Identification of influence factors in overweight population through an interpretable risk model based on machine learning: a large retrospective cohort

BACKGROUND

The identification of associated overweight risk factors is crucial to future health risk predictions and behavioral interventions. Several consensus problems remain in machine learning, such as cross-validation, and the resulting model may suffer from overfitting or poor interpretability.

METHODS

This study employed nine commonly used machine learning methods to construct overweight risk models. The general community are the target of this study, and a total of 10,905 Chinese subjects from Ningde City in Fujian province, southeast China, participated. The best model was selected through appropriate verification and validation and was suitably explained.

RESULTS

The overweight risk models employing machine learning exhibited good performance. It was concluded that CatBoost, which is used in the construction of clinical risk models, may surpass previous machine learning methods. The visual display of the Shapley additive explanation value for the machine model variables accurately represented the influence of each variable in the model.

CONCLUSIONS

The construction of an overweight risk model using machine learning may currently be the best approach. Moreover, CatBoost may be the best machine learning method. Furthermore, combining Shapley's additive explanation and machine learning methods can be effective in identifying disease risk factors for prevention and control.

Classification of Obesity among South African Female Adolescents: Comparative Analysis of Logistic Regression and Random Forest Algorithms

BACKGROUND

This study evaluates the performance of logistic regression (LR) and random forest (RF) algorithms to model obesity among female adolescents in South Africa.

METHODS

Data was analysed on 375 females aged 15-17 from the South African National Health and Nutrition Examination Survey 2011/2012. The primary outcome was obesity, defined as body mass index (BMI) ≥ 30 kg/m2. A total of 31 explanatory variables were included, ranging from socio-economic, demographic, family history, dietary and health behaviour. RF and LR models were run using imbalanced data as well as after oversampling, undersampling, and hybrid sampling of the data.

RESULTS

Using the imbalanced data, the RF model performed better with higher precision, recall, F1 score, and balanced accuracy. Balanced accuracy was highest with the hybrid data (0.618 for RF and 0.668 for LR). Using the hybrid balanced data, the RF model performed better (F1-score = 0.940 for RF vs. 0.798 for LR).

CONCLUSION

The model with the highest overall performance metrics was the RF model both before balancing the data and after applying hybrid balancing. Future work would benefit from using larger datasets on adolescent female obesity to assess the robustness of the models.

Trends in the ability of socioeconomic position to predict individual body mass index: an analysis of repeated cross-sectional data, 1991-2019

BACKGROUND

The widening of group-level socioeconomic differences in body mass index (BMI) has received considerable research attention. However, the predictive power of socioeconomic position (SEP) indicators at the individual level remains uncertain, as does the potential temporal variation in their predictive value. Examining this is important given the increasing incorporation of SEP indicators into predictive algorithms and calls to reduce social inequality to tackle the obesity epidemic. We thus investigated SEP differences in BMI over three decades of the obesity epidemic in England, comparing population-wide (SEP group differences in mean BMI) and individual-level (out-of-sample prediction of individuals' BMI) approaches to understanding social inequalities.

METHODS

We used repeated cross-sectional data from the Health Survey for England, 1991-2019. BMI (kg/m2) was measured objectively, and SEP was measured via educational attainment, occupational class, and neighbourhood index of deprivation. We ran random forest models for each survey year and measure of SEP adjusting for age and sex.

RESULTS

The mean and variance of BMI increased within each SEP group over the study period. Mean differences in BMI by SEP group also increased: differences between lowest and highest education groups were 1.0 kg/m2 (0.4, 1.6) in 1991 and 1.3 kg/m2 (0.7, 1.8) in 2019. At the individual level, the predictive capacity of SEP was low, though increased in later years: including education in models improved predictive accuracy (mean absolute error) by 0.14% (- 0.9, 1.08) in 1991 and 1.05% (0.18, 1.82) in 2019. Similar patterns were obtained for occupational class and neighbourhood deprivation and when analysing obesity as an outcome.

