Decision trees in epidemiological research

Identifying intersectional groups at risk for missing breast cancer screening: Comparing regression- and decision tree-based approaches

Malignant neoplasm of the breast was the fifth leading cause of death among women in Germany in 2020. To improve early detection, nationwide breast cancer screening (BCS) programmes for women 50-69 have been implemented since 2005. However, Germany has not reached the European benchmark of 70% participation, and socio-demographic inequalities persist. At the same time, challenges exist to identify groups of women at high risk for non-participation, since it is likely that this is due to disadvantages on multiple social dimensions. This study, therefore, aimed to identify intersectional groups of women at higher risk of not attending BCS by comparing two analytical strategies: a) evidence-informed regression and b) decision tree-based regression. Participants were drawn from the German 2019 European Health Interview Survey (N = 23,001; 21.6% response rate). Two logistic regressions using cross-classification intersectional groups based on relevant PROGRESS-Plus characteristics adjusted by age were built. The evidence-informed approach selected relevant variables based on the literature and the decision tree approach on the best-performing tree. The first identified low-income women born outside Germany, living in rural areas and not cohabiting with their partner at higher risk of never attending BCS (OR = 9.48, p = 0.002), whereas the second, based on a Classification and Regression Tree (61.91% balanced accuracy), determined widowed women living alone, with children, with a partner and children, or in other arrangements, and residing in specific federal states (i.e. Bavaria, Brandenburg, Bremen, Hamburg, or Saarland) (OR = 3.43, p < 0.001). Compared to the evidence-informed regression, the decision tree-based regression yielded higher discriminatory accuracy (AUC = 0.6726 vs AUC = 0.6618) and added relevant nuances in the identification of at-risk intersectional groups, going beyond known inequality dimensions and, therefore, helping the inclusion of under-studied populations in breast cancer screening.

Nothing means anything if everything means something: exploring the issues of coping profiles and the person-centered approach

BACKGROUND

Adopting a person-centered approach to coping potentially allows researchers to explore the multifaceted nature of the construct. However, this increasingly adopted approach also has limitations. Namely, employing cluster or latent profile analysis to investigate coping through a person-centered lens often brings a lack of generalizability and subjectivity in interpreting the generated profiles. As such, this study aimed to explore the impact of varied methodology in person-centered investigations of coping profiles.

METHODS

682 university students' (M = 21.3 years old, SD = 3.5) responses to the COPE Inventory were analyzed across item, subscale, and higher-order category levels using cluster and latent profile analysis to produce 6 finalized models for cross-method comparison.

RESULTS

Throughout 19 analyses, approach coping, avoidance coping, low coping, and help-seeking profiles were consistently identified, alluding to the potential of universal coping trends. However, membership overlap across COPE structures and methodology was largely inconsistent, with individual participants classified into theoretically distinct profiles based on the methodology employed.

CONCLUSION

While evidence suggests latent profile analysis provides a more rigorous approach, the significant impact of minor methodological variations urges a reevaluation of person-centered approaches and incorporation of multi-construct data to enhance the understanding of coping profiles.

Development of predictive model for post-stroke depression at discharge based on decision tree algorithm: A multi-center hospital-based cohort study

OBJECTIVE

Post-stroke depression (PSD) is one of the most common and severe neuropsychological sequelae after stroke. Using a prediction model composed of multiple predictors may be more beneficial than verifying the predictive performance of any single predictor. The primary objective of this study was to construct practical prediction tools for PSD at discharge utilizing a decision tree (DT) algorithm.

METHODS

A multi-center prospective cohort study was conducted from May 2018 to October 2019 and stroke patients within seven days of onset were consecutively recruited. The independent predictors of PSD at discharge were identified through multivariate logistic regression with backward elimination. Classification and regression tree (CART) algorithm was employed as the DT model's splitting method.

RESULTS

A total of 876 stroke patients who were discharged from the neurology departments of three large general Class A tertiary hospitals in Wuhan were eligible for analysis. Firstly, we divided these 876 patients into PSD and non-PSD groups, history of coronary heart disease (OR = 1.835; 95 % CI, 1.106-3.046; P = 0.019), length of hospital stay (OR = 1.040; 95 % CI, 1.013-1.069; P = 0.001), NIHSS score (OR = 1.124; 95 % CI, 1.052-1.201; P = 0.001), and Mini mental state examination (MMSE) score (OR = 0.935; 95 % CI, 0.893-0.978; P = 0.004) were significant predictors. The subgroup analysis results have shown that hemorrhagic stroke, history of hypertension and higher modified Rankin Scale score (mRS) score were associated with PSD at discharge in the young adult stroke patients.

CONCLUSIONS

Several predictors of PSD at discharge were identified and convenient DT models were constructed to facilitate clinical decision-making.

Intersectional inequities in colorectal cancer screening attendance in Sweden: Using decision trees for intersectional matrix reduction

Colorectal cancer (CRC) represents a significant health burden worldwide, with existing inequities in incidence and mortality. In Sweden, CRC screening programmes have varied regionally since the mid-2000s, but the significance of organised screening for counteracting complex inequities in screening attendance has not been investigated. This study aimed to assess patterns of inequities in lifetime CRC screening attendance in the Swedish population aged 60-69 years by identifying intersectional strata at higher risk of never attending CRC screening. The research question is answered using decision trees to reduce the complexity of a full intersectional matrix into a reduced intersectional matrix for risk estimation. Participants were drawn from the cross-sectional 2019 European Health Interview Survey (N = 9,757, response rate: 32.52%). The Conditional Inference Tree (CIT) (AUC = 0.7489, F-score = 0.7912, depth = 4, significance level = 0.05) identified region of residence (opportunistic vs organised screening), country of origin, gender, age and income as relevant variables in explaining lifetime CRC screening attendance in Sweden. Then, Poisson regression with robust standard errors estimated that EU-born women living in opportunistic screening regions belonging to the 2nd income quintile had the highest risk of never attending CRC screening (PR = 8.54, p < 0.001), followed by EU-born men living in opportunistic screening regions (PR = 7.41, p < 0.001) compared to the reference category (i.e. people aged 65-69 living in organised screening regions). In contrast, only age-related differences in attendance were found in regions with organised screening (i.e. people aged 60-64 living in regions with organised screening (PR = 2.01, p < 0.05)). The AUC of the reduced intersectional matrix model (0.7489) was higher than the full intersectional matrix model (0.6959) and slightly higher than the main effects model (0.7483), demonstrating intersectional effects of the reduced intersectional matrix compared with the main effects model and better discriminatory accuracy than the full intersectional matrix. In conclusion, regions with long-established organised CRC screening programmes display more limited socio-demographic inequities than regions with opportunistic CRC screening. This suggests that organised screening may be a crucial policy instrument to improve equity in CRC screening, which, in the long run, has the potential to prevent inequities in colorectal cancer mortality. Moreover, decision trees appear to be valuable statistical tools for efficient data-driven simplification of the analytical and empirical complexity that epidemiological intersectional analysis conventionally entails.

Health-related quality of life in children and adolescents born very preterm and its correlates: a cross-sectional study

OBJECTIVE

We aimed to assess health-related quality of life (HRQOL) in a cohort of very preterm born children and adolescents (aged 5-16), and to compare it with their fullterm born siblings and the general population. We also explored correlates of HRQOL among the very preterm born.

DESIGN

Cross-sectional survey.

PATIENTS

Children born <32 weeks gestation (N=442) as well as their fullterm born siblings (N=145).

MAIN OUTCOME MEASURES

Primary outcome was KINDL total score (0 worst to 100 best), a validated multidimensional measure of HRQOL in children and adolescents.

METHODS

Linear mixed models accounted for family unit. Secondary analysis compared very preterm born children to another cohort of healthy children from the same time period. A classification tree analysis explored potential correlates of HRQOL.

RESULTS

On average, preterm children, both <28 and 28-31 weeks gestational age, had similar KINDL total score to fullterm sibling controls (-2.3, 95% CI -3.6 to -0.6), and to population controls (+1.4, 95% CI 0.2 to 2.5). Chronic non-respiratory health conditions (such as attention deficit hyperactivity disorder or heart conditions, but not including cerebral palsy), age and respiratory symptoms affecting daily life were key correlates of HRQOL among very preterm born children.

CONCLUSIONS

Very preterm birth in children and adolescents was not associated with a relevant reduction in HRQOL compared with their fullterm born peers. However, lower HRQOL was explained by other factors, such as older age, and the presence of chronic non-respiratory health conditions, but also by possibly modifiable current respiratory symptoms. The influence of respiratory symptom amelioration and its potential influence on HRQOL needs to be investigated further.

TRIAL REGISTRATION NUMBER

NCT04448717.

Identifying combinations of long-term conditions associated with sarcopenia: a cross-sectional decision tree analysis in the UK Biobank study

OBJECTIVES

This study aims to determine whether machine learning can identify specific combinations of long-term conditions (LTC) associated with increased sarcopenia risk and hence address an important evidence gap-people with multiple LTC (MLTC) have increased risk of sarcopenia but it has not yet been established whether this is driven by specific combinations of LTC.

DESIGN

Decision trees were used to identify combinations of LTC associated with increased sarcopenia risk. Participants were classified as being at risk of sarcopenia based on maximum grip strength of <32 kg for men and <19 kg for women. The combinations identified were triangulated with logistic regression.

SETTING

UK Biobank.

