Classification and regression tree analysis vs. multivariable linear and logistic regression methods as statistical tools for studying haemophilia

Iron deficiency in new onset heart failure: association with clinical factors and quality of life

AIMS

The prevalence of iron deficiency (ID) in newly diagnosed heart failure (HF) and the progression of ID in patients after initiation of HF therapy are unknown. We aimed to describe the natural trajectory of ID in patients with new onset HF during the first year after HF diagnosis, assessing associations between ID, clinical factors, and quality of life (QoL).

METHODS AND RESULTS

A prospective cohort of patients with new onset HF in hospitals or outpatient clinics at five major hospitals in Stockholm, Sweden, during 2015-2018 were analysed with clinical assessment, electrocardiogram, blood samples including iron levels, Minnesota living with heart failure questionnaire (MLHFQ), and echocardiogram at baseline and after 12 months. Of 547 patients with new-onset HF, 482 (88%) had complete iron data at baseline. Mean age was 70 years (interquartile range 61-77) and 311 (65%) were men; 55% of patients had ejection fraction (EF) ≤ 40%, 19% had EF 41-49%, and 26% had HF with preserved EF (HFpEF). At baseline, 163 patients (34%) had ID defined as ferritin <100 μg/L or ferritin 100-299 μg/L and transferrin saturation <20%. After 12 months of follow-up, 119 (32%) had ID of the 368 patients who had complete iron data both at baseline and after 12 months and did not receive intravenous (i.v.) iron during follow-up. During the first year after HF diagnosis, 19% had persistent ID, 13% developed ID, 11% resolved ID, and 57% never had ID, consequently 24% changed their classification. Anaemia at baseline was the strongest independent predictor of ID 1 year after diagnosis [odds ratio (OR) 3.91, 95% confidence interval (CI) 1.88-8.13, P < 0.001], followed by HF hospitalization (OR 2.21, 95% CI 1.24-3.95, P < 0.01), female sex (OR 2.04, 95% CI 1.25-3.32, P < 0.01), HFpEF (OR 1.96, 95% CI 1.13-3.39, P < 0.05), and diabetes mellitus (OR 1.92, 95% CI 1.06-3.48, P < 0.05). ID was associated with low QoL at baseline (MLHFQ score mean difference 7.4 points, 95% CI 3.1-11.7, P < 0.001), but not at follow-up.

CONCLUSIONS

About one third of patients with new onset HF had ID both at the time of HF diagnosis and after 1 year, though a quarter of the patients changed their ID status. Patients with anaemia, HF hospitalization, female gender, HFpEF, or diabetes mellitus at baseline were more likely to have ID after 1 year implying that these should be carefully screened for ID to find those in need of i.v. iron treatment.

Diagnostic value of 18F-fluorodeoxyglucose positron emission tomography and computed tomography for differentiating polymyalgia rheumatica and rheumatoid arthritis: Using classification and regression tree analysis

OBJECTIVES

Determining which sites were important to differentiate polymyalgia rheumatica (PMR) from rheumatoid arthritis (RA) using 18F-fluorodeoxyglucose (FDG) positron emission tomography and computed tomography (PET-CT) is challenging.

METHODS

Patients with PMR or RA who were undergoing PET-CT were recruited at two mutual-aid hospitals in Japan between 2009 and 2018. Classification and regression tree (CART) analyses were performed to identify FDG uptake patterns that differentiated PMR from RA.

RESULTS

We enrolled 35 patients with PMR and 46 patients with RA. Univariate CART analysis showed that FDG uptake in the shoulder joints, spinous processes of the lumbar vertebrae, pubic symphysis, sternoclavicular joints, ischial tuberosities, greater trochanters, and hip joints differentiated PMR from RA. Multivariate CART analysis revealed that FDG uptake by at least one of the ischial tuberosities had the highest diagnostic value for distinguishing PMR from RA (sensitivity, 77.1%; specificity, 82.6%). We performed the same CART analysis to patients who had not undergone treatment (PMR, n = 28; RA, n = 9). Similar results were obtained, and sensitivity and specificity were increased (sensitivity, 89.3%; specificity, 88.8%).

CONCLUSIONS

In PET-CT, FDG uptake by at least one of the ischial tuberosities best discriminates between PMR and RA.

Evolution of in-hospital patient characteristics and predictors of death in the COVID-19 pandemic across four waves: are they moving targets with implications for patient care?

Objectives

The aim of this work was to study characteristics, outcomes and predictors of all-cause death in inpatients with SARS-CoV-2 infection across the pandemic waves in one large teaching hospital in Italy to optimize disease management.

Methods

All patients with SARS-CoV-2 infection admitted to our center from March 2020 to June 2022 were included in this retrospective observational cohort study. Both descriptive and regression tree analyses were applied to identify factors influencing all-cause mortality.

Results

527 patients were included in the study (65.3% with moderate and 34.7% with severe COVID-19). Significant evolutions of patient characteristics were found, and mortality increased in the last wave with respect to the third wave notwithstanding vaccination. Regression tree analysis showed that in-patients with severe COVID-19 had the greatest mortality across all waves, especially the older adults, while prognosis depended on the pandemic waves in patients with moderate COVID-19: during the first wave, dyspnea was the main predictor, while chronic kidney disease emerged as determinant factor afterwards.

Conclusion

Patients with severe COVID-19, especially the older adults during all waves, as well as those with moderate COVID-19 and concomitant chronic kidney disease during the most recent waves require more attention for monitoring and care. Therefore, our study drives attention towards the importance of co-morbidities and their clinical impact in patients with COVID-19 admitted to hospital, indicating that the healthcare system should adapt to the evolving features of the epidemic.

A cohort study of factors influencing the physical fitness of preschool children: a decision tree analysis

Objective

Based on the decision tree model, to explore the key influencing factors of children's physical fitness, rank the key influencing factors, and explain the complex interaction between the influencing factors.

Methods

A cohort study design was adopted. 1,276 children (ages 3-6) from 23 kindergartens in Nanchang, China, were chosen for the study to measure the children's physical fitness at baseline and a year later and to compare the physical fitness scores at the two stages. The study was conducted following the Chinese National Physical Fitness Testing Standard (Children Part); To identify the primary influencing factors of changes in physical fitness, a decision tree model was developed, and a questionnaire survey on birth information, feeding patterns, SB, PA, dietary nutrition, sleep, parental factors, and other relevant information was conducted.

Results

The levels of physical fitness indicators among preschool children showed a significant increase after 1 year. The accuracy of the CHAID model is 84.17%. It showed that 7 variables were strongly correlated with the physical changes of children's fitness, the order of importance of each variable was weekend PA, weekend MVPA, mother's BMI, mother's sports frequency, father's education, mother's education, and school day PA. Three factors are related to PA. Four factors are related to parental circumstances. In addition to the seven important variables mentioned, variables such as breakfast frequency on school day, puffed food, frequency of outing, school day MVPA, parental feeling of sports, father's occupation, and weekend breakfast frequency are all statistically significant leaf node variables.

Conclusion

PA, especially weekend PA, is the most critical factor in children's physical fitness improvement and the weekend MVPA should be increased to more than 30 min/d based on the improvement of weekend PA. In addition, parental factors and school day PA are also important in making decisions about changes in fitness for children. The mother's efforts to maintain a healthy BMI and engage in regular physical activity are crucial for enhancing the physical fitness of children. Additionally, other parental factors, such as the parents' educational levels and the father's occupation, can indirectly impact the level of physical fitness in children.

Improvement of predictive accuracies of functional outcomes after subacute stroke inpatient rehabilitation by machine learning models

OBJECTIVES

Stepwise linear regression (SLR) is the most common approach to predicting activities of daily living at discharge with the Functional Independence Measure (FIM) in stroke patients, but noisy nonlinear clinical data decrease the predictive accuracies of SLR. Machine learning is gaining attention in the medical field for such nonlinear data. Previous studies reported that machine learning models, regression tree (RT), ensemble learning (EL), artificial neural networks (ANNs), support vector regression (SVR), and Gaussian process regression (GPR), are robust to such data and increase predictive accuracies. This study aimed to compare the predictive accuracies of SLR and these machine learning models for FIM scores in stroke patients.

METHODS

Subacute stroke patients (N = 1,046) who underwent inpatient rehabilitation participated in this study. Only patients' background characteristics and FIM scores at admission were used to build each predictive model of SLR, RT, EL, ANN, SVR, and GPR with 10-fold cross-validation. The coefficient of determination (R2) and root mean square error (RMSE) values were compared between the actual and predicted discharge FIM scores and FIM gain.