CONCLUSIONS

SEP has become increasingly important at the population (group difference) and individual (prediction) levels. However, predictive ability remains low, suggesting limited utility of including SEP in prediction algorithms. Assuming links are causal, abolishing SEP differences in BMI could have a large effect on population health but would neither reverse the obesity epidemic nor reduce much of the variation in BMI.

Identifying Young Adults at High Risk for Weight Gain Using Machine Learning

INTRODUCTION

Weight gain among young adults continues to increase. Identifying adults at high risk for weight gain and intervening before they gain weight could have a major public health impact. Our objective was to develop and test electronic health record-based machine learning models to predict weight gain in young adults with overweight/class 1 obesity.

METHODS

Seven machine learning models were assessed, including three regression models, random forest, single-layer neural network, gradient-boosted decision trees, and support vector machine (SVM) models. Four categories of predictors were included: 1) demographics; 2) obesity-related health conditions; 3) laboratory data and vital signs; and 4) neighborhood-level variables. The cohort was split 60:40 for model training and validation. Area under the receiver operating characteristic curves (AUC) were calculated to determine model accuracy at predicting high-risk individuals, defined by ≥ 10% total body weight gain within 2 y. Variable importance was measured via generalized analysis of variance procedures.

RESULTS

Of the 24,183 patients (mean [SD] age, 32.0 [6.3] y; 55.1% females) in the study, 14.2% gained ≥10% total body weight. Area under the receiver operating characteristic curves varied from 0.557 (SVM) to 0.675 (gradient-boosted decision trees). Age, sex, and baseline body mass index were the most important predictors among the models except SVM and neural network.

CONCLUSIONS

Our machine learning models performed similarly and had modest accuracy for identifying young adults at risk of weight gain. Future models may need to incorporate behavioral and/or genetic information to enhance model accuracy.

The prevalence of anxiety and its key influencing factors among the elderly in China

INTRODUCTION

With the rapid aging population, the mental health of older adults is paid more and more attention. Anxiety is a common mental health illness in older adults. Therefore, the study aimed to explore the current situation of anxiety and its factors among the elderly in China.

METHODS

Based on the data from 2018 Chinese Longitudinal Healthy Longevity Survey (CLHLS), a total of 10,982 respondents aged 60 and above were selected. Generalized Anxiety Disorder (GAD-7) scale was used to assess the anxiety. Univariate and multivariate analysis were used to analyze the influencing factors of anxiety. Random forest was established to rank the importance of each influencing factors.

RESULTS

The results showed that the prevalence of anxiety among the elderly was 11.24%. Anxiety was mainly associated with 14 factors from five aspects: sociodemographic characteristics, health status, psychological state, social trust and social participation, among which loneliness related to psychological status was the most important factor.

DISCUSSION

The revelation of this study is that the present situation of anxiety among the elderly cannot be ignored, and it is necessary to take measures to prevent and control it from many aspects.

Predicting COVID-19 community infection relative risk with a Dynamic Bayesian Network

As COVID-19 continues to impact the United States and the world at large it is becoming increasingly necessary to develop methods which predict local scale spread of the disease. This is especially important as newer variants of the virus are likely to emerge and threaten community spread. We develop a Dynamic Bayesian Network (DBN) to predict community-level relative risk of COVID-19 infection at the census tract scale in the U.S. state of Indiana. The model incorporates measures of social and environmental vulnerability-including environmental determinants of COVID-19 infection-into a spatial temporal prediction of infection relative risk 1-month into the future. The DBN significantly outperforms five other modeling techniques used for comparison and which are typically applied in spatial epidemiological applications. The logic behind the DBN also makes it very well-suited for spatial-temporal prediction and for "what-if" analysis. The research results also highlight the need for further research using DBN-type approaches that incorporate methods of artificial intelligence into modeling dynamic processes, especially prominent within spatial epidemiologic applications.

Machine-Learning for Prescription Patterns: Random Forest in the Prediction of Dose and Number of Antipsychotics Prescribed to People with Schizophrenia

Objective

We aimed to predict antipsychotic prescription patterns for people with schizophrenia using machine learning (ML) algorithms.