PARTICIPANTS

UK Biobank participants with MLTC (two or more LTC) at baseline.

RESULTS

Of 140 001 participants with MLTC (55.3% women, median age 61 years), 21.0% were at risk of sarcopenia. Decision trees identified several LTC combinations associated with an increased risk of sarcopenia. These included drug/alcohol misuse and osteoarthritis, and connective tissue disease and osteoporosis in men, which showed the relative excess risk of interaction of 3.91 (95% CI 1.71 to 7.51) and 2.27 (95% CI 0.02 to 5.91), respectively, in age-adjusted models.

CONCLUSION

Knowledge of LTC combinations associated with increased sarcopenia risk could aid the identification of individuals for targeted interventions, recruitment of participants to sarcopenia studies and contribute to the understanding of the aetiology of sarcopenia.

Predictors of early child development for screening pregnant women most in need of support in Brazil

Background

Home visiting programmes can support child development and reduce inequalities, but failure to identify the most vulnerable families can undermine such efforts. We examined whether there are strong predictors of poor child development that could be used to screen pregnant women in primary health care settings to target early interventions in a Brazilian population. Considering selected predictors, we assessed coverage and focus of a large-scale home visiting programme named Primeira Infância Melhor (PIM).

Methods

We undertook a prospective cohort study on 3603 children whom we followed from gestation to age four years. We then used 27 potential socioeconomic, psychosocial, and clinical risk factors measurable during pregnancy to predict child development, which was assessed by the Battelle Developmental Inventory (BDI) at the age of four years. We compared the results from a Bonferroni-adjusted conditional inference tree with exploratory linear regression and principal component analysis (PCA), and we conducted external validation using data from a second cohort from the same population. Lastly, we assessed PIM coverage and focus by linking 2015 cohort data with PIM databases.

Results

The decision tree analyses identified maternal schooling as the most important variable for predicting BDI, followed by paternal schooling. Based on these variables, a group of 214 children who had the lowest mean BDI (BDI = -0.48; 95% confidence interval (CI) = -0.63, -0.33) was defined by mothers with ≤5 years and fathers with ≤4 years of schooling. Maternal and paternal schooling were also the strongest predictors in the exploratory analysis using regression and PCA, showing linear associations with the outcome. However, their capacity to explain outcome variance was low, with an adjusted R2 of 5.3% and an area under the receiver operating characteristic curve of 0.62 (95% CI = 0.60, 0.64). External validation showed consistent results. We also provided an online screening tool using parental schooling data to support programme's targeting. PIM coverage during pregnancy was low, but the focus was adequate, especially among families with longer enrolment, indicating families most in need received higher dosage.

Conclusions

Information on maternal and paternal schooling can improve the focus of home visiting programmes if used for initial population-level screening of pregnant women in Brazil. However, enrolment decisions require complementary information on parental resources and direct interactions with families to jointly decide on inclusion.

Assessing the Value of Imaging Data in Machine Learning Models to Predict Patient-Reported Outcome Measures in Knee Osteoarthritis Patients

Knee osteoarthritis (OA) affects over 650 million patients worldwide. Total knee replacement is aimed at end-stage OA to relieve symptoms of pain, stiffness and reduced mobility. However, the role of imaging modalities in monitoring symptomatic disease progression remains unclear. This study aimed to compare machine learning (ML) models, with and without imaging features, in predicting the two-year Western Ontario and McMaster Universities Arthritis Index (WOMAC) score for knee OA patients. We included 2408 patients from the Osteoarthritis Initiative (OAI) database, with 629 patients from the Multicenter Osteoarthritis Study (MOST) database. The clinical dataset included 18 clinical features, while the imaging dataset contained an additional 10 imaging features. Minimal Clinically Important Difference (MCID) was set to 24, reflecting meaningful physical impairment. Clinical and imaging dataset models produced similar area under curve (AUC) scores, highlighting low differences in performance AUC < 0.025). For both clinical and imaging datasets, Gradient Boosting Machine (GBM) models performed the best in the external validation, with a clinically acceptable AUC of 0.734 (95% CI 0.687-0.781) and 0.747 (95% CI 0.701-0.792), respectively. The five features identified included educational background, family history of osteoarthritis, co-morbidities, use of osteoporosis medications and previous knee procedures. This is the first study to demonstrate that ML models achieve comparable performance with and without imaging features.

Prediction model for lower limb amputation in hospitalized diabetic foot patients using classification and regression trees

BACKGROUND

The decision to perform amputation of a limb in a patient with diabetic foot ulcer (DFU) is not an easy task. Prediction models aim to help the surgeon in decision making scenarios. Currently there are no prediction model to determine lower limb amputation during the first 30 days of hospitalization for patients with DFU.

METHODS

Classification And Regression Tree analysis was applied on data from a retrospective cohort of patients hospitalized for the management of diabetic foot ulcer, using an existing database from two Orthopaedics and Traumatology departments. The secondary analysis identified independent variables that can predict lower limb amputation (mayor or minor) during the first 30 days of hospitalization.

RESULTS

Of the 573 patients in the database, 290 feet underwent a lower limb amputation during the first 30 days of hospitalization. Six different models were developed using a loss matrix to evaluate the error of not detecting false negatives. The selected tree produced 13 terminal nodes and after the pruning process, only one division remained in the optimal tree (Sensitivity: 69%, Specificity: 75%, Area Under the Curve: 0.76, Complexity Parameter: 0.01, Error: 0.85). Among the studied variables, the Wagner classification with a cut-off grade of 3 exceeded others in its predicting capacity.

CONCLUSIONS

Wagner classification was the variable with the best capacity for predicting amputation within 30 days. Infectious state and vascular occlusion described indirectly by this classification reflects the importance of taking quick decisions in those patients with a higher compromise of these two conditions. Finally, an external validation of the model is still required.

LEVEL OF EVIDENCE

III.

Risk factors for homelessness and housing instability in the first episode of mental illness: Initial findings from the AMONT study

AIM

People living with mental illness are more likely than the general population to experience adverse housing outcomes, including homelessness. The aim of the current study is to examine residential status when participants have their first contact with mental health services, and the correlates of residential status at that moment.

METHODS

First-time mental health service users were recruited from seven clinical sites across Québec. Data on residential status at entry in the project, as well as demographic, clinical and social variables, were collected using self-report and interviewer-rated questionnaires. Participants were classified as 'Homeless', 'At risk of homelessness' and 'Stably Housed', and correlates of residential status were identified through multivariate logistic regression and unbiased recursive partitioning.

RESULTS

Among the 478 participants, 206 (43.1%) were in stable housing, 171 (35.8%) were at risk of homelessness and 101 (21.1%) were classified as homeless. Placement in a youth protection facility was strongly associated with adverse housing outcomes, while having a high school diploma and more social support were associated with more stable housing situations.

CONCLUSIONS

First-time mental health service users are likely to experience a range of adverse housing situations, indicating the potential for clinical sites to implement homelessness primary prevention strategies. Factors related to family, foster care and schooling seem to be particularly salient in understanding risk of homelessness in first-time mental health service users, calling for intersectoral action to prevent adverse psychosocial outcomes in this population.

Use of machine learning to identify protective factors for death from COVID-19 in the ICU: a retrospective study

Background

Patients in serious condition due to COVID-19 often require special care in intensive care units (ICUs). This disease has affected over 758 million people and resulted in 6.8 million deaths worldwide. Additionally, the progression of the disease may vary from individual to individual, that is, it is essential to identify the clinical parameters that indicate a good prognosis for the patient. Machine learning (ML) algorithms have been used for analyzing complex medical data and identifying prognostic indicators. However, there is still an urgent need for a model to elucidate the predictors related to patient outcomes. Therefore, this research aimed to verify, through ML, the variables involved in the discharge of patients admitted to the ICU due to COVID-19.

Methods

In this study, 126 variables were collected with information on demography, hospital length stay and outcome, chronic diseases and tumors, comorbidities and risk factors, complications and adverse events, health care, and vital indicators of patients admitted to an ICU in southern Brazil. These variables were filtered and then selected by a ML algorithm known as decision trees to identify the optimal set of variables for predicting patient discharge using logistic regression. Finally, a confusion matrix was performed to evaluate the model's performance for the selected variables.

Results

Of the 532 patients evaluated, 180 were discharged: female (16.92%), with a central venous catheter (23.68%), with a bladder catheter (26.13%), and with an average of 8.46- and 23.65-days using bladder catheter and submitted to mechanical ventilation, respectively. In addition, the chances of discharge increase by 14% for each additional day in the hospital, by 136% for female patients, 716% when there is no bladder catheter, and 737% when no central venous catheter is used. However, the chances of discharge decrease by 3% for each additional year of age and by 9% for each other day of mechanical ventilation. The performance of the training data presented a balanced accuracy of 0.81, sensitivity of 0.74, specificity of 0.88, and the kappa value was 0.64. The test performance had a balanced accuracy of 0.85, sensitivity 0.75, specificity 0.95, and kappa value of 0.73. The McNemar test found that there were no significant differences in the error rates in the training and test data, suggesting good classification. This work showed that female, the absence of a central venous catheter and bladder catheter, shorter mechanical ventilation, and bladder catheter duration were associated with a greater chance of hospital discharge. These results may help develop measures that lead to a good prognosis for the patient.