RESULTS

Machine learning models (R2 of RT = 0.75, EL = 0.78, ANN = 0.81, SVR = 0.80, GPR = 0.81) outperformed SLR (0.70) to predict discharge FIM motor scores. The predictive accuracies of machine learning methods for FIM total gain (R2 of RT = 0.48, EL = 0.51, ANN = 0.50, SVR = 0.51, GPR = 0.54) were also better than of SLR (0.22).

CONCLUSIONS

This study suggested that the machine learning models outperformed SLR for predicting FIM prognosis. The machine learning models used only patients' background characteristics and FIM scores at admission and more accurately predicted FIM gain than previous studies. ANN, SVR, and GPR outperformed RT and EL. GPR could have the best predictive accuracy for FIM prognosis.

Machine Learning and Causal Approaches to Predict Readmissions and Its Economic Consequences Among Canadian Patients With Heart Disease: Retrospective Study

BACKGROUND

Unplanned patient readmissions within 30 days of discharge pose a substantial challenge in Canadian health care economics. To address this issue, risk stratification, machine learning, and linear regression paradigms have been proposed as potential predictive solutions. Ensemble machine learning methods, such as stacked ensemble models with boosted tree algorithms, have shown promise for early risk identification in specific patient groups.

OBJECTIVE

This study aims to implement an ensemble model with submodels for structured data, compare metrics, evaluate the impact of optimized data manipulation with principal component analysis on shorter readmissions, and quantitatively verify the causal relationship between expected length of stay (ELOS) and resource intensity weight (RIW) value for a comprehensive economic perspective.

METHODS

This retrospective study used Python 3.9 and streamlined libraries to analyze data obtained from the Discharge Abstract Database covering 2016 to 2021. The study used 2 sub-data sets, clinical and geographical data sets, to predict patient readmission and analyze its economic implications, respectively. A stacking classifier ensemble model was used after principal component analysis to predict patient readmission. Linear regression was performed to determine the relationship between RIW and ELOS.

RESULTS

The ensemble model achieved precision and slightly higher recall (0.49 and 0.68), indicating a higher instance of false positives. The model was able to predict cases better than other models in the literature. Per the ensemble model, readmitted women and men aged 40 to 44 and 35 to 39 years, respectively, were more likely to use resources. The regression tables verified the causality of the model and confirmed the trend that patient readmission is much more costly than continued hospital stay without discharge for both the patient and health care system.

CONCLUSIONS

This study validates the use of hybrid ensemble models for predicting economic cost models in health care with the goal of reducing the bureaucratic and utility costs associated with hospital readmissions. The availability of robust and efficient predictive models, as demonstrated in this study, can help hospitals focus more on patient care while maintaining low economic costs. This study predicts the relationship between ELOS and RIW, which can indirectly impact patient outcomes by reducing administrative tasks and physicians' burden, thereby reducing the cost burdens placed on patients. It is recommended that changes to the general ensemble model and linear regressions be made to analyze new numerical data for predicting hospital costs. Ultimately, the proposed work hopes to emphasize the advantages of implementing hybrid ensemble models in forecasting health care economic cost models, empowering hospitals to prioritize patient care while simultaneously decreasing administrative and bureaucratic expenses.

Population Pharmacokinetics and Dose Evaluation of Cycloserine among Patients with Multidrug-Resistant Tuberculosis under Standardized Treatment Regimens

Although cycloserine is a recommended drug for the treatment of multidrug-resistant tuberculosis (MDR-TB) according to World Health Organization (WHO), few studies have reported on pharmacokinetics (PK) and/or pharmacodynamics (PD) data of cycloserine in patients with standardized MDR-TB treatment. This study aimed to estimate the population PK parameters for cycloserine and to identify clinically relevant PK/PD thresholds, as well as to evaluate the current recommended dosage. Data from a large cohort with full PK curves was used to develop a population PK model. This model was used to estimate drug exposure in patients with MDR-TB from a multicentre prospective study in China. The classification and regression tree was used to identify the clinically relevant PK/PD thresholds. Probability of target attainment was analyzed to evaluate the currently recommended dosing strategy. Cycloserine was best described by a two-compartment disposition model. A percentage of time concentration above MICs (T>MIC) of 30% and a ratio of area under drug concentration-time curve (AUC0-24h) over MIC of 36 were the valid predictors for 6-month sputum culture conversion and final treatment outcome. Simulations showed that with WHO-recommended doses (500 mg and 750 mg for patients weighing <45 kg and ≥45 kg), the probability of target attainment exceeded 90% at MIC ≤16 mg/L in MGIT for both T>MIC of 30% and AUC0-24h/MIC of 36. New clinically relevant PK/PD thresholds for cycloserine were identified in patients with standardized MDR-TB treatment. WHO-recommended doses were considered adequate for the MGIT MIC distribution in our cohort of Chinese patients with MDR-TB.

Influencing factors for decision-making delay in seeking medical care among acute ischemic stroke patients in rural areas

OBJECTIVE

To explore the factors influencing decision-making delay in seeking medical care for patients with acute ischemic stroke (AIS) in rural areas.

METHODS

From September to December 2021, we conducted a questionnaire survey of 260 patients with AIS who were hospitalized in the neurology departments of four county-level hospitals in Daqing. A decision-tree and logistic regression model was used to investigate the elements contributing to decision-making delays.

RESULTS

This study found that the decision-making delay rate for rural patients with AIS was 71.5%. The results of the univariate analysis showed that factors associated with decision-making delay included educational level, National Institute of Health stroke scale (NIHSS) score, self-assessed health, monthly income, social support, attitude toward medical help-seeking, health belief, and family dynamics (P < 0.05). Further, we combined logistic regression (LR) and decision-tree (DT) models for multivariate analysis, and finally obtained five factors affecting decision-making delay in AIS patients in rural areas: disease severity, health belief, monthly income (common factors), educational level (only DT model), and social support (only LR model).

CONCLUSIONS

This study found that a few variables, including disease severity, educational level, monthly income, health belief, and social support, affected rural AIS patients' decision-making delay in seeking medical care.

PRACTICE IMPLICATIONS

To achieve the goal of reducing decision-delay and increasing thrombolysis rate, this study thoroughly examined the influencing factors of decision-making delay in seeking medical care of rural AIS patients from various angles. This analysis provides guidance for medical and healthcare professionals on how to best provide future health education for the high-risk population for stroke in rural areas.

Volumetric regional MRI and neuropsychological predictors of motor task variability in cognitively unimpaired, Mild Cognitive Impairment, and probable Alzheimer's disease older adults

INTRODUCTION

The mechanisms linking motor function to Alzheimer's disease (AD) progression have not been well studied, despite evidence of AD pathology within motor brain regions. Thus, there is a need for new motor measure that is sensitive and specific to AD.

METHODS

In a sample of 121 older adults (54 cognitive unimpaired [CU], 35 amnestic Mild Cognitive Impairment [aMCI], and 32 probable mild AD), intrasubject standard deviation (ISD) across six trials of a novel upper-extremity motor task was predicted with volumetric regional gray matter and neuropsychological scores using classification and regression tree (CART) analyses.

RESULTS

Both gray matter and neuropsychological CART models indicated that motor task ISD (our measure of motor learning) was related to cortical regions and cognitive test scores associated with memory, executive function, and visuospatial skills. CART models also accurately distinguished motor task ISD of MCI and probable mild AD from CU.

DISCUSSION

Variability in motor task performance across practice trials may be valuable for understanding preclinical and early-stage AD.

CART model to classify the drought status of diverse tomato genotypes by VPD, air temperature, and leaf-air temperature difference

Regular water management is crucial for the cultivation of tomato (Solanum lycopersicum L.). Inadequate irrigation leads to water stress and a reduction in tomato yield and quality. Therefore, it is important to develop an efficient classification method of the drought status of tomato for the timely application of irrigation. In this study, a simple classification and regression tree (CART) model that includes air temperature, vapor pressure deficit, and leaf-air temperature difference was established to classify the drought status of three tomato genotypes (i.e., cherry type 'Tainan ASVEG No. 19', large fruits breeding line '108290', and wild accession 'LA2093'). The results indicate that the proposed CART model exhibited a higher predictive sensitivity, specificity, geometric mean, and accuracy performance compared to the logistic model. In addition, the CART model was applicable not only to three tomato genotypes but across vegetative and reproductive stages. Furthermore, while the drought status was divided into low, medium, and high, the CART model provided a higher predictive performance than that of the logistic model. The results suggest that the drought status of tomato can be accurately classified by the proposed CART model. These results will provide a useful tool of the regular water management for tomato cultivation.

