Recursive Partitioning Research Articles

Radiomics of metastatic brain tumor as a predictive image biomarker of progression-free survival in patients with non-small-cell lung cancer with brain metastasis receiving tyrosine kinase inhibitors

Background and objectiveEpidermal growth factor receptor (EGFR)–targeted tyrosine kinase inhibitors (TKIs) are the first-line therapy for EGFR-mutant non-small-cell lung cancer (NSCLC). Early prediction of treatment failure in patients with brain metastases treated with EGFR–TKIs may help in making decisions for systemic drug therapy or local brain tumor control. This study examined the predictive power of the radiomics of both brain metastasis tumors and primary lung tumors. We propose a deep learning based CoxCC model based on quantitative brain magnetic resonance imaging (MRI), a prognostic index and clinical data; the model can be used to predict progression-free survival (PFS) after EGFR–TKI therapy in advanced EGFR-mutant NSCLC. MethodsThis retrospective single-center study included 271 patients receiving first-line EGFR–TKI targeted therapy in 2018–2019. Among them, 72 patients who had brain metastases before receiving first-line EGFR–TKI treatment. Three radiomic features were extracted from pretreatment brain MRI images. A CoxCC model for the progression risk stratification of EGFR–TKI treatment was proposed on the basis of MRI radiomics, clinical features, and a prognostic index. We performed time-dependent PFS predictions to evaluate the performance of the CoxCC model. ResultsThe CoxCC model based on a prognostic index, clinical features, and radiomic features of brain metastasis exhibited higher performance than clinical features combined with indexes previously proposed for determining the prognosis of brain metastasis, including recursive partitioning analysis, diagnostic-specific graded prognostic assessment, graded prognostic assessment for lung cancer using molecular markers (lung-molGPA), and modified lung-molGPA, with c-index values of 0.75, 0.67, 0.66, 0.65, and 0.65, respectively. The model achieved areas under the curve of 0.88, 0.73, 0.92, and 0.90 for predicting PFS at 3, 6, 9 and 12 months, respectively. PFS significantly differed between the high- and low-risk groups (p < 0.001). ConclusionsFor patients with advanced-stage NSCLC with brain metastasis, MRI radiomics of brain metastases may predict PFS. The CoxCC model integrating brain metastasis radiomics, clinical features, and a prognostic index provided reliable multi-time-point PFS predictions for patients with advanced NSCLC and brain metastases receiving EGFR–TKI treatment.

Development and Validation of a Predictive Model for Acute Kidney Injury in Sepsis Patients Based on Recursive Partition Analysis.

Sepsis-associated acute kidney injury (SA-AKI) is a critical condition with significant clinical implications, yet there is a need for a predictive model that can reliably assess the risk of its development. This study is undertaken to bridge a gap in healthcare by creating a predictive model for SA-AKI with the goal of empowering healthcare providers with a tool that can revolutionize patient care and ultimately lead to improved outcomes. A cohort of 615 patients afflicted with sepsis, who were admitted to the intensive care unit, underwent random stratification into 2 groups: a training set (n = 435) and a validation set (n = 180). Subsequently, a multivariate logistic regression model, imbued with nonzero coefficients via LASSO regression, was meticulously devised for the prognostication of SA-AKI. This model was thoughtfully rendered in the form of a nomogram. The salience of individual risk factors was assessed and ranked employing Shapley Additive Interpretation (SHAP). Recursive partition analysis was performed to stratify the risk of patients with sepsis. Among the panoply of clinical variables examined, hypertension, diabetes mellitus, C-reactive protein, procalcitonin (PCT), activated partial thromboplastin time, and platelet count emerged as robust and independent determinants of SA-AKI. The receiver operating characteristic curve analysis for SA-AKI risk discrimination in both the training set and validation set yielded an area under the curve estimates of 0.843 (95% CI: 0.805 to 0.882) and 0.834 (95% CI: 0.775 to 0.893), respectively. Notably, PCT exhibited the most conspicuous influence on the model's predictive capacity. Furthermore, statistically significant disparities were observed in the incidence of SA-AKI and the 28-day survival rate across high-risk, medium-risk, and low-risk cohorts (P < .05). The composite predictive model, amalgamating the quintet of SA-AKI predictors, holds significant promise in facilitating the identification of high-risk patient subsets.

Missing Data Imputation with High-Dimensional Data

Imputation of missing data in high-dimensional datasets with more variables P than samples N, P ≫ N , is hampered by the data dimensionality. For multivariate imputation, the covariance matrix is ill conditioned and cannot be properly estimated. For fully conditional imputation, the regression models for imputation cannot include all the variables. Thus, the high dimension requires special imputation approaches. In this article, we provide an overview and realistic comparisons of imputation approaches for high-dimensional data when applied to a linear mixed modeling (LMM) framework. We examine approaches from three different classes using simulation studies: multiple imputation with penalized regression, multiple imputation with recursive partitioning and predictive mean matching; and multiple imputation with Principal Component Analysis (PCA). We illustrate the methods on a real case study where a multivariate outcome (i.e., an extracted set of correlated biomarkers from human urine samples) was collected and monitored over time and we discuss the proposed methods with more standard imputation techniques that could be applied by ignoring either the multivariate or the longitudinal dimension. Our simulations demonstrate the superiority of the recursive partitioning and predictive mean matching algorithm over the other methods in terms of bias, mean squared error and coverage of the LMM parameter estimates when compared to those obtained from a data analysis without missingness, although it comes at the expense of high computational costs. It is worthwhile reconsidering much faster methodologies like the one relying on PCA.