Methods

In a cross-sectional design, a sample of community mental health service users (SUs; n = 368) with a primary diagnosis of schizophrenia was randomly selected. Socio-demographic and clinical features, including the number, total dose, and route of administration of the antipsychotic treatment were recorded. Information about the number and the length of psychiatric hospitalization was retrieved. Ordinary Least Square (OLS) regression and ML algorithms (i.e., random forest [RF], supported vector machine, K-nearest neighborhood, and Naïve Bayes) were used to estimate the predictors of total antipsychotic dosage and prescription of antipsychotic polytherapy (APP).

Results

The strongest predictor of the total dose was APP. The number of Community Mental Health Centers (CMHC) contacts was the most important predictor of APP and, with APP omitted, of dosage. Treatment with anticholinergics predicted APP, emphasizing the strong correlation between APP and higher antipsychotic dose. RF performed better than OLS regression and the other ML algorithms in predicting both antipsychotic dose (root square mean error = 0.70, R² = 0.31) and APP (area under the receiving operator curve = 0.66, true positive rate = 0.41, and true negative rate = 0.78).

Conclusion

APP is associated with the prescription of higher total doses of antipsychotics. Frequent attenders at CMHCs, and SUs recently hospitalized are often treated with APP and higher doses of antipsychotics. Future prospective studies incorporating standardized clinical assessments for both psychopathological severity and treatment efficacy are needed to confirm these findings.

Association of body composition with pubertal timing in children and adolescents from Guangzhou, China

Objective

An ongoing debate surrounds the relationship between body composition and pubertal timing, in particular for boys. This cross-sectional study aimed to investigate the association of body composition with pubertal timing among children and adolescents.

Methods

A total of 1,493 boys and 1,261 girls who entered puberty were enrolled in Guangzhou, China. Tanner stages were evaluated by examination of breast development for girls and testicular volume for boys. Fat mass (FM) and fat-free mass (FFM) were determined by bioelectrical impedance analysis. Parameters for body composition were transformed into age-and gender-specific Z-scores. The association of body composition with pubertal timing was examined using multinomial logistic regression with inverse probability weighting (IPW) based on the propensity score.

Results

For boys, IPW analysis showed Z-scores of body fat percentage (BF%) and FM index (FMI) were negatively associated with early puberty (OR = 0.75, 95% CI = 0.64-0.87; OR = 0.74, 95% CI = 0.63-0.88). As for girls, in contrast to boys, positive associations were seen between BF% and FMI with early puberty (OR = 1.39, 95% CI = 1.19-1.64; OR = 1.59, 95% CI = 1.33-1.90). With respect to appendicular skeletal muscle mass index (ASMI), there was a positive association with early puberty and a negative one with late puberty in boys (OR = 1.26, 95% CI = 1.07-1.49; OR = 0.82, 95% CI = 0.69-0.99).

Conclusion

There is a positive association of FM with early puberty for girls while negative for boys. FFM yields a positive association with early puberty and a negative one with late puberty in boys, but not in girls. Our findings highlight the gender differences in the connection between body composition and pubertal onset.

Daily distracted consumption patterns and their relationship with BMI

The rapidly increasing prevalence of overweight and obesity has heightened the need for a better understanding of obesity-related eating patterns and dietary behaviours. Recent work suggests that distracted eating is causally related to increased immediate and later food, pushing the need for a better understanding of the prevalence of distracted consumption and how this relates to body weight. To extract insights in the relationship between demographics, daily consumption settings, and BMI, we performed secondary data analyses on data from 1011 individuals representative of the Dutch population (adults, 507F, BMI 17-50 kg/m²). The most commonly reported distractions were talking to others (32.7%) and watching television (21.7%). Only 18.4% of respondents reported no distractions during meals. To examine how different distractions related to BMI, we performed OLS regression which showed, among other things, that watching tv while eating lunch (η² = 0.37) and working during dinner were associated with a higher BMI (η² = 1.63). To examine the robustness of these findings, machine learning techniques were used. A random forest analysis (RMSE = 4.09) showed that next to age and education level, distraction during lunch and snack was amongst the largest predictors of BMI. Multiple linear regression with lasso penalty (RMSE = 4.13) showed that specifically watching tv while eating lunch or snacks was associated with a higher BMI. In conclusion, our analyses confirmed the assumption that people are regularly distracted during their daily meals, with distinct distractors relating to BMI. These findings provide a starting point for evidence-based recommendations on which consumption settings are associated with healthier eating patterns and body weight.