The influencing factors of individual interest in physical education based on decision tree model: A cross-sectional study

To identify the key influencing factors and analyze the internal relationship among the factors of individual interest in PE, we conducted a cross-sectional survey of a large sample of Chinese young students based on the decision tree model. A total of 3,640 young students (M_age = 14.16; 7–18 years; SD = 2.66, 47% boys) were investigated by using six questionnaires, including individual interest in physical PE, self-efficacy, achievement goals, expectancy value in PE, PE knowledge and skills and PE learning environment. Results showed there were a total of seven variables entered into the decision tree model, which was 3 layers high, including 38 nodes. The root node was expectancy value which was divided by sports knowledge and skills and self-efficacy. The third layer included mastery-approach goal, family sports environment, performance-avoidance goal and gender. The results depict that expectancy value of PE was the most important influencing factors of adolescent students’ individual interest in PE in this study, and the other important factors were sports knowledge and skills, self-efficacy, mastery-approach goal, family sports environment, performance-avoidance goal, and gender, respectively. The implications for PE are: (1) Improve the status of the PE curriculum and enhance students’ recognition of the value of PE; (2) Strengthen the teaching of knowledge and skills to avoid low-level repetitive teaching; (3) Enhance success experience and foster sports self-efficacy; and (4) Establish reasonable sports goals to foster individual interest in sports learning.

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.

Classification and regression tree (CART) model to assist clinical prediction for tracheostomy in patients with traumatic cervical spinal cord injury: a 7-year study of 340 patients

OBJECTIVE

To develop a classification and regression tree (CART) model to predict the need of tracheostomy in patients with traumatic cervical spinal cord injury (TCSCI) and to quantify scores of risk factors to make individualized clinical assessments.

METHODS

The clinical characteristics of patients with TCSCI admitted to our hospital from January 2014 to December 2020 were retrospectively analyzed. The demographic characteristics (gender, age, smoking history), mechanism of injury, injury characteristics (ASIA impairment grades, neurological level of impairment, injury severity score), preexisting lung disease and preexisting medical conditions were statistically analyzed. The risk factors of tracheostomy were analyzed by univariate logistic regression analysis (ULRA) and multiple logistic regression analysis (MLRA). The CART model was established to predict tracheostomy.

RESULTS

Three hundred and forty patients with TCSCI met the inclusion criteria, in which 41 patients underwent the tracheostomy. ULRA and MLRA showed that age > 50, ISS > 16, NLI > C5 and AIS A were significantly associated with tracheostomy. The CART model showed that AIS A and NLI > C5 were at the first and second decision node, which had a significant influence on the decision of tracheostomy. The final scores for tracheostomy from CART algorithm, composed of age, ISS, NLI and AIS A with a sensitivity of 0.78 and a specificity of 0.96, could also predict tracheostomy.

CONCLUSION

The establishment of CART model provided a certain clinical guidance for the prediction of tracheostomy in TCSCI. Quantifications of risk factors enable accurate prediction of individual patient risk of need for tracheostomy.

"Decision tree analysis for assessing the risk of post-traumatic haemorrhage after mild traumatic brain injury in patients on oral anticoagulant therapy"

BACKGROUND

The presence of oral anticoagulant therapy (OAT) alone, regardless of patient condition, is an indication for CT imaging in patients with mild traumatic brain injury (MTBI). Currently, no specific clinical decision rules are available for OAT patients. The aim of the study was to identify which clinical risk factors easily identifiable at first ED evaluation may be associated with an increased risk of post-traumatic intracranial haemorrhage (ICH) in OAT patients who suffered an MTBI.

METHODS

Three thousand fifty-four patients in OAT with MTBI from four Italian centers were retrospectively considered. A decision tree analysis using the classification and regression tree (CART) method was conducted to evaluate both the pre- and post-traumatic clinical risk factors most associated with the presence of post-traumatic ICH after MTBI and their possible role in determining the patient's risk. The decision tree analysis used all clinical risk factors identified at the first ED evaluation as input predictor variables.

RESULTS

ICH following MTBI was present in 9.5% of patients (290/3054). The CART model created a decision tree using 5 risk factors, post-traumatic amnesia, post-traumatic transitory loss of consciousness, greater trauma dynamic, GCS less than 15, evidence of trauma above the clavicles, capable of stratifying patients into different increasing levels of ICH risk (from 2.5 to 61.4%). The absence of concussion and neurological alteration at admission appears to significantly reduce the possible presence of ICH.

CONCLUSIONS

The machine-learning-based CART model identified distinct prognostic groups of patients with distinct outcomes according to on clinical risk factors. Decision trees can be useful as guidance in patient selection and risk stratification of patients in OAT with MTBI.

Machine Learning Approaches in Traditional Chinese Medicine: A Systematic Review

Machine learning (ML), as a branch of artificial intelligence, acquires the potential and meaningful rules from the mass of data via diverse algorithms. Owing to all research of traditional Chinese medicine (TCM) belonging to the digitalization of clinical records or experimental works, a massive and complex amount of data has become an inextricable part of the related studies. It is thus not surprising that ML approaches, as novel and efficient tools to mine the useful knowledge from data, have created inroads in a diversity of scopes of TCM over the past decade of years. However, by browsing lots of literature, we find that not all of the ML approaches perform well in the same field. Upon further consideration, we infer that the specificity may inhere between the ML approaches and their applied fields. This systematic review focuses its attention on the four categories of ML approaches and their eight application scopes in TCM. According to the function, ML approaches are classified into four categories, including classification, regression, clustering, and dimensionality reduction, and into 14 models as follows in more detail: support vector machine, least square-support vector machine, logistic regression, partial least squares regression, k-means clustering, hierarchical cluster analysis, artificial neural network, back propagation neural network, convolutional neural network, decision tree, random forest, principal component analysis, partial least squares-discriminant analysis, and orthogonal partial least squares-discriminant analysis. The eight common applied fields are divided into two parts: one for TCM, such as the diagnosis of diseases, the determination of syndromes, and the analysis of prescription, and the other for the related researches of Chinese herbal medicine, such as the quality control, the identification of geographic origins, the pharmacodynamic material basis, the medicinal properties, and the pharmacokinetics and pharmacodynamics. Additionally, this paper discusses the function and feature difference among ML approaches when they are applied to the corresponding fields via comparing their principles. The specificity of each approach to its applied fields has also been affirmed, whereby laying a foundation for subsequent studies applying ML approaches to TCM.

Predicting medical usage rate at mass gathering events in Belgium: development and validation of a nonlinear multivariable regression model

Background

Every year, volunteers of the Belgian Red Cross provide onsite medical care at more than 8000 mass gathering events and other manifestations. Today standardized planning tools for optimal preventive medical resource use during these events are lacking. This study aimed to develop and validate a prediction model of patient presentation rate (PPR) and transfer to hospital rate (TTHR) at mass gatherings in Belgium.

Methods

More than 200,000 medical interventions from 2006 to 2018 were pooled in a database. We used a subset of 28 different mass gatherings (194 unique events) to develop a nonlinear prediction model. Using regression trees, we identified potential predictors for PPR and TTHR at these mass gatherings. The additional effect of ambient temperature was studied by linear regression analysis. Finally, we validated the prediction models using two other subsets of the database.

Results

The regression tree for PPR consisted of 7 splits, with mass gathering category as the most important predictor variable. Other predictor variables were attendance, number of days, and age class. Ambient temperature was positively associated with PPR at outdoor events in summer. Calibration of the model revealed an R² of 0.68 (95% confidence interval 0.60–0.75). For TTHR, the most determining predictor variables were mass gathering category and predicted PPR (R² = 0.48). External validation indicated limited predictive value for other events (R² = 0.02 for PPR; R² = 0.03 for TTHR).

Conclusions

Our nonlinear model performed well in predicting PPR at the events used to build the model on, but had poor predictive value for other mass gatherings. The mass gathering categories “outdoor music” and “sports event” warrant further splitting in subcategories, and variables such as attendance, temperature and resource deployment need to be better recorded in the future to optimize prediction of medical usage rates, and hence, of resources needed for onsite emergency medical care.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-022-12580-8.

Utilization of Decision Tree Algorithms for Supporting the Prediction of Intensive Care Unit Admission of Myasthenia Gravis: A Machine Learning-Based Approach

Myasthenia gravis (MG), an acquired autoimmune-related neuromuscular disorder that causes muscle weakness, presents with varying severity, including myasthenic crisis (MC). Although MC can cause significant morbidity and mortality, specialized neuro-intensive care can produce a good long-term prognosis. Considering the outcomes of MG during hospitalization, it is critical to conduct risk assessments to predict the need for intensive care. Evidence and valid tools for the screening of critical patients with MG are lacking. We used three machine learning-based decision tree algorithms, including a classification and regression tree, C4.5, and C5.0, for predicting intensive care unit (ICU) admission of patients with MG. We included 228 MG patients admitted between 2015 and 2018. Among them, 88.2% were anti-acetylcholine receptors antibody positive and 4.7% were anti-muscle-specific kinase antibody positive. Twenty clinical variables were used as predictive variables. The C5.0 decision tree outperformed the other two decision tree and logistic regression models. The decision rules constructed by the best C5.0 model showed that the Myasthenia Gravis Foundation of America clinical classification at admission, thymoma history, azathioprine treatment history, disease duration, sex, and onset age were significant risk factors for the development of decision rules for ICU admission prediction. The developed machine learning-based decision tree can be a supportive tool for alerting clinicians regarding patients with MG who require intensive care, thereby improving the quality of care.