Utility of artificial intelligence in the diagnosis and management of keratoconus: a systematic review

Introduction

The application of artificial intelligence (AI) systems in ophthalmology is rapidly expanding. Early detection and management of keratoconus is important for preventing disease progression and the need for corneal transplant. We review studies regarding the utility of AI in the diagnosis and management of keratoconus and other corneal ectasias.

Methods

We conducted a systematic search for relevant original, English-language research studies in the PubMed, Web of Science, Embase, and Cochrane databases from inception to October 31, 2023, using a combination of the following keywords: artificial intelligence, deep learning, machine learning, keratoconus, and corneal ectasia. Case reports, literature reviews, conference proceedings, and editorials were excluded. We extracted the following data from each eligible study: type of AI, input used for training, output, ground truth or reference, dataset size, availability of algorithm/model, availability of dataset, and major study findings.

Results

Ninety-three original research studies were included in this review, with the date of publication ranging from 1994 to 2023. The majority of studies were regarding the use of AI in detecting keratoconus or subclinical keratoconus (n=61). Among studies regarding keratoconus diagnosis, the most common inputs were corneal topography, Scheimpflug-based corneal tomography, and anterior segment-optical coherence tomography. This review also summarized 16 original research studies regarding AI-based assessment of severity and clinical features, 7 studies regarding the prediction of disease progression, and 6 studies regarding the characterization of treatment response. There were only three studies regarding the use of AI in identifying susceptibility genes involved in the etiology and pathogenesis of keratoconus.

Discussion

Algorithms trained on Scheimpflug-based tomography seem promising tools for the early diagnosis of keratoconus that can be particularly applied in low-resource communities. Future studies could investigate the application of AI models trained on multimodal patient information for staging keratoconus severity and tracking disease progression.

Tilapia consumption patterns and consumer preferences: Predictors and perspectives of consumers in Ghana

The objective of this study was to assess consumer behaviour towards tilapia and tilapia products and provide information linking production with consumption patterns and preferences as well as to predict factors that influence consumer preference, purchase behaviour, and willingness to patronize tilapia fillets using classification and regression trees. A total of 960 responses were obtained using convenient sampling. The findings of this survey indicate that tilapia is eaten mainly because of its taste. Regarding the various cooked tilapia options available in Ghana, 58.5 % preferred charcoal-grilled tilapia while sixty-six per cent (66 %) preferred to purchase their tilapia in the fresh state. Furthermore, sixty-five per cent (65 %) of the participants revealed that they consume tilapia at least once a month, indicating a link between production and consumption, as well as a continuous market for tilapia fish farmers. Most respondents (85 %) would prefer an easier way to prepare tilapia. The availability of tilapia in a fillet form appealed to about 50.8 % of respondents with 78 % indicating that they would purchase tilapia fillets if they were available on the market. For the parts of tilapia consumed, 70 % indicated that the head of tilapia was important to them and only 49 % of respondents indicated they would buy fillets without the head. The top three preferred fillet options in increasing order were chilled, frozen, and spiced. From the study of associations, income was the most important factor determining whether a consumer would purchase tilapia fillets or not. However, with regards to preference of head or tail region, age was the most important determining factor. Thus, consideration of all these factors would serve as a guide to businesspeople and actors within the tilapia value chain in Ghana and beyond.

Determinants of health-related quality of life in healthy children and adolescents during the COVID-19 pandemic: Results from a prospective longitudinal cohort study

Understanding health-related quality of life (HRQOL) in children and adolescents, during a pandemic and afterwards, aids in understanding how circumstances in their lives impact their well-being. We aimed to identify determinants of HRQOL from a broad range of biological, psychological, and social factors in a large longitudinal population-based sample. Data was taken from a longitudinal sample (n = 1843) of children and adolescents enrolled in the prospective school-based cohort study Ciao Corona in Switzerland. The primary outcome was HRQOL, assessed using the KINDL total score and its subscales (each from 0, worst, to 100, best). Potential determinants, including biological (physical activity, screen time, sleep, etc.), psychological (sadness, anxiousness, stress), and social (nationality, parents' education, etc.) factors, were assessed in 2020 and 2021 and HRQOL in 2022. Determinants were identified in a data-driven manner using recursive partitioning to define homogeneous subgroups, stratified by school level. Median KINDL total score in the empirically identified subgroups ranged from 68 to 83 in primary school children and from 69 to 82 in adolescents in secondary school. The psychological factors sadness, anxiousness, and stress in 2021 were identified as the most important determinants of HRQOL in both primary and secondary school children. Other factors, such as physical activity, screen time, chronic health conditions, or nationality, were determinants only in individual subscales.

CONCLUSION

Recent mental health, more than biological, physical, or social factors, played a key role in determining HRQOL in children and adolescents during pandemic times. Public health strategies to improve mental health may therefore be effective in improving HRQOL in this age group.

WHAT IS KNOWN

• Assessing health-related quality of life (HRQOL) in children and adolescents aids in understanding how life circumstances impact their well-being. • HRQOL is a complex construct, involving biological, psychological, and social factors. Factors driving HRQOL in children and adolescents are not often studied in longitudinal population-based samples.

WHAT IS NEW

• Mental health (stress, anxiousness, sadness) played a key role in determining HRQOL during the coronavirus pandemic, more than biological or social factors. • Public health strategies to improve mental health may be effective in improving HRQOL in children.

Dropout from exercise trials among cancer survivors-An individual patient data meta-analysis from the POLARIS study

INTRODUCTION

The number of randomized controlled trials (RCTs) investigating the effects of exercise among cancer survivors has increased in recent years; however, participants dropping out of the trials are rarely described. The objective of the present study was to assess which combinations of participant and exercise program characteristics were associated with dropout from the exercise arms of RCTs among cancer survivors.

METHODS

This study used data collected in the Predicting OptimaL cAncer RehabIlitation and Supportive care (POLARIS) study, an international database of RCTs investigating the effects of exercise among cancer survivors. Thirty-four exercise trials, with a total of 2467 patients without metastatic disease randomized to an exercise arm were included. Harmonized studies included a pre and a posttest, and participants were classified as dropouts when missing all assessments at the post-intervention test. Subgroups were identified with a conditional inference tree.

RESULTS

Overall, 9.6% of the participants dropped out. Five subgroups were identified in the conditional inference tree based on four significant associations with dropout. Most dropout was observed for participants with BMI >28.4 kg/m2 , performing supervised resistance or unsupervised mixed exercise (19.8% dropout) or had low-medium education and performed aerobic or supervised mixed exercise (13.5%). The lowest dropout was found for participants with BMI >28.4 kg/m2 and high education performing aerobic or supervised mixed exercise (5.1%), and participants with BMI ≤28.4 kg/m2 exercising during (5.2%) or post (9.5%) treatment.

CONCLUSIONS

There are several systematic differences between cancer survivors completing and dropping out from exercise trials, possibly affecting the external validity of exercise effects.

Intersectional analysis of inequalities in self-reported breast cancer screening attendance using supervised machine learning and PROGRESS-Plus framework

Background

Breast cancer is a critical public health concern in Spain, and organized screening programs have been in place since the 1990s to reduce its incidence. However, despite the bi-annual invitation for breast cancer screening (BCS) for women aged 45-69, significant attendance inequalities persist among different population groups. This study employs a quantitative intersectional perspective to identify intersectional positions at risk of not undergoing breast cancer screening in Spain.

Methods

Women were selected from the 2020 European Health Interview Survey in Spain, which surveyed the adult population (> 15 years old) living in private households (N = 22,072; 59% response rate). Inequality indicators based on the PROGRESS-Plus framework were used to disentangle existing social intersections. To identify intersectional groups, decision tree models, including classification and regression trees (CARTs), chi-squared automatic interaction detector (CHAID), conditional inference rees (CITs), and C5.0, along with an ensemble algorithm, extreme gradient boosting (XGBoost), were applied.

Results

XGBoost (AUC 78.8%) identified regional differences (Autonomous Community) as the most important factor for classifying BCS attendance, followed by education, age, and marital status. The C5.0 model (balanced accuracy 81.1%) highlighted that the relative importance of individual characteristics, such as education, marital status, or age, for attendance differs based on women's place of residence and their degree of interaction. The highest risk of not attending BCS was observed among illiterate older women in lower social classes who were born in Spain, were residing in Asturias, Cantabria, Basque Country, Castile and León, Extremadura, Galicia, Madrid, Murcia, La Rioja, or Valencian Community, and were married, divorced, or widowed. Subsequently, the risk of not attending BCS extends to three other groups of women: women living in Ceuta and Melilla; single or legally separated women living in the rest of Spain; and women not born in Spain who were married, divorced, or widowed and not residing in Ceuta or Melilla.

Conclusion

The combined use of decision trees and ensemble algorithms can be a valuable tool in identifying intersectional positions at a higher risk of not utilizing public resources and, thus, can aid substantially in developing targeted interventions to increase BCS attendance.

Exploratory analysis of the knowledge, attitudes and perceptions of healthcare workers about arboviruses in the context of surveillance in the Republic of Guinea

BACKGROUND

The escalating risk and contemporary occurrences of arbovirus infections prompt a critical inquiry into the ability of nations to execute efficient surveillance systems capable to detect, prevent and respond to arbovirus outbreaks. Healthcare workers (HCWs) are the major actors in the surveillance of infectious diseases with epidemic potential. The objective of this study was to evaluate the knowledge, attitudes and perceptions of HCWs regarding arboviruses in the public health facilities of Conakry, Guinea.