Applicability of machine learning techniques in food intake assessment: A systematic review

The evaluation of food intake is important in scientific research and clinical practice to understand the relationship between diet and health conditions of an individual or a population. Large volumes of data are generated daily in the health sector. In this sense, Artificial Intelligence (AI) tools have been increasingly used, for example, the application of Machine Learning (ML) algorithms to extract useful information, find patterns, and predict diseases. This systematic review aimed to identify studies that used ML algorithms to assess food intake in different populations. A literature search was conducted using five electronic databases, and 36 studies met all criteria and were included. According to the results, there has been a growing interest in the use of ML algorithms in the area of nutrition in recent years. Also, supervised learning algorithms were the most used, and the most widely used method of nutritional assessment was the food frequency questionnaire. We observed a trend in using the data analysis programs, such as R and WEKA. The use of ML in nutrition is recent and challenging. Therefore, it is encouraged that more studies are carried out relating these themes for the development of food reeducation programs and public policies.

Classifying Non-Dementia and Alzheimer's Disease/Vascular Dementia Patients Using Kinematic, Time-Based, and Visuospatial Parameters: The Digital Clock Drawing Test

BACKGROUND

Advantages of digital clock drawing metrics for dementia subtype classification needs examination.

OBJECTIVE

To assess how well kinematic, time-based, and visuospatial features extracted from the digital Clock Drawing Test (dCDT) can classify a combined group of Alzheimer's disease/Vascular Dementia patients versus healthy controls (HC), and classify dementia patients with Alzheimer's disease (AD) versus vascular dementia (VaD).

METHODS

Healthy, community-dwelling control participants (n = 175), patients diagnosed clinically with Alzheimer's disease (n = 29), and vascular dementia (n = 27) completed the dCDT to command and copy clock drawing conditions. Thirty-seven dCDT command and 37 copy dCDT features were extracted and used with Random Forest classification models.

RESULTS

When HC participants were compared to participants with dementia, optimal area under the curve was achieved using models that combined both command and copy dCDT features (AUC = 91.52%). Similarly, when AD versus VaD participants were compared, optimal area under the curve was, achieved with models that combined both command and copy features (AUC = 76.94%). Subsequent follow-up analyses of a corpus of 10 variables of interest determined using a Gini Index found that groups could be dissociated based on kinematic, time-based, and visuospatial features.

CONCLUSION

The dCDT is able to operationally define graphomotor output that cannot be measured using traditional paper and pencil test administration in older health controls and participants with dementia. These data suggest that kinematic, time-based, and visuospatial behavior obtained using the dCDT may provide additional neurocognitive biomarkers that may be able to identify and tract dementia syndromes.

A scoping review on the use of machine learning in research on social determinants of health: Trends and research prospects

Background

Machine learning (ML) has spread rapidly from computer science to several disciplines. Given the predictive capacity of ML, it offers new opportunities for health, behavioral, and social scientists. However, it remains unclear how and to what extent ML is being used in studies of social determinants of health (SDH).

Methods

Using four search engines, we conducted a scoping review of studies that used ML to study SDH (published before May 1, 2020). Two independent reviewers analyzed the relevant studies. For each study, we identified the research questions, Results, data, and algorithms. We synthesized our findings in a narrative report.

Results

Of the initial 8097 hits, we identified 82 relevant studies. The number of publications has risen during the past decade. More than half of the studies (n = 46) used US data. About 80% (n = 66) utilized surveys, and 70% (n = 57) employed ML for common prediction tasks. Although the number of studies in ML and SDH is growing rapidly, only a few studies used ML to improve causal inference, curate data, or identify social bias in predictions (i.e., algorithmic fairness).