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.

Non-Lactobacillus-Dominated Vaginal Microbiota Is Associated With a Tubal Pregnancy in Symptomatic Chinese Women in the Early Stage of Pregnancy: A Nested Case-Control Study

The features of the vaginal microbiota (VM) community can reflect health status, and they could become new biomarkers for disease diagnosis. During pregnancy, domination of bacteria of the genus Lactobacillus in the VM community is regarded as a keystone because they stabilize the VM by producing antimicrobial compounds and competing adhesion. An altered VM composition provides a marker for adverse pregnancy outcomes. This nested case–control study aimed to characterize the VM in women with a tubal pregnancy (TP) presenting with pain and/or uterine bleeding in early pregnancy. Chinese women with a symptomatic early pregnancy of unknown location were the study cohort. 16S rDNA gene-sequencing of V3–V4 variable regions was done to assess the diversity, structures, taxonomic biomarkers, and classification of the VM community. The primary outcome was the location of the early pregnancy. The VM community in women with a TP showed higher diversity (PD-whole-tree, median: 8.26 vs. 7.08, P = 0.047; Shannon Diversity Index, median: 1.43 vs 0.99, P = 0.03) and showed different structures to those in women with an intrauterine pregnancy (IUP) (R = 0.23, P < 0.01). Bacteria of the genus Lactobacillus were significantly enriched in the IUP group, whereas bacteria of the genera Gardnerella and Prevotella were significantly enriched in the TP group. Lactobacillus abundance could be used to classify the pregnancy location (AUC = 0.81). Non-Lactobacillus-dominated microbiota (≤ 0.85% Lactobacillus) was significantly associated with a TP (adjusted odds ratio: 4.42, 95% confidence interval: 1.33 to 14.71, P = 0.02). In conclusion, among women with a symptomatic early pregnancy, a higher diversity and lower abundance of Lactobacillus in the VM is associated with a TP.

Fresh Perspectives on Hospital-Acquired Neonatal Skin Injury Period Prevalence From a Multicenter Study: Length of Stay, Acuity, and Incomplete Course of Antenatal Steroids

The objective of this study was to explore neonatal skin injury period prevalence, classification, and risk factors. Skin injury period prevalence over 9 months and χ2, Mann-Whitney U, and independent-samples t tests compared injured and noninjured neonates, with P values less than .05 considered statistically significant. Injury prediction models were developed using Classification and Regression Tree (CART) analysis for the entire cohort and separately for those classified as high or low acuity. The study took place in 3 Australian and New Zealand units. Neonates enrolled (N = 501) had a mean birth gestational age of 33.48 ± 4.61 weeks and weight of 2138.81 ± 998.92 g. Of the 501 enrolled neonates, 206 sustained skin injuries (41.1%), resulting in 391 injuries to the feet (16.4%; n = 64), cheek (12.5%; n = 49), and nose (11.3%; n = 44). Medical devices were directly associated with 61.4% (n = 240) of injuries; of these medical devices, 50.0% (n = 120) were unable to be repositioned and remained in a fixed position for treatment duration. The strongest predictor of skin injury was birth gestation of 30 weeks or less, followed by length of stay of more than 12 days, and birth weight of less than 1255 g. Prediction for injury based on illness acuity identified neonates less than 30 weeks' gestation and length of stay more than 39 days were at a greater risk (high acuity), as well as neonates less than 33 weeks' gestation and length of stay of more than 9 days (low acuity). More than 40% of hospitalized neonates acquired skin injury, of which the majority skin injuries were associated with medical devices required to sustain life. Increased neonatal clinician education and improved skin injury frameworks, informed by neonatal epidemiological data, are vital for the development of effective prevention strategies.

Random forest analysis identifies change in serum creatinine and listing status as the most predictive variables of an outcome for young children on liver transplant waitlist

Young children listed for liver transplant have high waitlist mortality (WL), which is not fully predicted by the PELD score. SRTR database was queried for children < 2 years listed for initial LT during 2002-17 (n = 4973). Subjects were divided into three outcome groups: bad (death or removal for too sick to transplant), good (spontaneous improvement), and transplant. Demographic, clinical, listing history, and laboratory variables at the time of listing (baseline variables), and changes in variables between listing and prior to outcome (trajectory variables) were analyzed using random forest (RF) analysis. 81.5% candidates underwent LT, and 12.3% had bad outcome. RF model including both baseline and trajectory variables improved prediction compared to model using baseline variables alone. RF analyses identified change in serum creatinine and listing status as the most predictive variables. 80% of subjects listed with a PELD score at time of listing and outcome underwent LT, while ~70% of subjects in both bad and good outcome groups were listed with either Status 1 (A or B) prior to an outcome, regardless of initial listing status. Increase in creatinine on LT waitlist was predictive of bad outcome. Longer time spent on WL was predictive of good outcome. Subjects with biliary atresia, liver tumors, and metabolic disease had LT rate >85%, while >20% of subjects with acute liver failure had a bad outcome. Change in creatinine, listing status, need for RRT, time spent on LT waitlist, and diagnoses were the most predictive variables.

Three machine learning algorithms and their utility in exploring risk factors associated with primary cesarean section in low-risk women: A methods paper

Machine learning, a branch of artificial intelligence, is increasingly used in health research, including nursing and maternal outcomes research. Machine learning algorithms are complex and involve statistics and terminology that are not common in health research. The purpose of this methods paper is to describe three machine learning algorithms in detail and provide an example of their use in maternal outcomes research. The three algorithms, classification and regression trees, least absolute shrinkage and selection operator, and random forest, may be used to understand risk groups, select variables for a model, and rank variables' contribution to an outcome, respectively. While machine learning has plenty to contribute to health research, it also has some drawbacks, and these are discussed as well. To provide an example of the different algorithms' function, they were used on a completed cross-sectional study examining the association of oxytocin total dose exposure with primary cesarean section. The results of the algorithms are compared to what was done or found using more traditional methods.

Derivation and validation of risk groups in patients with osteosarcoma utilizing regression tree analysis

BACKGROUND

For patients with osteosarcoma, apart from stage and primary site, we lack reliable prognostic factors for risk stratification at diagnosis. There is a need for further defined, discrete prognostic groups using presenting clinical features.

METHODS

We analyzed a cohort of 3069 patients less than 50 years of age, diagnosed with primary osteosarcoma of the bone between 1986 and 2013 from the Surveillance, Epidemiology, and End Results (SEER) database. Patients were randomly split into test and validation cohorts. Optimal cut points for age, tumor size, and grade were identified using classification and regression tree analysis. Manual recursive partitioning was used to identify discrete prognostic groups within the test cohort. These groups were applied to the validation cohort, and overall survival was analyzed using Cox models, Kaplan Meier methods, and log-rank tests.

RESULTS

After applying recursive partitioning to the test cohort, our initial model included six groups. Application of these groups to the validation cohort resulted in four final groups. Key risk factors included presence of metastases, tumor site, tumor grade, age, and tumor size. Patients with localized axial tumors were identified as having similar outcomes to patients with metastases. Age and tumor size were only prognostically important in patients with extremity tumors when assessed in the validation cohort.

CONCLUSIONS

This analysis supports prior reports that patients with axial tumors are a high-risk group, and demonstrates the importance of age and tumor size in patients with appendicular tumors. Biologic and genetic markers are needed to further define subgroups in osteosarcoma.

Preoperative Stratification of Liver Transplant Recipients: Validation of the LTRS

BACKGROUND

The liver transplant risk score (LTRS) was developed to stratify 90-day mortality of patients referred for liver transplantation (LT). We aimed to validate the LTRS using a new cohort of patients.

METHODS

The LTRS stratifies the risk of 90-day mortality of LT recipients based on their age, body mass index, diabetes, model for end-stage liver disease (MELD) score, and need for dialysis. We assessed the performance of the LTRS using a new cohort of patients transplanted in the United States between July 2013 and June 2017. Exclusion criteria were age <18 years, ABO incompatibility, redo or multivisceral transplants, partial grafts, malignancies other than hepatocellular carcinoma and fulminant hepatitis.

