of this otherwise cryptic method. Of a total of 9650 eligible patients, the mean age was 41.02 (SD = 22.16), 4792 (50%) males, 4858 (50%) female, 3407 (35%) White patients, 2567 (27%) Black patients, 2108 (22%) Hispanic patients, and 981 (10%) Asian patients. From evaluation of model gain statistics, age was found to be the single strongest predictor of hypertension, with a gain of 53.1%. Additionally, demographic factors such as poverty and Black race were also strong predictors of hypertension, with gain of 4.33% and 4.18%, respectively. Nutritional Covariates contributed 37% to the overall prediction: Sodium, Caffeine, Potassium, and Alcohol intake being significantly represented within the model. Machine Learning can be used to predict hypertension.

Technical Report: Machine-Learning Pipeline for Medical Research and Quality-Improvement Initiatives.

Machine-learning techniques have been increasing in popularity within medicine during the past decade. However, these computational techniques are not presented in statistical lectures throughout medical school and are perceived to have a high barrier to entry. The objective is to develop a concise pipeline with publicly available data to decrease the learning time towards using machine learning for medical research and quality-improvement initiatives. This report utilized a publicly available machine-learning data package in R (MLDataR) and computational packages (XGBoost) to highlight techniques for machine-learning model development and visualization with SHaply Additive exPlanations (SHAP). A simple six-step process along with example code was constructed to build and visualize machine-learning models. A concrete set of three steps was developed to help with interpretation. Further teaching of these methods could benefit researchers by providing alternative methods for data analysis in medical studies. These could help researchers without computational experience to get a feel for machine learning to better understand the literature and technique.

Use of feature importance statistics to accurately predict asthma attacks using machine learning: A cross-sectional cohort study of the US population.

BACKGROUND: Asthma attacks are a major cause of morbidity and mortality in vulnerable populations, and identification of associations with asthma attacks is necessary to improve public awareness and the timely delivery of medical interventions. OBJECTIVE: The study aimed to identify feature importance of factors associated with asthma in a representative population of US adults. METHODS: A cross-sectional analysis was conducted using a modern, nationally representative cohort, the National Health and Nutrition Examination Surveys (NHANES 2017-2020). All adult patients greater than 18 years of age (total of 7,922 individuals) with information on asthma attacks were included in the study. Univariable regression was used to identify significant nutritional covariates to be included in a machine learning model and feature importance was reported. The acquisition and analysis of the data were authorized by the National Center for Health Statistics Ethics Review Board. RESULTS: 7,922 patients met the inclusion criteria in this study. The machine learning model had 55 out of a total of 680 features that were found to be significant on univariate analysis (P<0.0001 used). In the XGBoost model the model had an Area Under the Receiver Operator Characteristic Curve (AUROC) = 0.737, Sensitivity = 0.960, NPV = 0.967. The top five highest ranked features by gain, a measure of the percentage contribution of the covariate to the overall model prediction, were Octanoic Acid intake as a Saturated Fatty Acid (SFA) (gm) (Gain = 8.8%), Eosinophil percent (Gain = 7.9%), BMXHIP-Hip Circumference (cm) (Gain = 7.2%), BMXHT-standing height (cm) (Gain = 6.2%) and HS C-Reactive Protein (mg/L) (Gain 6.1%). CONCLUSION: Machine Learning models can additionally offer feature importance and additional statistics to help identify associations with asthma attacks.

Exploring Depression and Nutritional Covariates Amongst US Adults using Shapely Additive Explanations.

Background: Depression affects personal and public well-being and identification of natural therapeutics such as nutrition is necessary to help alleviate this public health concern. Objective: The study aimed to identify feature importance in a machine learning model using solely nutrition covariates. Methods: A retrospective analysis was conducted using a modern, nationally representative cohort, the National Health and Nutrition Examination Surveys (NHANES 2017-2020). Depressive symptoms were evaluated using the validated 9-item Patient Health Questionnaire (PHQ-9), and all adult patients (total of 7929 individuals) who completed the PHQ-9 and total nutritional intake questionnaire were included in the study. Univariable regression was used to identify significant nutritional covariates to be included in a machine learning model and feature importance was reported. The acquisition and analysis of the data were authorized by the National Center for Health Statistics Ethics Review Board. Results: 7929 patients met the inclusion criteria in this study. The machine learning model had 24 out of a total of 60 features that were found to be significant on univariate analysis (p < 0.01 used). In the XGBoost model the model had an Area Under the Receiver Operator Characteristic Curve (AUROC) = 0.603, Sensitivity = 0.943, Specificity = 0.163. The top four highest ranked features by gain, a measure of the percentage contribution of the covariate to the overall model prediction, were Potassium Intake (Gain = 6.8%), Vitamin E Intake (Gain = 5.7%), Number of Foods and Beverages Reported (Gain = 5.7%), and Vitamin K Intake (Gain 5.6%). Conclusion: Machine learning models with feature importance can be utilized to identify nutritional covariates for further study in patients with clinical symptoms of depression.