METHODS

A cross-sectional survey was conducted during the from December 27, 2022, to January 31, 2023, encompassing from public health facilities in Conakry. The data collection process encompassed various aspects, including the characteristics of health facilities, socio-demographic and professional attributes of HCWs, the information received concerning arboviruses and the sources of information, as well as a series of inquiries designed to evaluate their knowledge, attitudes and perceptions. Subsequently, scores were computed for knowledge, attitude and perception. To identify the factors influencing the knowledge, attitudes, and perceptions of HCWs regarding arboviruses, Decision Tree and Inference Conditional Tree models were used.

RESULTS

A total of 352 HCWs participated in the study, comprising 219 from national hospitals, 72 from municipal hospitals and 61 from primary health centers. More than half of the respondents (54.3%) had never received information on arboviruses. Only 1% of the respondents had good knowledge about arboviruses, 95.7% had a negative attitude about arboviruses. Moreover, nearly 60% of the respondents had a moderate perception and 24.1% had a good perception. The analysis revealed significant associations between the knowledge and attitudes of respondents concerning arboviruses and their years of professional experience and age.

CONCLUSION

This study highlights the imperative requirement for comprehensive training targeting HCWs to enhance their capacity for early case detection within healthcare facilities. Additionally, there is a crucial need for analogous studies adopting a mixed-methods approach across all healthcare regions in Guinea.

Solid waste management and Aedes aegypti infestation interconnections: A regression tree application

Public health is at the core of all environmental and anthropic impacts. Urban and territorial planners should include public health concerns in their plans. Basic sanitation infrastructure is essential to maintaining public health and social and economic development. This infrastructure deficiency causes diseases, death and economic losses in developing countries. Framing interconnections among health, sanitation, urbanization and circular economy will assist sustainable development goal achievements. This study aims to identify the relationships between solid waste management indicators in Brazil and the Aedes aegypti mosquito infestation index. Regression trees were employed for modelling due to the complexity and characteristics of the data. The analyses were performed separately from data collected from 3501 municipalities and 42 indicators from the country's five regions. Results show that expenses and personnel indicators were the most critical indicators (in the mid-western, southeastern and southern regions), operational (northeastern (NE) region) and management (northern region). The mean absolute errors ranged from 0.803 (southern region) to 2.507 (NE region). Regional analyses indicate that the municipalities with better SWM results display lower infestation rates in buildings and residences. This research is innovative as it analyses infestation rates rather than dengue prevalence, using a machine learning method, in a multidisciplinary research field that needs further study.

A machine learning approach to predict the glaucoma filtration surgery outcome

This study aimed at predicting the filtration surgery (FS) outcome using a machine learning (ML) approach. 102 glaucomatous patients undergoing FS were enrolled and underwent ocular surface clinical tests (OSCTs), determination of surgical site-related biometric parameters (SSPs) and conjunctival vascularization. Break-up-time, Schirmer test I, corneal fluorescein staining, Meibomian gland expressibility; conjunctival hyperemia, upper bulbar conjunctiva area of exposure, limbus to superior eyelid distance; and conjunctival epithelial and stromal (CET, CST) thickness and reflectivity (ECR, SCR) at AS-OCT were considered. Successful FS required a 30% baseline intraocular pressure reduction, with values ≤ 18 mmHg with or without medications. The classification tree (CT) was the ML algorithm used to analyze data. At the twelfth month, FS was successful in 60.8% of cases, whereas failed in 39.2%. At the variable importance ranking, CST and SCR were the predictors with the greater relative importance to the CART tree construction, followed by age. CET and ECR showed less relative importance, whereas OSCTs and SSPs were not important features. Within the CT, CST turned out the most important variable for discriminating success from failure, followed by SCR and age, with cut-off values of 75 µm, 169 on gray scale, and 62 years, respectively. The ROC curve for the classifier showed an AUC of 0.784 (0.692-0.860). In this ML approach, CT analysis found that conjunctival stroma thickness and reflectivity, along with age, can predict the FS outcome with good accuracy. A pre-operative thick and hyper-reflective stroma, and a younger age increase the risk of FS failure.

'Forest malaria' in Myanmar? Tracking transmission landscapes in a diversity of environments

BACKGROUND

In the Greater Mekong Subregion, case-control studies and national-level analyses have shown an association between malaria transmission and forest activities. The term 'forest malaria' hides the diversity of ecosystems in the GMS, which likely do not share a uniform malaria risk. To reach malaria elimination goals, it is crucial to document accurately (both spatially and temporally) the influence of environmental factors on malaria to improve resource allocation and policy planning within given areas. The aim of this ecological study is to characterize the association between malaria dynamics and detailed ecological environments determined at village level over a period of several years in Kayin State, Myanmar.

METHODS

We characterized malaria incidence profiles at village scale based on intra- and inter-annual variations in amplitude, seasonality, and trend over 4 years (2016-2020). Environment was described independently of village localization by overlaying a 2-km hexagonal grid over the region. Specifically, hierarchical classification on principal components, using remote sensing data of high spatial resolution, was used to assign a landscape and a climate type to each grid cell. We used conditional inference trees and random forests to study the association between the malaria incidence profile of each village, climate and landscape. Finally, we constructed eco-epidemiological zones to stratify and map malaria risk in the region by summarizing incidence and environment association information.

RESULTS

We identified a high diversity of landscapes (n = 19) corresponding to a gradient from pristine to highly anthropogenically modified landscapes. Within this diversity of landscapes, only three were associated with malaria-affected profiles. These landscapes were composed of a mosaic of dense and sparse forest fragmented by small agricultural patches. A single climate with moderate rainfall and a temperature range suitable for mosquito presence was also associated with malaria-affected profiles. Based on these environmental associations, we identified three eco-epidemiological zones marked by later persistence of Plasmodium falciparum, high Plasmodium vivax incidence after 2018, or a seasonality pattern in the rainy season.

CONCLUSIONS

The term forest malaria covers a multitude of contexts of malaria persistence, dynamics and populations at risk. Intervention planning and surveillance could benefit from consideration of the diversity of landscapes to focus on those specifically associated with malaria transmission.

Using decision tree analysis to identify population groups at risk of subjective unmet need for assistance with activities of daily living

BACKGROUND

Identifying predictors of subjective unmet need for assistance with activities of daily living (ADL) is necessary to allocate resources in social care effectively to the most vulnerable populations. In this study, we aimed at identifying population groups at risk of subjective unmet need for assistance with ADL and instrumental ADL (IADL) taking complex interaction patterns between multiple predictors into account.

METHODS

We included participants aged 55 or older from the cross-sectional German Health Update Study (GEDA 2019/2020-EHIS). Subjective unmet need for assistance was defined as needing any help or more help with ADL (analysis 1) and IADL (analysis 2). Analysis 1 was restricted to participants indicating at least one limitation in ADL (N = 1,957). Similarly, analysis 2 was restricted to participants indicating at least one limitation in IADL (N = 3,801). Conditional inference trees with a Bonferroni-corrected type 1 error rate were used to build classification models of subjective unmet need for assistance with ADL and IADL, respectively. A total of 36 variables representing sociodemographics and impairments of body function were used as covariates for both analyses. In addition, the area under the receiver operating characteristics curve (AUC) was calculated for each decision tree.

RESULTS

Depressive symptoms according to the PHQ-8 was the most important predictor of subjective unmet need for assistance with ADL. Further classifiers that were selected from the 36 independent variables were gender identity, employment status, severity of pain, marital status, and educational level according to ISCED-11. The AUC of this decision tree was 0.66. Similarly, depressive symptoms was the most important predictor of subjective unmet need for assistance with IADL. In this analysis, further classifiers were severity of pain, social support according to the Oslo-3 scale, self-reported prevalent asthma, and gender identity (AUC = 0.63).

CONCLUSIONS

Reporting depressive symptoms was the most important predictor of subjective unmet need for assistance among participants with limitations in ADL or IADL. Our findings do not allow conclusions on causal relationships. Predictive performance of the decision trees should be further investigated before conclusions for practice can be drawn.

Hepatic Involvement across the Metabolic Syndrome Spectrum: Non-Invasive Assessment and Risk Prediction Using Machine Learning

Metabolic-dysfunction-associated steatotic liver disease (MASLD) and metabolic syndrome (MetS) are inextricably linked conditions, both of which are experiencing an upward trend in prevalence, thereby exerting a substantial clinical and economic burden. The presence of MetS should prompt the search for metabolic-associated liver disease. Liver fibrosis is the main predictor of liver-related morbidity and mortality. Non-invasive tests (NIT) such as the Fibrosis-4 index (FIB4), aspartate aminotransferase-to-platelet ratio index (APRI), aspartate aminotransferase-to-alanine aminotransferase ratio (AAR), hepatic steatosis index (HIS), transient elastography (TE), and combined scores (AGILE3+, AGILE4) facilitate the detection of liver fibrosis or steatosis. Our study enrolled 217 patients with suspected MASLD, 109 of whom were diagnosed with MetS. We implemented clinical and biological evaluations complemented by transient elastography (TE) to discern the most robust predictors for liver disease manifestation patterns. Patients with MetS had significantly higher values of FIB4, APRI, HSI, liver stiffness, and steatosis parameters measured by TE, as well as AGILE3+ and AGILE4 scores. Machine-learning algorithms enhanced our evaluation. A two-step cluster algorithm yielded three clusters with reliable model quality. Cluster 1 contained patients without significant fibrosis or steatosis, while clusters 2 and 3 showed a higher prevalence of significant liver fibrosis or at least moderate steatosis as measured by TE. A decision tree algorithm identified age, BMI, liver enzyme levels, and metabolic syndrome characteristics as significant factors in predicting cluster membership with an overall accuracy of 89.4%. Combining NITs improves the accuracy of detecting patterns of liver involvement in patients with suspected MASLD.