RESULTS

We found a linear correlation between the number of points of the LTRS and 90-day mortality. Among 18 635 recipients, 90-day mortality was 2.7%, 3.8%, 5.2%, 4.8%, 6.7%, and 9.3% for recipients with 0, 1, 2, 3, 4, and ≥5 points (P < 0.001). The LTRS also stratified 1-year mortality that was 5.5%, 7.7%, 9.9%, 9.3%, 10.8%, and 15.4% for 0, 1, 2, 3, 4, and ≥5 points (P < 0.001). An inverse correlation was found between the LTRS and 4-year survival that was 82%, 79%, 78%, 82%, 78%, and 66% for patients with 0, 1, 2, 3, 4, and ≥5 points (P < 0.001). The LTRS remained an independent predictor after accounting for recipient sex, ethnicity, cause of liver disease, donor age, cold ischemia time, and waiting time.

CONCLUSIONS

The LTRS can stratify the short- and long-term outcomes of LT recipients at the time of their evaluations irrespective of their gender, ethnicity, and primary cause of liver disease.

Feasibility study on exhaled-breath analysis by untargeted Selected-Ion Flow-Tube Mass Spectrometry in children with cystic fibrosis, asthma, and healthy controls: Comparison of data pretreatment and classification techniques

Selected-Ion Flow-Tube Mass Spectrometry (SIFT-MS) has been applied in a clinical context as diagnostic tool for breath samples using target biomarkers. Exhaled breath sampling is non-invasive and therefore much more patient friendly compared to bronchoscopy, which is the golden standard for evaluating airway inflammation. In the actual pilot study, 55 exhaled breath samples of children with asthma, cystic-fibrosis and healthy individuals were included. Rather than focusing on the analysis of target biomarkers or on the identification of biomarkers, different data analysis strategies, including a variety of pretreatment, classification and discrimination techniques, are evaluated regarding their capacity to distinguish the three classes based on subtle differences in their full scan SIFT-MS spectra. Proper data-analysis strategies are required because these full scan spectra contain much external, i.e. unwanted, variation. Each SIFT-MS analysis generates three spectra resulting from ion-molecule reactions of analyte molecules with H₃O⁺, NO⁺ and O₂⁺. Models were built with Linear Discriminant Analysis, Quadratic Discriminant Analysis, Soft Independent Modelling by Class Analogy, Partial Least Squares - Discriminant Analysis, K-Nearest Neighbours, and Classification and Regression Trees. Perfect models, concerning overall sensitivity and specificity (100% for both) were found using Direct Orthogonal Signal Correction (DOSC) pretreatment. Given the uncertainty related to the classification models associated with DOSC pretreatments (i.e. good classification found also for random classes), other models are built applying other preprocessing approaches. A Partial Least Squares - Discriminant Analysis model with a combined pre-processing method considering single value imputation results in 100% sensitivity and specificity for calibration, but was less good predictive. Pareto scaling prior to Quadratic Discriminant Analysis resulted in 41/55 correctly classified samples for calibration and 34/55 for cross-validation. In future, the uncertainty with DOSC and the applicability of the promising preprocessing methods and models must be further studied applying a larger representative data set with a more extensive number of samples for each class. Nevertheless, this pilot study showed already some potential for the untargeted SIFT-MS application as a rapid pattern-recognition technique, useful in the diagnosis of clinical breath samples.

Clinical and Biological Markers Predict the Efficacy of Glucocorticoid- and Macrolide-Based Postoperative Therapy in Patients With Chronic Rhinosinusitis

BACKGROUND

Although subsequent anti-inflammatory treatments are indispensable for patients with chronic rhinosinusitis (CRS) undergoing sinus surgery, few studies have explored the factors influencing the efficacy of postoperative anti-inflammatory treatment.

OBJECTIVE

We aimed to develop prediction models for the response to glucocorticoid- and macrolide-based postoperative therapy in CRS patients.

METHODS

We performed a post-hoc analysis of our previous study comparing the efficacy of fluticasone propionate and clarithromycin in the postoperative treatment of CRS patients. Clinical characteristics and treatment outcome information were collected. In addition, diseased sinonasal mucosal tissues obtained during surgery were processed for Bio-Plex analysis of protein levels of 34 biomarkers. Classification trees were built to predict refractory CRS based on clinical characteristics and biological markers for patients treated with fluticasone propionate or clarithromycin. A random forest algorithm was used to confirm the discriminating factors that formed the classification trees.

RESULTS

One year after surgery, 22.7% of the patients (17/75) treated with fluticasone propionate, and 24.3% of those (18/74) treated with clarithromycin were diagnosed with refractory CRS. Nasal tissue IL-8 and IgG3 levels and headache VAS scores in the fluticasone propionate group, and nasal tissue IgG4 levels and overall burden of symptoms VAS scores in the clarithromycin group, were identified as discriminating factors forming the classification tree to predict refractory CRS. The overall predictive accuracy of the model was 89.3% and 87.8% for fluticasone propionate- and clarithromycin-based postsurgical treatment, respectively.

CONCLUSIONS

Classification trees built using clinical and biological parameters could be helpful in identifying patients with poor response to fluticasone propionate- and clarithromycin-based postoperative treatment.

A practical algorithm to predict postsurgical recurrence and progression of pituitary neuroendocrine tumours (PitNET)s

OBJECTIVE

Pituitary neuroendocrine tumours (PitNET)s can be aggressive, thus presenting local invasion, postsurgical recurrence and/or resistance to treatment, responsible for significant morbidity. The study aimed at identifying prognostic factors of postsurgical outcome using data-driven classification of patients.

DESIGN

Retrospective observational study.

METHODS

Clinicopathological and radiological data of patients with PitNET treated via endoscopic endonasal surgery were collected. Tumour recurrence/progression and progression-free survival were assessed by classification tree analysis (CTA) and Kaplan-Meier curves, respectively. Histological subtype, cavernous/sphenoid sinus invasion, mitosis, Ki-67, p53, Trouillas' grading, degree of tumour exeresis and postsurgery disease activity were also evaluated.

RESULTS

A total of 1066 (466 gonadotroph, 287 somatotroph, 148 lactotroph, 157 corticotroph and 8 thyrotroph) tumours were included; 21.7% invaded the cavernous/sphenoid sinus. Based on Trouillas' classification, 64.3% were grade 1a, 14.2% 1b, 16.1% 2a, and 5.4% 2b; 18.3% had >2/10 HPF mitoses, 24.9% had Ki-67 ≥3%; 15.8% were positive for p53. Exeresis was radical in 81.2% of the cases. Median follow-up was 59.2 months. At last evaluation, 79.4% of the patients were cured; 20.6% had disease persistence, controlled by medical treatment in 18.3% of them. Disease recurrence/progression was recorded in 10.9% of the cases. CTA identified 5 distinct patient subgroups with different risk of disease recurrence/progression. Grade 2 of the Trouillas' grading, >2/10 HPF mitoses, Ki-67 ≥3%, p53 protein expression (P < .001), tumour invasion (P = .002) and ACTH-subtype (P = .003) were identified as risk factors of disease recurrence/progression.

CONCLUSIONS

The combined evaluation of Trouillas' grading, proliferation indexes and immunohistochemistry appears promising in the prediction of surgical outcome in PitNET.

Validation of a type 1 diabetes algorithm using electronic medical records and administrative healthcare data to study the population incidence and prevalence of type 1 diabetes in Ontario, Canada

INTRODUCTION

We aimed to develop algorithms distinguishing type 1 diabetes (T1D) from type 2 diabetes in adults ≥18 years old using primary care electronic medical record (EMRPC) and administrative healthcare data from Ontario, Canada, and to estimate T1D prevalence and incidence.

RESEARCH DESIGN AND METHODS

The reference population was a random sample of patients with diabetes in EMRPC whose charts were manually abstracted (n=5402). Algorithms were developed using classification trees, random forests, and rule-based methods, using electronic medical record (EMR) data, administrative data, or both. Algorithm performance was assessed in EMRPC. Administrative data algorithms were additionally evaluated using a diabetes clinic registry with endocrinologist-assigned diabetes type (n=29 371). Three algorithms were applied to the Ontario population to evaluate the minimum, moderate and maximum estimates of T1D prevalence and incidence rates between 2010 and 2017, and trends were analyzed using negative binomial regressions.