Quantification of the Relationship of Pyridoxine and Spirometry Measurements in the United States Population.

Background: There has been evidence to suggest associations between vitamins and lung function. Objective: This study aimed to examine the association between vitamin B6 and spirometry values. Methods: A cross-sectional study was done using National Health and Nutritional Examination Surveys (NHANES) 2007-2012, which is a nationally representative, modern cohort. Spirometry, a clinical pulmonary function test, measured the amount and speed of air a person could exhale after taking the deepest possible breath after forceful expiratory volume at 1 s (FEV1) and forced vital capacity (FVC). After determination of the relationship of the linearity of variables, univariable and multivariable models were fitted to investigate the effect of vitamin B6 on FEV1 and FVC. The National Center for Health Statistics Ethics Review Board granted permission for the study's data collection and analysis. Results: Of 19,160 individuals who had complete information on vitamin B6 intake, FEV1, and FVC, it was found each mg of vitamin B6 intake was associated with increase in 166.41 mL of FEV1 (95% CI: 156.71, 176.12; P < 0.01) and 221.6 mL of FVC (95% CI: 209.62, 233.57; P < 0.01). After controlling for potential confounders (age, race, sex, body mass index, education, and income), multiple linear regression found that each mg of vitamin B6 was associated with increase in 25.98 mL of FEV1 (95% CI: 19.15, 32.80, P < 0.01) and 38.97 mL of FVC (95% CI: 30.65, 47.30, P < 0.01). Conclusion: Increased vitamin B6 intake is associated with improvement in lung function. Further prospective studies are required to ascertain whether increased vitamin B6 can lead to increased long-term spirometry measurements and the specific therapeutic dose-response relationship.

Increased vigorous exercise and decreased sedentary activities are associated with decreased depressive symptoms in United States adults: Analysis of The National Health and Nutrition Examination Survey (NHANES) 2017-2020.

Background and aims: Depression is a major public health concern that affects over 4% of the global population. Identification of new nonpharmacologic recommendations will help decrease the burden of disease. The overarching of this study was to examine the association between physical activity and depressive symptoms in a large sample of adults in the United States. Methods: Presently, researchers utilized data from the National Health and Nutrition Examination Surveys (NHANES 2017-2020), which is a retrospective, complex, multistage, representative, and modern cohort of the United States. Adult patients ( > 18 years; N = 8091) with complete 9-item Patient Health Questionnaire (PHQ-9) information were included in the study. The PHQ-9 is a well-validated survey, per literature, scores ≥10 are considered to have clinically relevant depression. Univariable and multivariable logistic regression was fit for active and sedentary activities on clinical depression (PHQ-9 ≥ 10). The acquisition and analysis of the data within this study were approved by the National Center for Health Statistics Ethics Review Board. Results: After adjusting for potential confounders like age, race, sex, and income, we found that increased vigorous exercise was associated with lower rates of depressive symptoms. Each extra day of vigorous exercise was associated with 11% decreased odds of depression (odd ratio [OR]: 0.89, confidence interval [CI]: 0.83-0.96, p < 0.01). Increased sedentary activity was associated with increased depression. Each extra hour per day of sedentary activity was associated with a 6% increase in odds of depression (OR: 1.06, (1.02-1.10, p < 0.01). Conclusion: To conclude, exercise appears to be protective against depressive symptoms; however, further prospective studies are required to ascertain whether exercise causes decreased depressive symptoms.

Dendrogram of transparent feature importance machine learning statistics to classify associations for heart failure: A reanalysis of a retrospective cohort study of the Medical Information Mart for Intensive Care III (MIMIC-III) database.