Health, Economic, and Social Impacts of Substandard and Falsified Medicines in Low- and Middle-Income Countries: A Systematic Review of Methodological Approaches

Little is known about the adverse health, economic, and social impacts of substandard and falsified medicines (SFMs). This systematic review aimed to identify the methods used in studies to measure the impact of SFMs in low- and middle-income countries (LMICs), summarize their findings, and identify gaps in the reviewed literature. A search of eight databases for published papers, and a manual search of references in the relevant literature were conducted using synonyms of SFMs and LMICs. Studies in the English language that estimated the health, social, or economic impacts of SFMs in LMICs published before June 17, 2022 were considered eligible. Search results generated 1,078 articles, and 11 studies were included after screening and quality assessment. All included studies focused on countries in sub-Saharan Africa. Six studies used the Substandard and Falsified Antimalarials Research Impact model to estimate the impact of SFMs. This model is an important contribution. However, it is technically challenging and data demanding, which poses challenges to its adoption by national academics and policymakers alike. The included studies estimate that substandard and falsified antimalarial medicines can account from 10% to ∼40% of total annual malaria costs, and SFMs affect rural and poor populations disproportionately. Evidence on the impact of SFMs is limited in general and nonexistent regarding social outcomes. Further research needs to focus on practical methods that can serve local authorities without major investments in terms of technical capacity and data collection.

Fever and health-seeking behaviour among migrants living along the Thai-Myanmar border: a mixed-methods study

BACKGROUND

Fever is a common reason to seek healthcare in Southeast Asia, and the decline of malaria has complexified how is perceived, and what actions are taken towards it. We investigated the concept of fever and the determinants influencing health-seeking behaviours among migrants on the Thai-Myanmar border, where rapid economic development collides with precarious political and socio-economic conditions.

METHODS

We implemented a mixed-methods study between August to December 2019. Phase I used a qualitative approach, with in-depth interviews and focus group discussions. Phase II used a quantitative approach with a close-ended questionnaire based on Phase I findings. A conditional inference tree (CIT) model first identified geographic and socio-demographic determinants, which were then tested using a logistic regression model.

RESULTS

Fever corresponded to a high diversity of conceptions, symptoms and believed causes. Self-medication was the commonest behaviour at fever onset. If fever persisted, migrants primarily sought care in humanitarian cost-free clinics (45.5%, 92/202), followed by private clinics (43.1%, 87/202), health posts (36.1%, 73/202), public hospitals (33.7%, 68/202) and primary care units (30, 14.9%). The qualitative analysis identified distance and legal status as key barriers for accessing health care. The quantitative analysis further investigated determinants influencing health-seeking behaviour: living near a town where a cost-free clinic operated was inversely associated with seeking care at health posts (adjusted odds ratio [aOR], 0.40, 95% confidence interval [95% CI] [0.19-0.86]), and public hospital attendance (aOR 0.31, 95% CI [0.14-0.67]). Living further away from the nearest town was associated with health posts attendance (aOR 1.05, 95% CI [1.00-1.10] per 1 km). Having legal status was inversely associated with cost-free clinics attendance (aOR 0.27, 95% CI [0.10-0.71]), and positively associated with private clinic and public hospital attendance (aOR 2.56, 95% CI [1.00-6.54] and 5.15, 95% CI [1.80-14.71], respectively).

CONCLUSIONS

Fever conception and believed causes are context-specific and should be investigated prior to any intervention. Distance to care and legal status were key determinants influencing health-seeking behaviour. Current economic upheavals are accelerating the unregulated flow of undocumented migrants from Myanmar to Thailand, warranting further inclusiveness and investments in the public health system.

Practical Guide to Honest Causal Forests for Identifying Heterogeneous Treatment Effects

"Heterogeneous treatment effects" is a term which refers to conditional average treatment effects (i.e., CATEs) that vary across population subgroups. Epidemiologists are often interested in estimating such effects because they can help detect populations that may particularly benefit from or be harmed by a treatment. However, standard regression approaches for estimating heterogeneous effects are limited by preexisting hypotheses, test a single effect modifier at a time, and are subject to the multiple-comparisons problem. In this article, we aim to offer a practical guide to honest causal forests, an ensemble tree-based learning method which can discover as well as estimate heterogeneous treatment effects using a data-driven approach. We discuss the fundamentals of tree-based methods, describe how honest causal forests can identify and estimate heterogeneous effects, and demonstrate an implementation of this method using simulated data. Our implementation highlights the steps required to simulate data sets, build honest causal forests, and assess model performance across a variety of simulation scenarios. Overall, this paper is intended for epidemiologists and other population health researchers who lack an extensive background in machine learning yet are interested in utilizing an emerging method for identifying and estimating heterogeneous treatment effects.

Evaluation of maternal and child care continuum in Guinea: a secondary analysis of two demographic and health surveys using the composite coverage index (CCI)

INTRODUCTION

The composite coverage index (CCI) is the weighted average coverage of eight preventive and curative interventions received along the maternal and childcare continuum. This study aimed to analyse maternal and child health indicators using CCI.

METHODS

We performed a secondary analysis of demographic and health surveys (DHS) focused on women aged 15 to 49 and their children aged 1 to 4. This study took place in Guinea. The CCI (meeting the need for planning, childbirth assisted by qualified healthcare workers, antenatal care assisted by qualified healthcare workers, vaccination against diphtheria, pertussis, tetanus, measles and Bacillus Calmette-Guérin, taking oral rehydration salts during diarrhoea and seeking care for pneumonia) is optimal if the weighted proportion of interventions is > 50%; otherwise, it is partial. We identified the factors associated with CCI using the descriptive association tests, the spatial autocorrelation statistic and multivariate logistic regression.

RESULTS

The analyses involved two DHS surveys, with 3034 included in 2012 and 4212 in 2018. The optimal coverage of the CCI has increased from 43% in 2012 to 61% in 2018. In multivariate analysis, in 2012: the poor had a lower probability of having an optimal CCI than the richest; OR = 0.11 [95% CI; 0.07, 0.18]. Those who had done four antenatal care visits (ANC) were 2.78 times more likely to have an optimal CCI than those with less OR = 2.78 [95% CI;2.24, 3.45]. In 2018: the poor had a lower probability of having an optimal CCI than the richest OR = 0.27 [95% CI; 0.19, 0.38]. Women who planned their pregnancies were 28% more likely to have an optimal CCI than those who had not planned OR = 1.28 [95% CI;1.05, 1.56]. Finally, women with more than 4 ANC were 2.43 times more likely to have an optimal CCI than those with the least OR = 2.43 [95% CI; 2.03, 2.90]. The spatial analysis reveals significant disparities with an aggregation of high partial CCI in Labé between 2012 and 2018.

CONCLUSION

This study showed an increase in CCI between 2012 and 2018. Policies should improve access to care and information for poor women. Besides, strengthening ANC visits and reducing regional inequalities increases optimal CCI.

Perinatal predictors of clinical instability at birth in late-preterm and term infants

To identify characteristics associated with delivery room clinical instability in at-risk infants. Prospective cohort study. Two perinatal centres in Melbourne, Australia. Infants born at ≥ 35⁺⁰ weeks' gestation with a first-line paediatric doctor requested to attend. Clinical instability defined as any one of heart rate < 100 beats per minute for ≥ 20 s in the first 10 min after birth, maximum fraction of inspired oxygen of ≥ 0.70 in the first 10 min after birth, 5-min Apgar score of < 7, intubated in the delivery room or admitted to the neonatal unit for respiratory support. Four hundred and seventy-three infants were included. The median (IQR) gestational age at birth was 39⁺⁴ (38⁺⁴-40⁺⁴) weeks. Eighty (17%) infants met the criteria for clinical instability. Independent risk factors for clinical instability were labour without oxytocin administration, presence of a medical pregnancy complication, difficult extraction at birth and unplanned caesarean section in labour. Decision tree analysis determined that infants at highest risk were those whose mothers did not receive oxytocin during labour (25% risk). Infants at lowest risk were those whose mothers received oxytocin during labour and did not have a medical pregnancy complication (7% risk).

CONCLUSIONS

We identified characteristics associated with clinical instability that may be useful in alerting less experienced clinicians to call for senior assistance early. The decision trees provide intuitive visual aids but require prospective validation.

WHAT IS KNOWN

• First-line clinicians attending at-risk births may need to call senior colleagues for assistance depending on the infant's condition. • Delays in effectively supporting a compromised infant at birth is an important cause of neonatal morbidity and infant-mother separation.

WHAT IS NEW

• This study identifies risk factors for delivery room clinical instability in at-risk infants born at ≥ 35⁺⁰ weeks' gestation. • The decision trees presented provide intuitive visual tools to aid in determining the need for senior paediatric presence.

Examining the use of decision trees in population health surveillance research: an application to youth mental health survey data in the COMPASS study

INTRODUCTION

In population health surveillance research, survey data are commonly analyzed using regression methods; however, these methods have limited ability to examine complex relationships. In contrast, decision tree models are ideally suited for segmenting populations and examining complex interactions among factors, and their use within health research is growing. This article provides a methodological overview of decision trees and their application to youth mental health survey data.