RESULTS

Of 5402 individuals with diabetes in EMRPC, 195 had T1D. Sensitivity, specificity, positive predictive value and negative predictive value for the best performing algorithms were 80.6% (75.9-87.2), 99.8% (99.7-100), 94.9% (92.3-98.7), and 99.3% (99.1-99.5) for EMR, 51.3% (44.0-58.5), 99.5% (99.3-99.7), 79.4% (71.2-86.1), and 98.2% (97.8-98.5) for administrative data, and 87.2% (81.7-91.5), 99.9% (99.7-100), 96.6% (92.7-98.7) and 99.5% (99.3-99.7) for combined EMR and administrative data. Administrative data algorithms had similar sensitivity and specificity in the diabetes clinic registry. Of 11 499 711 adults in Ontario in 2017, there were 24 789 (0.22%, minimum estimate) to 102 140 (0.89%, maximum estimate) with T1D. Between 2010 and 2017, the age-standardized and sex-standardized prevalence rates per 1000 person-years increased (minimum estimate 1.7 to 2.56, maximum estimate 7.48 to 9.86, p<0.0001). In contrast, incidence rates decreased (minimum estimate 0.1 to 0.04, maximum estimate 0.47 to 0.09, p<0.0001).

CONCLUSIONS

Primary care EMR and administrative data algorithms performed well in identifying T1D and demonstrated increasing T1D prevalence in Ontario. These algorithms may permit the development of large, population-based cohort studies of T1D.

Live poultry market closure and avian influenza A (H7N9) infection in cities of China, 2013-2017: an ecological study

Background

Previous studies have proven that the closure of live poultry markets (LPMs) was an effective intervention to reduce human risk of avian influenza A (H7N9) infection, but evidence is limited on the impact of scale and duration of LPMs closure on the transmission of H7N9.

Method

Five cities (i.e., Shanghai, Suzhou, Shenzhen, Guangzhou and Hangzhou) with the largest number of H7N9 cases in mainland China from 2013 to 2017 were selected in this study. Data on laboratory-confirmed H7N9 human cases in those five cities were obtained from the Chinese National Influenza Centre. The detailed information of LPMs closure (i.e., area and duration) was obtained from the Ministry of Agriculture. We used a generalized linear model with a Poisson link to estimate the effect of LPMs closure, reported as relative risk reduction (RRR). We used classification and regression trees (CARTs) model to select and quantify the dominant factor of H7N9 infection.

Results

All five cities implemented the LPMs closure, and the risk of H7N9 infection decreased significantly after LPMs closure with RRR ranging from 0.80 to 0.93. Respectively, a long-term LPMs closure for 10–13 weeks elicited a sustained and highly significant risk reduction of H7N9 infection (RRR = 0.98). Short-time LPMs closure with 2 weeks in every epidemic did not reduce the risk of H7N9 infection (p > 0.05). Partially closed LPMs in some suburbs contributed only 35% for reduction rate (RRR = 0.35). Shenzhen implemented partial closure for first 3 epidemics (p > 0.05) and all closure in the latest 2 epidemic waves (RRR = 0.64).

Conclusion

Our findings suggest that LPMs all closure in whole city can be a highly effective measure comparing with partial closure (i.e. only urban closure, suburb and rural remain open). Extend the duration of closure and consider permanently closing the LPMs will help improve the control effect. The effect of LPMs closure seems greater than that of meteorology on H7N9 transmission.

Spleen volumetry and liver transient elastography: Predictors of persistent posthepatectomy decompensation in patients with hepatocellular carcinoma

BACKGROUND

Posthepatectomy decompensation remains a frequent and poor outcome after hepatectomy, but its prediction is still inaccurate. Liver stiffness measurement can predict posthepatectomy decompensation, but there is a so-called "gray zone" that requires another predictor. Because splenomegaly is an objective sign of portal hypertension, we hypothesized that spleen volumetry could improve the identification of patients at risk.

METHODS

Patients with hepatocellular carcinoma who underwent hepatectomy in our tertiary center between August 2014 and December 2017 were reviewed. The primary endpoint was to determine if the spleen volumetry and liver stiffness measurement were independent predictors of posthepatectomy decompensation, and secondarily, to determine if they were synergistic through a theoretic predictive model.

RESULTS

One hundred and seven patients were included. The median follow-up time was 3 months (3-5). Postoperative 90-day mortality was 4.7%. By multivariate analysis, liver stiffness measurement and spleen volumetry predicted posthepatectomy decompensation. The liver stiffness measurement had a cutoff point of 11.6 kPa (area under receiver operating curve = 0.71 confidence interval 95% 0.71-0.88, sensitivity: 89%, specificity: 47%). The spleen volumetry cutoff point was 381.1 cm³ (area under receiver operating curve = 0.78, 95% confidence interval 0.77-0.93, sensitivity: 55%, specificity: 91%). The spleen volumetry improved prediction of posthepatectomy decompensation, because use of the spleen volumetry increased sensitivity (from 62% to 97%) and the negative predictive value (from 96% to 100%) along with a negligible decrease in specificity (from 96.7 to 93.4) and positive predictive value (from 64% to 59%) (P = .003).

CONCLUSION

Spleen volumetry (>380 cm³) and liver stiffness measurement (>12 kPa) are non-invasive, independent, and synergistic tools that appear to be able to predict posthepatectomy decompensation. The importance of this finding is that these measurements may help to anticipate posthepatectomy decompensation and may possibly be used to direct alternative treatments to resection.

Medical Costs of Patients with Type 2 Diabetes in a Single Payer System: A Classification and Regression Tree Analysis

BACKGROUND AND OBJECTIVES

Many studies and systematic reviews have estimated the healthcare costs of diabetes using a cost-of-illness approach. However, in the studies based on this approach patients' heterogeneity is rarely taken into account. The aim of this study was to stratify patients with type 2 diabetes into homogeneous cost groups based on demographic and clinical characteristics.

METHODS

We conducted a retrospective cost-of-illness study by linking individual data on health services utilization retrieved from the administrative databases of Emilia-Romagna Region (Italy). Direct medical costs (either all-cause or diabetes-related) were calculated from the perspective of the regional health service, using tariffs for hospitalizations and outpatient services and the unit costs of prescriptions for drugs. The determinants of costs identified in a generalized linear regression model were used to characterize subgroups of patients with homogeneous costs in a classification and regression tree analysis.

RESULTS

The study population consisted of a cohort of 101,334 patients with type 2 diabetes, followed up for 1 year, with a mean age of 70.9 years. Age, gender, complications, comorbidities and living area accounted significantly for cost variability. The classification tree identified ten patient subgroups with different costs, ranging from a median of €483 to €39,578. The two subgroups with highest costs comprised dialysis patients, and the largest subgroup (57.9%) comprised patients aged ≥ 65 years without renal, cardiovascular and cerebrovascular complications.

CONCLUSIONS

Classification of patients into homogeneous cost subgroups can be used to improve the management of, and budget allocation for, patients with type 2 diabetes.

Decision Tree Analysis: A Retrospective Analysis of Postoperative Recurrence of Adhesions in Patients with Moderate-to-Severe Intrauterine

OBJECTIVE

To establish and validate a decision tree model to predict the recurrence of intrauterine adhesions (IUAs) in patients after separation of moderate-to-severe IUAs.

DESIGN

A retrospective study.

SETTING

A tertiary hysteroscopic center at a teaching hospital.

POPULATION

Patients were retrospectively selected who had undergone hysteroscopic adhesion separation surgery for treatment of moderate-to-severe IUAs.

INTERVENTIONS

Hysteroscopic adhesion separation surgery and second-look hysteroscopy 3 months later.

MEASUREMENTS AND MAIN RESULTS

Patients' demographics, clinical indicators, and hysteroscopy data were collected from the electronic database of the hospital. The patients were randomly apportioned to either a training or testing set (332 and 142 patients, respectively). A decision tree model of adhesion recurrence was established with a classification and regression tree algorithm and validated with reference to a multivariate logistic regression model. The decision tree model was constructed based on the training set. The classification node variables were the risk factors for recurrence of IUAs: American Fertility Society score (root node variable), isolation barrier, endometrial thickness, tubal opening, uterine volume, and menstrual volume. The accuracies of the decision tree model and multivariate logistic regression analysis model were 75.35% and 76.06%, respectively, and areas under the receiver operating characteristic curve were 0.763 (95% CI 0.681-0.846) and 0.785 (95% CI 0.702-0.868).

CONCLUSIONS

The decision tree model can readily predict the recurrence of IUAs and provides a new theoretical basis upon which clinicians can make appropriate clinical decisions.

Clinical indices and mortality of hospitalized avian influenza A (H7N9) patients in Guangdong, China

Background:

Six epidemic waves of human infection with avian influenza A (H7N9) virus have emerged in China with high mortality. However, study on quantitative relationship between clinical indices in ill persons and H7N9 outcome (fatal and non-fatal) is still unclear. A retrospective cohort study was conducted to collect laboratory-confirmed cases with H7N9 viral infection from 2013 to 2015 in 23 hospitals across 13 cities in Guangdong Province, China.