BACKGROUND: There is a continual push for developing accurate predictors for Intensive Care Unit (ICU) admitted heart failure (HF) patients and in-hospital mortality. OBJECTIVE: The study aimed to utilize transparent machine learning and create hierarchical clustering of key predictors based off of model importance statistics gain, cover, and frequency. METHODS: Inclusion criteria of complete patient information for in-hospital mortality in the ICU with HF from the MIMIC-III database were randomly divided into a training (n = 941, 80%) and test (n = 235, 20%). A grid search was set to find hyperparameters. Machine Learning with XGBoost were used to predict mortality followed by feature importance with Shapely Additive Explanations (SHAP) and hierarchical clustering of model metrics with a dendrogram and heat map. RESULTS: Of the 1,176 heart failure ICU patients that met inclusion criteria for the study, 558 (47.5%) were males. The mean age was 74.05 (SD = 12.85). XGBoost model had an area under the receiver operator curve of 0.662. The highest overall SHAP explanations were urine output, leukocytes, bicarbonate, and platelets. Average urine output was 1899.28 (SD = 1272.36) mL/day with the hospital mortality group having 1345.97 (SD = 1136.58) mL/day and the group without hospital mortality having 1986.91 (SD = 1271.16) mL/day. The average leukocyte count in the cohort was 10.72 (SD = 5.23) cells per microliter. For the hospital mortality group the leukocyte count was 13.47 (SD = 7.42) cells per microliter and for the group without hospital mortality the leukocyte count was 10.28 (SD = 4.66) cells per microliter. The average bicarbonate value was 26.91 (SD = 5.17) mEq/L. Amongst the group with hospital mortality the average bicarbonate value was 24.00 (SD = 5.42) mEq/L. Amongst the group without hospital mortality the average bicarbonate value was 27.37 (SD = 4.98) mEq/L. The average platelet value was 241.52 platelets per microliter. For the group with hospital mortality the average platelet value was 216.21 platelets per microliter. For the group without hospital mortality the average platelet value was 245.47 platelets per microliter. Cluster 1 of the dendrogram grouped the temperature, platelets, urine output, Saturation of partial pressure of Oxygen (SPO2), Leukocyte count, lymphocyte count, bicarbonate, anion gap, respiratory rate, PCO2, BMI, and age as most similar in having the highest aggregate gain, cover, and frequency metrics. CONCLUSION: Machine Learning models that incorporate dendrograms and heat maps can offer additional summaries of model statistics in differentiating factors between in patient ICU mortality in heart failure patients.

Diabetes is associated with increased risk of death in COVID-19 hospitalizations in Mexico 2020: A retrospective cohort study.

Background and Aim: The COVID-19 disease course can be thought of as a function of prior risk factors consisting of comorbidities and outcomes. Survival analysis data for diabetic patients with COVID-19 from an up to date and representative sample can increase efficiency in resource allocation. The study aimed to quantify mortality in Mexico for individuals with diabetes in the setting of COVID-19 hospitalization. Methods: This retrospective cohort study utilized publicly available data from the Mexican Federal Government, covering the period from April 14, 2020, to December 20, 2020 (last accessed). Survival analysis techniques were applied, including Kaplan-Meier curves to estimate survival probabilities, log-rank tests to compare survival between groups, Cox proportional hazard models to assess the association between diabetes and mortality risk, and restricted mean survival time (RMST) analyses to measure the average survival time. Results: A total of 402,388 adults age greater than 18 with COVID-19 were used in the analysis. Mean age = 16.16 (SD = 15.55), 214,161 males (53%). Twenty-day Kaplan-Meier estimates of mortality were 32% for COVID-19 patients with diabetes and 10.2% for those without diabetes with log-rank p < 0.01. Univariable analysis showed increased mortality in diabetic patients (hazard ratio [HR]: 3.61, 95% confidence interval [CI]: 3.54-3.67, p < 0.01) showing a 254% increase in death. After controlling for confounding variables, multivariate analysis continued to show increased mortality in diabetics (HR: 1.37, 95% CI: 1.29-1.44, p < 0.01) indicating a 37% increase in death. Multivariable RMST at Day 20 showed in Mexico, hospitalized COVID-19 patients were associated with less mean survival time by 2.01 days (p < 0.01) and a 10% increased mortality (p < 0.01). Conclusions: In the present analysis, COVID-19 patients with diabetes in Mexico had shorter survival times. Further interventions aimed at improving comorbidities in the population, particularly in individuals with diabetes, may contribute to better outcomes in COVID-19 patients.

Quantification of the Effect of Vitamin E Intake on Depressive Symptoms in United States Adults Using Restricted Cubic Splines.