METHODS

The performance of two popular decision tree techniques, the classification and regression tree (CART) and conditional inference tree (CTREE) techniques, is compared to traditional linear and logistic regression models through an application to youth mental health outcomes in the COMPASS study. Data were collected from 74 501 students across 136 schools in Canada. Anxiety, depression and psychosocial well-being outcomes were measured along with 23 sociodemographic and health behaviour predictors. Model performance was assessed using measures of prediction accuracy, parsimony and relative variable importance.

RESULTS

Decision tree and regression models consistently identified the same sets of most important predictors for each outcome, indicating a general level of agreement between methods. Tree models had lower prediction accuracy but were more parsimonious and placed greater relative importance on key differentiating factors.

CONCLUSION

Decision trees provide a means of identifying high-risk subgroups to whom prevention and intervention efforts can be targeted, making them a useful tool to address research questions that cannot be answered by traditional regression methods.

Psychosocial Predictors of Drop-Out from Organised Sport: A Prospective Study in Adolescent Soccer

In recent years an increased drop-out rate in adolescents' soccer participation has been observed. Given the potentially adverse consequences of drop-out from soccer, more information about risk factors for drop-out is warranted. In the current study, Classification and Regression Tree (CRT) analysis was used to investigate demographic and motivational factors associated with an increased risk of drop-out from adolescent soccer. The results of this study indicate that older age, experiencing less autonomy support from the coach, less intrinsic motivation, being female, and lower socioeconomic status are factors associated with an increased risk of drop-out. An interpretation of the results of this study is that coaches play a central part in creating a sports context that facilitates motivation and continued soccer participation. Based on the findings of the current study we propose that soccer clubs implement theoretically informed coach education programs to help coaches adopt autonomy-supportive coaching strategies.

Predicting Low Cognitive Ability at Age 5-Feature Selection Using Machine Learning Methods and Birth Cohort Data

Objectives: In this study, we applied the random forest (RF) algorithm to birth-cohort data to train a model to predict low cognitive ability at 5 years of age and to identify the important predictive features. Methods: Data was from 1,070 participants in the Irish population-based BASELINE cohort. A RF model was trained to predict an intelligence quotient (IQ) score ≤90 at age 5 years using maternal, infant, and sociodemographic features. Feature importance was examined and internal validation performed using 10-fold cross validation repeated 5 times. Results The five most important predictive features were the total years of maternal schooling, infant Apgar score at 1 min, socioeconomic index, maternal BMI, and alcohol consumption in the first trimester. On internal validation a parsimonious RF model based on 11 features showed excellent predictive ability, correctly classifying 95% of participants. This provides a foundation suitable for external validation in an unseen cohort. Conclusion: Machine learning approaches to large existing datasets can provide accurate feature selection to improve risk prediction. Further validation of this model is required in cohorts representative of the general population.

Comparison of machine learning models for bluetongue risk prediction: a seroprevalence study on small ruminants

Background

Bluetongue (BT) is a disease of concern to animal breeders, so the question on their minds is whether they can predict the risk of the disease before it occurs. The main objective of this study is to enhance the accuracy of BT risk prediction by relying on machine learning (ML) approaches to help in fulfilling this inquiry. Several risk factors of BT that affect the occurrence and magnitude of animal infection with the virus have been reported globally. Additionally, risk factors, such as sex, age, species, and season, unevenly affect animal health and welfare. Therefore, the seroprevalence study data of 233 apparently healthy animals (125 sheep and 108 goats) from five different provinces in Egypt were used to analyze and compare the performance of the algorithms in predicting BT risk.

Results

Logistic regression (LR), decision tree (DT), random forest (RF), and a feedforward artificial neural network (ANN) were used to develop predictive BT risk models and compare their performance to the base model (LR). Model performance was assessed by the area under the receiver operating characteristics curve (AUC), accuracy, true positive rate (TPR), false positive rate (FPR), false negative rate (FNR), precision, and F1 score. The results indicated that RF performed better than other models, with an AUC score of 81%, ANN of 79.6%, and DT of 72.85%. In terms of performance and prediction, LR showed a much lower value (AUC = 69%). Upon further observation of the results, it was discovered that age and season were the most important predictor variables reported in classification and prediction.

Conclusion

The findings of this study can be utilized to predict and control BT risk factors in sheep and goats, with better diagnostic discrimination in terms of accuracy, TPR, FNR, FPR, and precision of ML models over traditional and commonly used LR models. Our findings advocate that the implementation of ML algorithms, mainly RF, in farm decision making and prediction is a promising technique for analyzing cross-section studies, providing adequate predictive power and significant competence in identifying and ranking predictors representing potential risk factors for BT.

Machine learning integration of multimodal data identifies key features of blood pressure regulation

BACKGROUND

Association studies have identified several biomarkers for blood pressure and hypertension, but a thorough understanding of their mutual dependencies is lacking. By integrating two different high-throughput datasets, biochemical and dietary data, we aim to understand the multifactorial contributors of blood pressure (BP).

METHODS

We included 4,863 participants from TwinsUK with concurrent BP, metabolomics, genomics, biochemical measures, and dietary data. We used 5-fold cross-validation with the machine learning XGBoost algorithm to identify features of importance in context of one another in TwinsUK (80% training, 20% test). The features tested in TwinsUK were then probed using the same algorithm in an independent dataset of 2,807 individuals from the Qatari Biobank (QBB).

FINDINGS

Our model explained 39·2% [4·5%, MAE:11·32 mmHg (95%CI, +/- 0·65)] of the variance in systolic BP (SBP) in TwinsUK. Of the top 50 features, the most influential non-demographic variables were dihomo-linolenate, cis-4-decenoyl carnitine, lactate, chloride, urate, and creatinine along with dietary intakes of total, trans and saturated fat. We also highlight the incremental value of each included dimension. Furthermore, we replicated our model in the QBB [SBP variance explained = 45·2% (13·39%)] cohort and 30 of the top 50 features overlapped between cohorts.

INTERPRETATION

We show that an integrated analysis of omics, biochemical and dietary data improves our understanding of their in-between relationships and expands the range of potential biomarkers for blood pressure. Our results point to potentially key biological pathways to be prioritised for mechanistic studies.

FUNDING

Chronic Disease Research Foundation, Medical Research Council, Wellcome Trust, Qatar Foundation.

Estimation of Litho-Fluid Facies Distribution from Zero-Offset Acoustic and Shear Impedances

Seismic data are considered crucial sources of data that help identify the litho-fluid facies distributions in reservoir rocks. However, different facies mostly have similar responses to seismic attributes. In addition, seismic anisotropy negatively affects the facies predictors extracted from seismic data. Accordingly, this study aims at estimating zero-offset acoustic and shear impedances based on partial-stack inversion by two methods: statistical modeling and a multilayer feed-forward neural network (MLFN). The resulting impedance volumes are compared to those obtained from isotropic simultaneous inversion by using impedance logs. The best impedance volumes are applied to Thomsen’s anisotropy equations to solve for the anisotropy parameters Epsilon and Delta. Finally, the shear and acoustic impedances are transformed into elastic properties from which the facies and fluid distributions are predicted by using the logistic regression and decision tree algorithms. The results obtained from the MLFN show better matching with the impedance and facies logs compared to those obtained from isotropic inversion and statistical modeling.

Decision Tree Analyses to Explore the Relevance of Multiple Sex/Gender Dimensions for the Exposure to Green Spaces: Results from the KORA INGER Study

Recently, attention has been drawn to the need to integrate sex/gender more comprehensively into environmental health research. Considering theoretical approaches, we define sex/gender as a multidimensional concept based on intersectionality. However, operationalizing sex/gender through multiple covariates requires the usage of statistical methods that are suitable for handling such complex data. We therefore applied two different decision tree approaches: classification and regression trees (CART) and conditional inference trees (CIT). We explored the relevance of multiple sex/gender covariates for the exposure to green spaces, measured both subjectively and objectively. Data from 3742 participants from the Cooperative Health Research in the Region of Augsburg (KORA) study were analyzed within the INGER (Integrating gender into environmental health research) project. We observed that the participants’ financial situation and discrimination experience was relevant for their access to high quality public green spaces, while the urban/rural context was most relevant for the general greenness in the residential environment. None of the covariates operationalizing the individual sex/gender self-concept were relevant for differences in exposure to green spaces. Results were largely consistent for both CART and CIT. Most importantly we showed that decision tree analyses are useful for exploring the relevance of multiple sex/gender dimensions and their interactions for environmental exposures. Further investigations in larger urban areas with less access to public green spaces and with a study population more heterogeneous with respect to age and social disparities may add more information about the relevance of multiple sex/gender dimensions for the exposure to green spaces.

Applied Machine Learning for Spine Surgeons: Predicting Outcome for Patients Undergoing Treatment for Lumbar Disc Herniation Using PRO Data

STUDY DESIGN

Retrospective/prospective study.

OBJECTIVE

Models based on preoperative factors can predict patients' outcome at 1-year follow-up. This study measures the performance of several machine learning (ML) models and compares the results with conventional methods.