Methods:

Multivariable logistic regression model and classification tree model analyses were used to detect the threshold of selected clinical indices and risk factors for H7N9 death. The receiver operating characteristic curve (ROC) and analyses were used to compare survival and death distributions and differences between indices. A total of 143 cases with 90 survivors and 53 deaths were investigated.

Results:

Average age (Odds Ratio (OR) = 1.036, 95% Confidence Interval (CI) = 1.016–1.057), interval days between dates of onset and confirmation (OR = 1.078, 95% CI = 1.004–1.157), interval days between onset and oseltamivir treatment (OR = 5.923, 95% CI = 1.877–18.687), body temperature (BT) (OR = 3.612, 95% CI = 1.914–6.815), white blood cell count (WBC) (OR = 1.212, 95% CI = 1.092–1.346) were significantly associated with H7N9 death after adjusting for confounders. The chance of death from H7N9 infection was 80.0% if BT was over 38.1 °C, and chance of death is 67.4% if WBC count was higher than 9.5 (10⁹/L). Only 27.1% of patients who began oseltamivir treatment less than 9.5 days after disease onset died, compared to 68.8% of those who started treatment more than 15.5 days after onset.

Conclusions:

The intervals between date of onset and confirmation of diagnosis, between date of onset to oseltamivir treatment, age, BT and WBC are found to be the best predictors of H7N9 mortality.

The development of a short form of the DePaul Symptom Questionnaire

PURPOSE/OBJECTIVE

The DePaul Symptom Questionnaire (DSQ) is a widely used instrument that assesses common symptoms of myalgic encephalomyelitis (ME) and chronic fatigue syndrome (CFS). The DSQ has strong psychometric properties; however, it consists of 99 items, and the energy limitations and cognitive difficulties experienced by individuals with ME and CFS may hinder their ability to easily complete the questionnaire.

METHOD

The current study examined symptom prevalence and discriminative ability to develop a short form of the DSQ (DSQ-SF).

RESULTS

The resulting short form questionnaire consists of 14 items that were highly prevalent among individuals with ME and CFS. Additionally, the items demonstrated the ability to differentiate individuals with ME and CFS from adult controls and, to a lesser extent, individuals with multiple sclerosis.

CONCLUSIONS/IMPLICATIONS

The DSQ-SF may serve as an effective, brief screening tool for symptoms of ME and CFS. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Risk stratification integrating genetic data for factor VIII inhibitor development in patients with severe hemophilia A

Replacement therapy in severe hemophilia A leads to factor VIII (FVIII) inhibitors in 30% of patients. Factor VIII gene (F8) mutation type, a family history of inhibitors, ethnicity and intensity of treatment are established risk factors, and were included in two published prediction tools based on regression models. Recently investigated immune regulatory genes could also play a part in immunogenicity. Our objective is to identify bio-clinical and genetic markers for FVIII inhibitor development, taking into account potential genetic high order interactions. The study population consisted of 593 and 79 patients with hemophilia A from centers in Bonn and Frankfurt respectively. Data was collected in the European ABIRISK tranSMART database. A subset of 125 severely affected patients from Bonn with reliable information on first treatment was selected as eligible for risk stratification using a hybrid tree-based regression model (GPLTR). In the eligible subset, 58 (46%) patients developed FVIII inhibitors. Among them, 49 (84%) were “high risk” F8 mutation type. 19 (33%) had a family history of inhibitors. The GPLTR model, taking into account F8 mutation risk, family history of inhibitors and product type, distinguishes two groups of patients: a high-risk group for immunogenicity, including patients with positive HLA-DRB1*15 and genotype G/A and A/A for IL-10 rs1800896, and a low-risk group of patients with negative HLA-DRB1*15 / HLA-DQB1*02 and T/T or G/T for CD86 rs2681401. We show associations between genetic factors and the occurrence of FVIII inhibitor development in severe hemophilia A patients taking into account for high-order interactions using a generalized partially linear tree-based approach.

Machine Learning Accurately Predicts Short-Term Outcomes Following Open Reduction and Internal Fixation of Ankle Fractures

Ankle fractures are common orthopedic injuries with favorable outcomes when managed with open reduction and internal fixation (ORIF). Several patient-related risk factors may contribute to poor short-term outcomes, and machine learning may be a valuable tool for predicting outcomes. The objective of this study was to evaluate machine-learning algorithms for accurately predicting short-term outcomes after ORIF for ankle fractures. The Nationwide Inpatient Sample and Nationwide Readmissions Database were queried for adult patients ≥18 years old who underwent ORIF of an ankle fracture during 2013 or 2014. Morbidity and mortality, length of stay >3 days, and 30-day all-cause readmission were the outcomes of interest. Two machine-learning models were created to identify patient and hospital characteristics associated with the 3 outcomes. The machine learning models were evaluated using confusion matrices and receiver operating characteristic area under the curve values. A total of 16,501 cases were drawn from the Nationwide Inpatient Sample and used to assess morbidity and mortality and length of stay >3 days, and 33,504 cases were drawn from the Nationwide Readmissions Database to assess 30-day readmission. Older age, Medicaid, Medicare, deficiency anemia, congestive heart failure, chronic lung disease, diabetes, hypertension, and renal failure were the variables associated with a statistically significant increased risk of developing all 3 adverse events. Logistic regression and gradient boosting had similar area under the curve values for each outcome, but gradient boosting was more accurate and more specific for predicting each outcome. Our results suggest that several comorbidities may be associated with adverse short-term outcomes after ORIF of ankle fractures, and that machine learning can accurately predict these outcomes.

A Stroke Risk Detection: Improving Hybrid Feature Selection Method

Background

Stroke is one of the most common diseases that cause mortality. Detecting the risk of stroke for individuals is critical yet challenging because of a large number of risk factors for stroke.

Objective

This study aimed to address the limitation of ineffective feature selection in existing research on stroke risk detection. We have proposed a new feature selection method called weighting- and ranking-based hybrid feature selection (WRHFS) to select important risk factors for detecting ischemic stroke.

Methods

WRHFS integrates the strengths of various filter algorithms by following the principle of a wrapper approach. We employed a variety of filter-based feature selection models as the candidate set, including standard deviation, Pearson correlation coefficient, Fisher score, information gain, Relief algorithm, and chi-square test and used sensitivity, specificity, accuracy, and Youden index as performance metrics to evaluate the proposed method.

Results

This study chose 792 samples from the electronic records of 13,421 patients in a community hospital. Each sample included 28 features (24 blood test features and 4 demographic features). The results of evaluation showed that the proposed method selected 9 important features out of the original 28 features and significantly outperformed baseline methods. Their cumulative contribution was 0.51. The WRHFS method achieved a sensitivity of 82.7% (329/398), specificity of 80.4% (317/394), classification accuracy of 81.5% (645/792), and Youden index of 0.63 using only the top 9 features. We have also presented a chart for visualizing the risk of having ischemic strokes.

Conclusions

This study has proposed, developed, and evaluated a new feature selection method for identifying the most important features for building effective and parsimonious models for stroke risk detection. The findings of this research provide several novel research contributions and practical implications.

Mobility Analysis of AmpuTees (MAAT 4): classification tree analysis for probability of lower limb prosthesis user functional potential

Purpose: To develop a predictive model to inform the probability of lower limb prosthesis users' functional potential for ambulation.Materials and Methods: A retrospective analysis of a database of outcomes for 2770 lower limb prosthesis users was used to inform a classification and regression tree analysis. Gender, age, height, weight, body mass index adjusted for amputation, amputation level, cause of amputation, comorbid health status and functional mobility score [Prosthetic Limb Users Survey of Mobility (PLUS-M™)] were entered as potential predictive variables. Patient K-Level was used to assign dependent variable status as unlimited community ambulator (i.e., K3 or K4) or limited community/household ambulator (i.e., K1 or K2). The classification tree was initially trained from 20% of the sample and subsequently tested with the remaining sample.Results: A classification tree was successfully developed, able to accurately classify 87.4% of individuals within the model's training group (standard error 1.4%), and 81.6% within the model's testing group (standard error 0.82%). Age, PLUS-M™ T-score, cause of amputation and body weight were retained within the tree logic.Conclusions: The resultant classification tree has the ability to provide members of the clinical care team with predictive probabilities of a patient's functional potential to help assist care decisions.Implications for RehabilitationClassification and regression tree analysis is a simple analytical tool that can be used to provide simple predictive models for patients with a lower limb prosthesis.The resultant classification tree had an 81.6% (standard error 0.82%) accuracy predicting functional potential as an unlimited community ambulator (i.e., K3 or K4) or limited community/ household ambulator (i.e., K1 or K2) in an unknown group of 2770 lower limb prosthesis users.The resultant classification tree can assist with the rehabilitation team's care planning providing probabilities of functional potential for the lower limb prosthesis user.