Background: Depression is a rapidly increasing public health concern, affecting >4% of the global population. Identification of new nutritional recommendations is needed to help combat this increasing public health concern. Objectives: The study aimed to examine the association between vitamin E intake and depressive symptoms. Methods: A retrospective study was conducted by using a nationally representative, modern cohort (NHANES 2017-2020). Depressive symptoms were assessed through the validated 9-item Patient Health Questionnaire (PHQ-9). All adult patients ([≥18 y old], 8091 total adults) who answered the PHQ-9 and daily nutritional values questionnaires were selected for this study. Per literature, patients with PHQ-9 scores ≥10 were considered to have depressive symptoms. Univariable and multivariable logistic regressions were used to investigate the effect of vitamin E on depressive symptoms as ascertained by PHQ-9. The acquisition and analysis of the data within this study was approved by the NCHS ethics review board. Results: After controlling for potential confounders (age, race, sex, and income), we observed that increased vitamin E (up until 15 mg/d) was associated with decreased rates of depressive symptoms, with each 5 mg increase in vitamin E associated with 13% decreased odds of symptoms of depression (OR: 0.87; 95% CI: 0.77, 0.97; P < 0.01). Additional intake above 15 mg/d, the daily recommended amount by the Food and Nutrition Board, did not change the odds of depression (OR: 1.05; 95% CI: 0.92, 1.16; P = 0.44). Conclusions: Increased vitamin E intake (up to 15 mg/d) is associated with decreased depressive symptoms. Further prospective studies are required to ascertain whether increased vitamin E can protect against depressive symptoms and the specific therapeutic dose-response relationship.

Computation of the distribution of model accuracy statistics in machine learning: Comparison between analytically derived distributions and simulation-based methods.

Background and Aims: All fields have seen an increase in machine-learning techniques. To accurately evaluate the efficacy of novel modeling methods, it is necessary to conduct a critical evaluation of the utilized model metrics, such as sensitivity, specificity, and area under the receiver operator characteristic curve (AUROC). For commonly used model metrics, we proposed the use of analytically derived distributions (ADDs) and compared it with simulation-based approaches. Methods: A retrospective cohort study was conducted using the England National Health Services Heart Disease Prediction Cohort. Four machine learning models (XGBoost, Random Forest, Artificial Neural Network, and Adaptive Boost) were used. The distribution of the model metrics and covariate gain statistics were empirically derived using boot-strap simulation (N = 10,000). The ADDs were created from analytic formulas from the covariates to describe the distribution of the model metrics and compared with those of bootstrap simulation. Results: XGBoost had the most optimal model having the highest AUROC and the highest aggregate score considering six other model metrics. Based on the Anderson-Darling test, the distribution of the model metrics created from bootstrap did not significantly deviate from a normal distribution. The variance created from the ADD led to smaller SDs than those derived from bootstrap simulation, whereas the rest of the distribution remained not statistically significantly different. Conclusions: ADD allows for cross study comparison of model metrics, which is usually done with bootstrapping that rely on simulations, which cannot be replicated by the reader.

Use of machine learning to identify risk factors for insomnia.

IMPORTANCE: Sleep is critical to a person's physical and mental health, but there are few studies systematically assessing risk factors for sleep disorders. OBJECTIVE: The objective of this study was to identify risk factors for a sleep disorder through machine-learning and assess this methodology. DESIGN, SETTING, AND PARTICIPANTS: A retrospective, cross-sectional cohort study using the publicly available National Health and Nutrition Examination Survey (NHANES) was conducted in patients who completed the demographic, dietary, exercise, and mental health questionnaire and had laboratory and physical exam data. METHODS: A physician diagnosis of insomnia was the outcome of this study. Univariate logistic models, with insomnia as the outcome, were used to identify covariates that were associated with insomnia. Covariates that had a p<0.0001 on univariate analysis were included within the final machine-learning model. The machine learning model XGBoost was used due to its prevalence within the literature as well as its increased predictive accuracy in healthcare prediction. Model covariates were ranked according to the cover statistic to identify risk factors for insomnia. Shapely Additive Explanations (SHAP) were utilized to visualize the relationship between these potential risk factors and insomnia. RESULTS: Of the 7,929 patients that met the inclusion criteria in this study, 4,055 (51% were female, 3,874 (49%) were male. The mean age was 49.2 (SD = 18.4), with 2,885 (36%) White patients, 2,144 (27%) Black patients, 1,639 (21%) Hispanic patients, and 1,261 (16%) patients of another race. The machine learning model had 64 out of a total of 684 features that were found to be significant on univariate analysis (P<0.0001 used). These were fitted into the XGBoost model and an AUROC = 0.87, Sensitivity = 0.77, Specificity = 0.77 were observed. The top four highest ranked features by cover, a measure of the percentage contribution of the covariate to the overall model prediction, were the Patient Health Questionnaire depression survey (PHQ-9) (Cover = 31.1%), age (Cover = 7.54%), physician recommendation of exercise (Cover = 3.86%), weight (Cover = 2.99%), and waist circumference (Cover = 2.70%). CONCLUSION: Machine learning models can effectively predict risk for a sleep disorder using demographic, laboratory, physical exam, and lifestyle covariates and identify key risk factors.