METHODS

Inclusion criteria were patients who had lumbar disc herniation (LDH) surgery, identified in the Danish national registry for spine surgery. Initial training of models included 16 independent variables, including demographics and presurgical patient-reported measures. Patients were grouped by reaching minimal clinically important difference or not for EuroQol, Oswestry Disability Index, Visual Analog Scale (VAS) Leg, and VAS Back and by their ability to return to work at 1 year follow-up. Data were randomly split into training, validation, and test sets by 50%/35%/15%. Deep learning, decision trees, random forest, boosted trees, and support vector machines model were trained, and for comparison, multivariate adaptive regression splines (MARS) and logistic regression models were used. Model fit was evaluated by inspecting area under the curve curves and performance during validation.

RESULTS

Seven models were arrived at. Classification errors were within ±1% to 4% SD across validation folds. ML did not yield superior performance compared with conventional models. MARS and deep learning performed consistently well. Discrepancy was greatest among VAS Leg models.

CONCLUSIONS

Five predictive ML and 2 conventional models were developed, predicting improvement for LDH patients at the 1-year follow-up. We demonstrate that it is possible to build an ensemble of models with little effort as a starting point for further model optimization and selection.

Describing Intersectional Health Outcomes: An Evaluation of Data Analysis Methods

BACKGROUND

Intersectionality theoretical frameworks have been increasingly incorporated into quantitative research. A range of methods have been applied to describing outcomes and disparities across large numbers of intersections of social identities or positions, with limited evaluation.

METHODS

Using data simulated to reflect plausible epidemiologic data scenarios, we evaluated methods for intercategorical intersectional analysis of continuous outcomes, including cross-classification, regression with interactions, multilevel analysis of individual heterogeneity (MAIHDA), and decision-tree methods (classification and regression trees [CART], conditional inference trees [CTree], random forest). The primary outcome was estimation accuracy of intersection-specific means. We applied each method to an illustrative example using National Health and Nutrition Examination Study (NHANES) systolic blood pressure data.

RESULTS

When studying high-dimensional intersections at smaller sample sizes, MAIHDA, CTree, and random forest produced more accurate estimates. In large samples, all methods performed similarly except CART, which produced less accurate estimates. For variable selection, CART performed poorly across sample sizes, although random forest performed best. The NHANES example demonstrated that different methods resulted in meaningful differences in systolic blood pressure estimates, highlighting the importance of selecting appropriate methods.

CONCLUSIONS

This study evaluates some of a growing toolbox of methods for describing intersectional health outcomes and disparities. We identified more accurate methods for estimating outcomes for high-dimensional intersections across different sample sizes. As estimation is rarely the only objective for epidemiologists, we highlight different outputs each method creates, and suggest the sequential pairing of methods as a strategy for overcoming certain technical challenges.

COVID-19 Phenotypes and Comorbidity: A Data-Driven, Pattern Recognition Approach Using National Representative Data from the United States

The aim of our study was to determine COVID-19 syndromic phenotypes in a data-driven manner using the survey results based on survey results from Carnegie Mellon University’s Delphi Group. Monthly survey results (>1 million responders per month; 320,326 responders with a certain COVID-19 test status and disease duration <30 days were included in this study) were used sequentially in identifying and validating COVID-19 syndromic phenotypes. Logistic Regression-weighted multiple correspondence analysis (LRW-MCA) was used as a preprocessing procedure, in order to weigh and transform symptoms recorded by the survey to eigenspace coordinates, capturing a total variance of >75%. These scores, along with symptom duration, were subsequently used by the Two Step Clustering algorithm to produce symptom clusters. Post-hoc logistic regression models adjusting for age, gender, and comorbidities and confirmatory linear principal components analyses were used to further explore the data. Model creation, based on August’s 66,165 included responders, was subsequently validated in data from March−December 2020. Five validated COVID-19 syndromes were identified in August: 1. Afebrile (0%), Non-Coughing (0%), Oligosymptomatic (ANCOS); 2. Febrile (100%) Multisymptomatic (FMS); 3. Afebrile (0%) Coughing (100%) Oligosymptomatic (ACOS); 4. Oligosymptomatic with additional self-described symptoms (100%; OSDS); 5. Olfaction/Gustatory Impairment Predominant (100%; OGIP). Our findings indicate that the COVID-19 spectrum may be undetectable when applying current disease definitions focusing on respiratory symptoms alone.

Conditional associations between childhood cat ownership and psychotic experiences in adulthood: A retrospective study

Ownership of cats in childhood has been inconsistently associated with psychosis in adulthood. Parasitic exposure, the putative mechanism of this association, may be more common with rodent-hunting cats, and its association with psychosis may depend on other environmental exposures. We examined the conditional associations between childhood cat ownership and the frequency of psychotic experiences in adulthood. Adults (n = 2206) were recruited in downtown Montreal to complete a survey about childhood cat ownership (non-hunting or rodent-hunting), winter birth, residential moves in childhood, head trauma history, and tobacco smoking. The frequency of psychotic experiences (PE) was measured with the 15-item positive subscale of the Community Assessment of Psychic Experiences. Associations between exposures and PE were examined in linear regressions adjusted for age and sex. Interactions among variables were explored using a conditional inference tree. Rodent-hunting cat ownership was associated with higher PE scores in male participants (vs. non-hunting or no cat ownership: SMD = 0.57; 95% CI: 0.27, 0.86), but not in female participants (SMD = 0.10; 95% CI: -0.18, 0.38). In the conditional inference tree, the highest mean PE score was in the class comprised of non-smokers with >1 residential move, head trauma history, and rodent-hunting cat ownership (n = 22; mean standard score = 0.96). The interaction between rodent-hunting cat ownership and head trauma history was supported by a post-hoc linear regression model. Our findings suggest childhood cat ownership has conditional associations with psychotic experiences in adulthood.

In-hospital Outcomes of Infective Endocarditis from 1978 to 2015: Analysis Through Machine-Learning Techniques

BACKGROUND

Early identification of patients with infective endocarditis (IE) at higher risk for in-hospital mortality is essential to guide management and improve prognosis.

METHODS

A retrospective analysis was conducted of a cohort of patients followed up from 1978 to 2015, classified according to the modified Duke criteria. Clinical parameters, echocardiographic data, and blood cultures were assessed. Techniques of machine learning, such as the classification tree, were used to explain the association between clinical characteristics and in-hospital mortality. Additionally, the log-linear model and graphical random forests (GRaFo) representation were used to assess the degree of dependence among in-hospital outcomes of IE.

RESULTS

This study analyzed 653 patients: 449 (69.0%) with definite IE; 204 (31.0%) with possible IE; mean age, 41.3 ± 19.2 years; 420 (64%) men. Mode of IE acquisition: community-acquired (67.6%), nosocomial (17.0%), undetermined (15.4%). Complications occurred in 547 patients (83.7%), the most frequent being heart failure (47.0%), neurologic complications (30.7%), and dialysis-dependent renal failure (6.5%). In-hospital mortality was 36.0%. The classification tree analysis identified subgroups with higher in-hospital mortality: patients with community-acquired IE and peripheral stigmata on admission; and patients with nosocomial IE. The log-linear model showed that surgical treatment was related to higher in-hospital mortality in patients with neurologic complications.

CONCLUSIONS

The use of a machine-learning model allowed identification of subgroups of patients at higher risk for in-hospital mortality. Peripheral stigmata, nosocomial IE, absence of vegetation, and surgery in the presence of neurologic complications are predictors of fatal outcomes in machine learning-based analysis.

Tree-Values: Selective Inference for Regression Trees

We consider conducting inference on the output of the Classification and Regression Tree (CART) (Breiman et al., 1984) algorithm. A naive approach to inference that does not account for the fact that the tree was estimated from the data will not achieve standard guarantees, such as Type 1 error rate control and nominal coverage. Thus, we propose a selective inference framework for conducting inference on a fitted CART tree. In a nutshell, we condition on the fact that the tree was estimated from the data. We propose a test for the difference in the mean response between a pair of terminal nodes that controls the selective Type 1 error rate, and a confidence interval for the mean response within a single terminal node that attains the nominal selective coverage. Efficient algorithms for computing the necessary conditioning sets are provided. We apply these methods in simulation and to a dataset involving the association between portion control interventions and caloric intake.

The Role of the Results of Functional Tests and Psychological Factors on Prediction of Injuries in Adolescent Female Football Players

Football is a popular sport among adolescent females. Given the rate of injuries in female footballers, identifying factors that can predict injuries are important. These injuries are often caused by complex reasons. The aim of this study was to investigate if the combination of demographic (age, number of training and match play hours/week), psychosocial (perceived stress, adaptive coping strategies) and physiological factors (functional performance) can predict a traumatic injury in adolescent female footballers. A cohort consisting of 419 female football players aged 13-16 years was established. Baseline questionnaires covered potential risk factors for sport injuries, and measurements included football-related functional performance tests. Data were collected prospectively with a weekly online questionnaire for 52 weeks covering, e.g., injuries, training, and match play hours/week. A total of 62% of the players reported at least one traumatic injury during the 52 weeks. The coping strategy "positive reframing" had the strongest association with the risk of traumatic injuries. The combination of more frequent use of the coping strategy, positive reframing, and high levels of physical performance capacity may prevent a traumatic injury in adolescent female footballers. Coaches are encouraged to adopt both physiological and psychological factors when preventing injuries in young female footballers.