Conclusions

While ML equips researchers with new ways to measure health outcomes and their determinants from non-conventional sources such as text, audio, and image data, most studies still rely on traditional surveys. Although there are no guarantees that ML will lead to better social epidemiological research, the potential for innovation in SDH research is evident as a result of harnessing the predictive power of ML for causality, data curation, or algorithmic fairness.

Application of machine learning to understand child marriage in India

BACKGROUND

Prior research documents that India has the greatest number of girls married as minors of any nation in the world, increasing social and health risks for both these young wives and their children. While the prevalence of child marriage has declined in the nation, more work is needed to accelerate this decline and the negative consequences of the practice. Expanded targets for intervention require greater identification of these targets. Machine learning can offer insight into identification of novel factors associated with child marriage that can serve as targets for intervention.

METHODS

We applied machine learning methods to retrospective cross-sectional survey data from India on demographics and health, the nationally-representative National Family Health Survey, conducted in 2015-16. We analyzed data using a traditional regression model, with child marriage as the dependent variable, and 4000+ variables from the survey as the independent variables. We also used three commonly used machine learning algorithms- Least Absolute Shrinkage and Selection Operator (lasso) or L-1 regularized logistic regression models; L2 regularized logistic regression or ridge models; and neural network models. Finally, we developed and applied a novel and rigorous approach involving expert qualitative review and coding of variables generated from an iterative series of regularized models to assess thematically key variable groupings associated with child marriage.

FINDINGS

Analyses revealed that regularized logistic and neural network applications demonstrated better accuracy and lower error rates than traditional logistic regression, with a greater number of features and variables generated. Regularized models highlight higher fertility and contraception, longer duration of marriage, geographic, and socioeconomic vulnerabilities as key correlates; findings shown in prior research. However, our novel method involving expert qualitative coding of variables generated from iterative regularized models and resultant thematic generation offered clarity on variables not focused upon in prior research, specifically non-utilization of health system benefits related to nutrition for mothers and infants.

INTERPRETATION

Machine learning appears to be a valid means of identifying key correlates of child marriage in India and, via our innovative iterative thematic approach, can be useful to identify novel variables associated with this outcome. Findings related to low nutritional service uptake also demonstrate the need for more focus on public health outreach for nutritional programs tailored to this population.

Fertility is a key predictor of the double burden of malnutrition among women of child-bearing age in sub-Saharan Africa

BACKGROUND

The ongoing nutrition transition in sub-Saharan Africa (SSA) is exhibiting spatial heterogeneity and temporal variability leading to different forms of malnutrition burden across SSA, with some regions exhibiting the double burden of malnutrition. This study aimed to develop a predictive understanding of the malnutrition burden among women of child-bearing age.

METHODS

Data from 34 SSA countries were acquired from the Demographic and Health Survey, World Bank, and Swiss Federal Institute of Technology. The SSA countries were classified into malnutrition classes based on their national prevalence of underweight, overweight, and obesity using a 10% threshold. Next, random forest analysis was used to examine the association between country-level demographic variables and the national prevalence of underweight, overweight and obesity. Finally, random forest analysis and multinomial logistic regression models were utilized to investigate the association between individual-level social and demographic variables and Body Mass Index (BMI) categories of underweight, normal weight, and combined overweight and obesity.

RESULTS

Four malnutrition classes were identified: Class A had 5 countries with ≥10% of the women underweight; Class B had 11 countries with ≥10% each of underweight and overweight; Class C1 had 7 countries with ≥10% overweight; and Class C2 had 11 countries with ≥10% obesity. At the country-level, fertility rate predicted underweight, overweight and obesity prevalence, but economic indicators were also important, including the gross domestic product per capita - a measure of economic opportunity that predicted both overweight and obesity prevalence, and the GINI coefficient - a measure of economic inequality that predicted both underweight and overweight prevalence. At the individual-level, parity was a risk factor for underweight in underweight burdened countries and a risk factor for overweight/obesity in overweight/obesity burdened countries, whereas age and wealth were protective factors for underweight but risk factors for overweight/obesity.