Improved Rod Sensitivity as Assessed by Two-Color Dark-Adapted Perimetry in Patients With RPE65-Related Retinopathy Treated With Voretigene Neparvovec-rzyl.

Purpose: The purpose of this study was to evaluate rod-mediated function with two-color dark-adapted perimetry (2cDAP) in patients with RPE65-related retinopathy treated with voretigene neparvovec-rzyl. Methods: Following dilation and dark adaptation, 2cDAP and FST were performed. The 2cDAP was measured on an Octopus 900 perimeter (Haag-Streit) with cyan (500 nm wavelength) and red (650 nm wavelength) stimuli. Hill of vision (HOV) analysis was performed on 2cDAP perimetry with Visual Field Modeling and Analysis (VFMA). Full field threshold stimulus testing (FST) was also measured as a secondary measure of rod-mediated function, and assessed on a Diagnosys Espion with the ColorDome stimulator (Diagnosys LLC). Results: Eight eyes from 4 patients who were treated with voretigene bilaterally had rod function assessed by 2cDAP testing at least 1 year after treatment. There was statistically significant improvement in 2cDAP following gene augmentation therapy. HOV VFMA analysis showed widespread improvements that extended beyond the treatment bleb and statistically significant improvement in HOV analysis volumetric measurements post-treatment to cyan and red stimuli. FST testing performed in six eyes from three patients demonstrated statistically significant improvement to all chromatic stimuli following treatment. Conclusions: These findings demonstrated statistically significant improvement in 2cDAP and FST following treatment with voretigene. Translational Relevance: These findings provide a sensitive method of assessing rod-mediated function in a topographic manner that may be useful in future clinical trials for inherited retinal dystrophies.

Use of machine learning to identify risk factors for coronary artery disease.

Coronary artery disease (CAD) is the leading cause of death in both developed and developing nations. The objective of this study was to identify risk factors for coronary artery disease through machine-learning and assess this methodology. A retrospective, cross-sectional cohort study using the publicly available National Health and Nutrition Examination Survey (NHANES) was conducted in patients who completed the demographic, dietary, exercise, and mental health questionnaire and had laboratory and physical exam data. Univariate logistic models, with CAD as the outcome, were used to identify covariates that were associated with CAD. Covariates that had a p<0.0001 on univariate analysis were included within the final machine-learning model. The machine learning model XGBoost was used due to its prevalence within the literature as well as its increased predictive accuracy in healthcare prediction. Model covariates were ranked according to the Cover statistic to identify risk factors for CAD. Shapely Additive Explanations (SHAP) explanations were utilized to visualize the relationship between these potential risk factors and CAD. Of the 7,929 patients that met the inclusion criteria in this study, 4,055 (51%) were female, 2,874 (49%) were male. The mean age was 49.2 (SD = 18.4), with 2,885 (36%) White patients, 2,144 (27%) Black patients, 1,639 (21%) Hispanic patients, and 1,261 (16%) patients of other race. A total of 338 (4.5%) of patients had coronary artery disease. These were fitted into the XGBoost model and an AUROC = 0.89, Sensitivity = 0.85, Specificity = 0.87 were observed (Fig 1). The top four highest ranked features by cover, a measure of the percentage contribution of the covariate to the overall model prediction, were age (Cover = 21.1%), Platelet count (Cover = 5.1%), family history of heart disease (Cover = 4.8%), and Total Cholesterol (Cover = 4.1%). Machine learning models can effectively predict coronary artery disease using demographic, laboratory, physical exam, and lifestyle covariates and identify key risk factors.