Identifying the women most vulnerable to intimate partner violence: A decision tree analysis from 48 low and middle-income countries

BACKGROUND

Primary prevention strategies are needed to reduce high rates of intimate partner violence (IPV) in low- and middle-income countries (LMICs). The effectiveness of population-based approaches may be improved by adding initiatives targeted at the most vulnerable groups and tailored to context-specificities.

METHODS

We applied a decision-tree approach to identify subgroups of women at higher risk of IPV in 48 LMICs and in all countries combined. Data from the most recent Demographic and Health Survey carried out between 2010 and 2019 with available information on IPV and sociodemographic indicators was used. To create the trees, we selected 15 recognized risk factors for IPV in the literature which had a potential for targeting interventions. Exposure to IPV was defined as having experienced physical and/or sexual IPV in the past 12 months.

FINDINGS

In the pooled decision tree, witnessing IPV during childhood, a low or medium empowerment level and alcohol use by the partner were the strongest markers of IPV vulnerability. IPV prevalence amongst the most vulnerable women was 43% compared to 21% in the overall sample. This high-risk group included women who witnessed IPV during childhood and had lower empowerment levels. These were 12% of the population and 1 in 4 women who experienced IPV in the selected LMICs. Across the individual national trees, subnational regions emerged as the most frequent markers of IPV occurrence.

INTERPRETATION

Starting with well-known predictors of IPV, the decision-tree approach provides important insights about subpopulations of women where IPV prevalence is high. This information can help designing targeted interventions. For a large proportion of women who experienced IPV, however, no particular risk factors were identified, emphasizing the need for population wide approaches conducted in parallel, including changing social norms, strengthening laws and policies supporting gender equality and women´s rights as well as guaranteeing women´s access to justice systems and comprehensive health services.

FUNDING

Bill and Melinda Gates Foundation (Grant INV-010051/OPP1199234), Wellcome Trust (Grant Number: 101815/Z/13/Z) and Associação Brasileira de Saúde Coletiva (ABRASCO).

Smart Materials Enabled with Artificial Intelligence for Healthcare Wearables

Contemporary medicine suffers from many shortcomings in terms of successful disease diagnosis and treatment, both of which rely on detection capacity and timing. The lack of effective, reliable, and affordable detection and real‐time monitoring limits the affordability of timely diagnosis and treatment. A new frontier that overcomes these challenges relies on smart health monitoring systems that combine wearable sensors and an analytical modulus. This review presents the latest advances in smart materials for the development of multifunctional wearable sensors while providing a bird's eye‐view of their characteristics, functions, and applications. The review also presents the state‐of‐the‐art on wearables fitted with artificial intelligence (AI) and support systems for clinical decision in early detection and accurate diagnosis of disorders. The ongoing challenges and future prospects for providing personal healthcare with AI‐assisted support systems relating to clinical decisions are presented and discussed.

Reasons for Utilizing Telemedicine during and after the COVID-19 Pandemic: An Internet-Based International Study

The COVID-19 pandemic challenges healthcare services. Concomitantly, this pandemic had a stimulating effect on technological expansions related to telehealth and telemedicine. We sought to elucidate the principal patients' reasons for using telemedicine during the COVID-19 pandemic and the propensity to use it thereafter. Our primary objective was to identify the reasons of the survey participants' disparate attitudes toward the use of telemedicine. We performed an online, multilingual 30-question survey for 14 days during March-April 2021, focusing on the perception and usage of telemedicine and their intent to use it after the pandemic. We analyzed the data to identify the attributes influencing the intent to use telemedicine and built decision trees to highlight the most important related variables. We examined 473 answers: 272 from Israel, 87 from Uruguay, and 114 worldwide. Most participants were women (64.6%), married (63.8%) with 1-2 children (52.9%), and living in urban areas (84.6%). Only a third of the participants intended to continue using telemedicine after the COVID-19 pandemic. Our main findings are that an expected substitution effect, technical proficiency, reduced queueing times, and peer experience are the four major factors in the overall adoption of telemedicine. Specifically, (1) for most participants, the major factor influencing their telemedicine usage is the implicit expectation that such a visit will be a full substitute for an in-person appointment; (2) another factor affecting telemedicine usage by patients is their overall technical proficiency and comfort level in the use of common web-based tools, such as social media, while seeking relevant medical information; (3) time saving as telemedicine can allow for asynchronous communications, thereby reducing physical travel and queuing times at the clinic; and finally (4) some participants have also indicated that telemedicine seems more attractive to them after watching family and friends (peer experience) use it successfully.

Comparing organ donation decisions for next-of-kin versus the self: results of a national survey

OBJECTIVES

Intensive care audits point to family refusal as a major barrier to organ donation. In this study, we sought to understand refusal by accounting for the decision-maker's mindset. This focused on: (1) how decisions compare when made on behalf of a relative (vs the self); and (2) confidence in decisions made for family members.

DESIGN

Cross-sectional survey in Singapore.

SETTING

Participants were recruited from community settings via door-to-door sampling and community eateries.

PARTICIPANTS

973 adults who qualified as organ donors in Singapore.

RESULTS

Although 68.1% of participants were willing to donate their own organs, only 51.8% were willing to donate a relative's organs. Using machine learning, we found that consistency was predicted by: (1) religion, and (2) fears about organ donation. Conversely, participants who were willing to donate their own organs but not their relative's were less driven by these factors, and may instead have resorted to heuristics in decision-making. Finally, we observed how individuals were overconfident in their decision-making abilities: although 78% had never discussed organ donation with their relatives, the large majority expressed high confidence that they would respect their relatives' wishes on death.

CONCLUSIONS

These findings underscore the distinct psychological processes involved when donation decisions are made for family members. Amidst a global shortage of organ donors, addressing the decision-maker's mindset (eg, overconfidence, the use of heuristics) may be key to actualizing potential donors identified in intensive care units.

Occupational stress among farm and ranch operators in the midwestern United States

BACKGROUND

This study used surveillance data from 2018 and 2020 to test the stability of work-related strain symptoms (high stress, sleep deprivation, exhaustion) with demographic factors, work characteristics, and musculoskeletal symptoms among farm and ranch operators in seven midwestern states of the United States.

METHODS

Cross-sectional surveys were conducted among farm and ranch operators in 2018 (n = 4423) and 2020 (n = 3492). Operators were asked whether, in the past 12 months, they experienced extended work periods that resulted in high stress levels, sleep deprivation, exhaustion/fatigue, or other work-related strain symptoms. Covariates included personal and demographic factors, work characteristics, number of injuries, work-related health conditions, and exposures on the operation. Summary statistics were tabulated for explanatory and outcome variables. The classification (decision) tree approach was used to assess what variables would best separate operators with and without reported strain symptoms, based on a set of explanatory variables. Regularized regression was used to generate effect estimates between the work strain variables and explanatory variables.

RESULTS

High stress level, sleep deprivation, and exhaustion were reported more frequently in 2018 than 2020. The classification tree reproduced the 2018 model using 2020 data with approximately 80% accuracy. The mean number of reported MSD symptoms increased slightly from 1.23 in 2018 to 1.41 in 2020. Older age, more time spent in farm work, higher gross farm income (GFI), and MSD symptoms in six body regions (ankles/feet, knees, lower back, neck, shoulders, wrists/hands) were associated with all three work strain symptoms.

CONCLUSIONS

Musculoskeletal pain and discomfort was a strong predictor for stress, sleep deprivation, and exhaustion among farmers and ranchers. This finding indicates that reducing MSD pain and discomfort is beneficial for both physical and mental health.

Classification tree analysis for an intersectionality-informed identification of population groups with non-daily vegetable intake

BACKGROUND

Daily vegetable intake is considered an important behavioural health resource associated with improved immune function and lower incidence of non-communicable disease. Analyses of population-based data show that being female and having a high educational status is most strongly associated with increased vegetable intake. In contrast, men and individuals with a low educational status seem to be most affected by non-daily vegetable intake (non-DVI). From an intersectionality perspective, health inequalities are seen as a consequence of an unequal balance of power such as persisting gender inequality. Unravelling intersections of socially driven aspects underlying inequalities might be achieved by not relying exclusively on the male/female binary, but by considering different facets of gender roles as well. This study aims to analyse possible interactions of sex/gender or sex/gender related aspects with a variety of different socio-cultural, socio-demographic and socio-economic variables with regard to non-DVI as the health-related outcome.

METHOD

Comparative classification tree analyses with classification and regression tree (CART) and conditional inference tree (CIT) as quantitative, non-parametric, exploratory methods for the detection of subgroups with high prevalence of non-DVI were performed. Complete-case analyses (n = 19,512) were based on cross-sectional data from a National Health Telephone Interview Survey conducted in Germany.

RESULTS

The CART-algorithm constructed overall smaller trees when compared to CIT, but the subgroups detected by CART were also detected by CIT. The most strongly differentiating factor for non-DVI, when not considering any further sex/gender related aspects, was the male/female binary with a non-DVI prevalence of 61.7% in men and 42.7% in women. However, the inclusion of further sex/gender related aspects revealed a more heterogenous distribution of non-DVI across the sample, bringing gendered differences in main earner status and being a blue-collar worker to the foreground. In blue-collar workers who do not live with a partner on whom they can rely on financially, the non-DVI prevalence was 69.6% in men and 57.4% in women respectively.

CONCLUSIONS

Public health monitoring and reporting with an intersectionality-informed and gender-equitable perspective might benefit from an integration of further sex/gender related aspects into quantitative analyses in order to detect population subgroups most affected by non-DVI.