Population level changes in schistosome-specific antibody levels following chemotherapy

AIMS

Previous studies have reported that chemotherapy of schistosomiasis by praziquantel in humans boosts protective antibody responses against S mansoni and S haematobium. A number of studies have reported schistosome-specific antibody levels before and after chemotherapy. Using these reports, a meta-analysis was conducted to identify predictors of population level change in schistosome-specific antibody levels after chemotherapy.

METHODS AND RESULTS

Following a systematic review, 92 observations from 26 articles published between 1988 and 2013 were included in this study. Observations were grouped by antigen type and antibody isotypes for the classification and regression tree (CART) analysis. The study showed that the change in antibody levels was variable: (a) between different human populations and (b) according to the parasite antigen and antibody isotypes. Thus, while anti-worm responses predominantly increased after chemotherapy, anti-egg responses decreased or did not show a significant trend. The change in antibody levels depended on a combination of age and infection intensity for anti-egg IgA, IgM, IgG1, IgG2 and anti-worm IgM and IgG.

CONCLUSION

The study results are consistent with praziquantel treatment boosting anti-worm antibody responses. However, there is considerable heterogeneity in post-treatment changes in specific antibody levels that is related to host age and pre-treatment infection intensity.

NMR metabolomics of cerebrospinal fluid differentiates inflammatory diseases of the central nervous system

BACKGROUND

Myriad infectious and noninfectious causes of encephalomyelitis (EM) have similar clinical manifestations, presenting serious challenges to diagnosis and treatment. Metabolomics of cerebrospinal fluid (CSF) was explored as a method of differentiating among neurological diseases causing EM using a single CSF sample.

METHODOLOGY/PRINCIPAL FINDINGS

1H NMR metabolomics was applied to CSF samples from 27 patients with a laboratory-confirmed disease, including Lyme disease or West Nile Virus meningoencephalitis, multiple sclerosis, rabies, or Histoplasma meningitis, and 25 controls. Cluster analyses distinguished samples by infection status and moderately by pathogen, with shared and differentiating metabolite patterns observed among diseases. CART analysis predicted infection status with 100% sensitivity and 93% specificity.

CONCLUSIONS/SIGNIFICANCE

These preliminary results suggest the potential utility of CSF metabolomics as a rapid screening test to enhance diagnostic accuracies and improve patient outcomes.

Patient scheduling based on a service-time prediction model: a data-driven study for a radiotherapy center

With the growth of the population, access to medical care is in high demand, and queues are becoming longer. The situation is more critical when it concerns serious diseases such as cancer. The primary problem is inefficient management of patients rather than a lack of resources. In this work, we collaborate with the Centre Intégré de Cancérologie de Laval (CICL). We present a data-driven study based on a nonblock approach to patient appointment scheduling. We use data mining and regression methods to develop a prediction model for radiotherapy treatment duration. The best model is constructed by a classification and regression tree; its accuracy is 84%. Based on the predicted duration, we design new workday divisions, which are evaluated with various patient sequencing rules. The results show that with our approach, 40 additional patients are treated daily in the cancer center, and a considerable improvement is noticed in patient waiting times and technologist overtime.

Evaluation of potential predictor variables for PCR assay diagnosis of Anaplasma phagocytophilum infection in equids in Northern California

OBJECTIVE To identify clinical or clinicopathologic variables that can be used to predict a positive PCR assay result for Anaplasma phagocytophilum infection in equids. ANIMALS 162 equids. PROCEDURES Medical records were reviewed to identify equids that underwent testing for evidence of A phagocytophilum infection by PCR assay between June 1, 2007, and December 31, 2015. For each equid that tested positive (case equid), 2 time-matched equids that tested negative for the organism (control equids) were identified. Data collected included age, sex, breed, geographic location (residence at the time of testing), physical examination findings, and CBC and plasma biochemical analysis results. Potential predictor variables were analyzed by stepwise logistic regression followed by classification and regression tree analysis. Generalized additive models were used to evaluate identified predictors of a positive test result for A phagocytophilum. RESULTS Total lymphocyte count, plasma total bilirubin concentration, plasma sodium concentration, and geographic latitude were linear predictors of a positive PCR assay result for A phagocytophilum. Plasma creatine kinase activity was a nonlinear predictor of a positive result. CONCLUSIONS AND CLINICAL RELEVANCE Assessment of predictors identified in this study may help veterinarians identify equids that could benefit from early treatment for anaplasmosis while definitive test results are pending. This information may also help to prevent unnecessary administration of oxytetracycline to equids that are unlikely to test positive for the disease.

Risk factors can classify individuals who develop accelerated knee osteoarthritis: Data from the osteoarthritis initiative

We assessed which combinations of risk factors can classify adults who develop accelerated knee osteoarthritis (KOA) or not and which factors are most important. We conducted a case-control study using data from baseline and the first four annual visits of the Osteoarthritis Initiative. Participants had no radiographic KOA at baseline (Kellgren-Lawrence [KL]<2). We classified three groups (matched on sex): (i) accelerated KOA: >1 knee developed advance-stage KOA (KL = 3 or 4) within 48 months; (ii) typical KOA: >1 knee increased in radiographic scoring (excluding those with accelerated KOA); and (iii) No KOA: no change in KL grade by 48 months. We selected eight predictors: Serum concentrations for C-reactive protein, glycated serum protein (GSP), and glucose; age; sex; body mass index; coronal tibial slope, and femorotibial alignment. We performed a classification and regression tree (CART) analysis to determine rules for classifying individuals as accelerated KOA or not (no KOA and typical KOA). The most important baseline variables for classifying individuals with incident accelerated KOA (in order of importance) were age, glucose concentrations, BMI, and static alignment. Individuals <63.5 years were likely not to develop accelerated KOA, except when overweight. Individuals >63.5 years were more likely to develop accelerated KOA except when their glucose levels were >81.98 mg/dl and they did not have varus malalignment. The unexplained variance of the CART = 69%. These analyses highlight the complex interactions among four risk factors that may classify individuals who will develop accelerated KOA but more research is needed to uncover novel risk factors. © 2017 Orthopaedic Research Society. Published by Wiley Periodicals, Inc. J Orthop Res 36:876-880, 2018.

Advances in the genome-wide association study of chronic hepatitis B susceptibility in Asian population

Chronic hepatitis B (CHB) is the most common chronic liver disease resulting from viral infection and has become a serious threat to human health. Each year, about 1.2 million people in the world die from diseases caused by chronic infection of hepatitis B virus. The genetic polymorphism is significantly associated with the susceptibility to chronic hepatitis B. Genome-wide association study was recently developed and has become an important tool to detect susceptibility genes of CHB. To date, a number of CHB-associated susceptibility loci and regions have been identified by scientists over the world. To clearly understand the role of susceptibility loci in the occurrence of CHB is important for the early diagnosis and prevention of CHB.

Conditional Inference Tree for Multiple Gene-Environment Interactions on Myocardial Infarction

BACKGROUND AND AIMS

Identifying gene-environment interaction in the context of multiple environmental factors has been a challenging task. We aimed to use conditional inference tree (CTREE) to strata myocardial infarction (MI) risk synthesizing information from both genetic and environmental factors.

METHODS

We conducted a case-control study including 1440 Chinese men (730 MI patients and 710 controls). We first calculated a weighted genetic risk score (GRS) by combining 25 single nucleotide polymorphisms (SNPs) that had been identified to be associated with coronary artery diseases in previous genome wide association studies. We then developed a CTREE model to interpret the gene-environment interaction network in predicting MI.

RESULTS

We detected high-order interactions between dyslipidemia, GRS, smoking status, age and diabetes. Of all the variables examined, high density lipoprotein cholesterol (HDL-C) of 1.25 mmlo/L was identified as the key discriminator. The subsequent splits of MI were low density lipoprotein cholesterol (LDL-C) of 4.01 mmol/L and GRS of 20.9. We found that individuals with HDL-C ≤1.25 mmol/L, GRS >20.9 and lipoprotein (a) > 0.09 g/L had a higher risk of MI than those who at the lowest risk group (OR: 5.89, 95% CI: 3.99-8.69). This magnitude of MI risk was similar to the combination of HDL-C ≤1.25 mmol/L, GRS ≤20.9, smoking and lipoprotein (a) > 0.15 g/L (OR: 5.49, 95% CI: 3.51-8.58).

CONCLUSIONS

The multiple interactions between genetic and environmental factors can be visually present via the CTREE approach. The tree diagram also simplifies the decision making procedure by answering a sequence of questions along the branches.