CONCLUSIONS

Beyond the effect of economic indicators, this study revealed the important role of fertility rate and parity, which may represent risk factors for both underweight and combined overweight and obesity among women of child-bearing age. Health professionals should consider combining reproductive health services with nutritional programs when addressing the challenge of malnutrition in SSA.

Individualized predictions of changes in knee pain, quality of life and walking speed following patient education and exercise therapy in patients with knee osteoarthritis - a prognostic model study

OBJECTIVE

To facilitate shared decision-making for patients with knee osteoarthritis (OA), we aimed at building clinically applicable models to predict the individual change in pain intensity (VAS scale 0-100), knee-related quality of life (QoL) (KOOS QoL score 0-100) and walking speed (m/sec) immediately following two educational and 12 supervised exercise therapy sessions.

METHODS

We used data from patients with knee OA from the 'Good Life with osteoArthritis in Denmark' (GLA:D®) registry (n = 6,767). From 51 patient characteristics, we selected the best performing variables to predict the outcomes via random forest regression. We evaluated model performance via R². Lastly, we validated and compared our models with the average improvements via the mean differences in an independent validation data set from the GLA:D® registry (n = 2,896) collected 1 year later than the data used to build the models.

RESULTS

Validating our models including the best performing variables yielded R²s of 0.34 for pain intensity, 0.18 for knee-related QoL, and 0.07 for walking speed. The absolute mean differences between model predictions and the true outcomes were 14.65 mm, 10.32 points, and 0.14 m/s, respectively, and similar to the absolute mean differences of 17.64, 11.28 and 0.14 observed when we subtracted the average improvements from the true outcomes.

CONCLUSION

Despite including 51 potential predictors, we were unable to predict changes in individuals' pain intensity, knee-related QoL and walking speed with clinically relevant greater precision than the respective group average outcomes. Therefore, average prediction values can be used to inform patients about expected outcomes.

Machine learning as a strategy to account for dietary synergy: an illustration based on dietary intake and adverse pregnancy outcomes

BACKGROUND

Conventional analytic approaches for studying diet patterns assume no dietary synergy, which can lead to bias if incorrectly modeled. Machine learning algorithms can overcome these limitations.

OBJECTIVES

We estimated associations between fruit and vegetable intake relative to total energy intake and adverse pregnancy outcomes using targeted maximum likelihood estimation (TMLE) paired with the ensemble machine learning algorithm Super Learner, and compared these with results generated from multivariable logistic regression.

METHODS

We used data from 7572 women in the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be. Usual daily periconceptional intake of total fruits and total vegetables was estimated from an FFQ. We calculated the marginal risk of preterm birth, small-for-gestational-age (SGA) birth, gestational diabetes, and pre-eclampsia according to density of fruits and vegetables (cups/1000 kcal) ≥80th percentile compared with <80th percentile using multivariable logistic regression and Super Learner with TMLE. Models were adjusted for confounders, including other Healthy Eating Index-2010 components.

RESULTS

Using logistic regression, higher fruit and high vegetable densities were associated with 1.1% and 1.4% reductions in pre-eclampsia risk compared with lower densities, respectively. They were not associated with the 3 other outcomes. Using Super Learner with TMLE, high fruit and vegetable densities were associated with fewer cases of preterm birth (-4.0; 95% CI: -4.9, -3.0 and -3.7; 95% CI: -5.0, -2.3), SGA (-1.7; 95% CI: -2.9, -0.51 and -3.8; 95% CI: -5.0, -2.5), and pre-eclampsia (-3.2; 95% CI: -4.2, -2.2 and -4.0; 95% CI: -5.2, -2.7) per 100 births, respectively, and high vegetable densities were associated with a 0.9% increase in risk of gestational diabetes.

CONCLUSIONS

The differences in results between Super Learner with TMLE and logistic regression suggest that dietary synergy, which is accounted for in machine learning, may play a role in pregnancy outcomes. This innovative methodology for analyzing dietary data has the potential to advance the study of diet patterns.