EPID-12. SURVIVAL IS ASSOCIATED WITH CIRCULATING B-CELLS AND AGE AMONG OLIGODENDROGLIOMA PATIENTS

Abstract BACKGROUND Though patients with oligodendroglioma have improved outcomes versus astrocytomas, there remains marked variability in their survival experience that is unexplained by known prognostic factors. Immunomethylomics uses DNA methylation as a substitute for flow cytometry to characterize immune cells in whole blood. METHODS We applied immunomethylomics to archived blood samples from patients enrolled in the UCSF Adult Glioma Study including oligodendrogliomas (141); other adult diffuse glioma subtypes (347); and controls (454). RESULTS Median age of oligodendrogliomas was 44 years (range 19 – 83), 57% male, 75% white, and 60% received chemotherapy prior to blood draw. Oligodendroglioma patients had lower percent of B cell fractions (total, memory, naïve) than people without glioma. They also had higher percent of B cell (total, memory) and lymphocyte fractions (total, CD4-total and memory), basophil, eosinophil, and NK cells versus other glioma subtypes. Median overall survival in oligodendrogliomas was 17.5 years (95% CI 16.9 - NA) with median follow-up of 12 years (95% CI 11 – 14). Recursive partitioning analysis delineated two survival groups for oligodendroglioma patients based on age and percent B naïve cells. Best outcome, with median survival of 24.2 years (95% CI 20.4 – NR), was seen in patients &amp;lt; /= 42 years of age with a higher percent of B naïve cells ( &amp;gt;1.91%) (N = 37). Worst outcome, with median survival of 16.9 years (14.8 – NR), was seen in patients &amp;lt; /= 42 years of age with a lower percent of B naïve cells (&amp;lt; /= 1.91%) (N = 28) OR patients &amp;gt; 42 years of age (N = 76) (p = 0.00032). CONCLUSIONS An interaction between age and B naïve cell percentages may predict survival for oligodendroglioma patients. Immunomethylomics is a powerful tool for assessing immune status and these findings suggest that blood-immune profiles may be prognostic glioma patients.

Termination-of-resuscitation rule in the emergency department for patients with initial pulseless electrical activity after out-of-hospital cardiac arrest

Abstract Background/Introduction In several Asian countries including Japan, emergency medical service (EMS) providers are prohibited from performing field termination-of-resuscitation (TOR) in patients with refractory out-of-hospital cardiac arrest (OHCA). In 2022, we developed a TOR rule in the emergency department for patients with initial asystole after refractory OHCA, which includes three criteria: unwitnessed arrest, EMS-initiated cardiopulmonary resuscitation (CPR) duration was more than 20 min, and there was no prehospital return of spontaneous circulation (ROSC). However, until now a TOR rule in the emergency department for patients with initial pulseless electrical activity (PEA) has not been developed. Purpose This study aimed to develop and validate a TOR rule in the emergency department for patients with initial PEA after refractory OHCA. Methods We analysed the records of 137,230 adult patients (≥18 years-old) with initial PEA from the nationwide all-Japan Utstein registry from 2016 to 2020. Patients were divided into two groups: development (n=109,642, 2016 to 2019) and validation (n=27,588, 2020). The primary outcome measures were specificity, false positive rate (FPR), and positive predictive value (PPV) of the TOR rule for patients with initial PEA in the emergency department for predicting 1-month mortality. Using recursive partitioning analysis, we developed a new TOR rule for patients with initial PEA after OHCA. Results The overall 1-month survival rate of the participants was 7.4% (95% confidence interval [CI], 7.2–7.5%). Recursive partitioning analysis for the development group in predicting 1-month mortality revealed that a TOR rule for patients with PEA could be defined if patients with OHCA met all the following four criteria immediately after hospital arrival: (1) unwitnessed arrest by any laypersons, (2) EMS-initiated CPR duration more than 20 min, (3) no prehospital ROSC, and (4) age greater than or equal to 80 years. The specificity, FPR, and PPV of the rule for predicting 1-month mortality were 99.1% (95% CI, 98.9–99.3%), 0.87% (95% CI, 0.69–1.09%), and 99.2% (95% CI, 98.9–99.3%), respectively. The proportion of patients who fulfilled the rule and the area under the receiver operating curve (AUC) was 7.8% (95% CI, 7.7–8.0%), and 0.80 (95% CI, 0.79–0.80), respectively. In the validation group, the specificity, FPR, PPV, proportion of patients who met the rule, and AUC were 99.1% (95% CI, 98.5–99.4%), 0.93% (95% CI, 0.58–1.49%), 99.3% (95% CI, 98.8–99.5%), 8.3% (95% CI, 8.0–8.6%), and 0.789 (95% CI, 0.775–0.797), respectively. Conclusions We developed and validated a TOR rule in the emergency department for patients with initial PEA after OHCA. The rule includes four criteria (unwitnessed arrest, EMS-initiated CPR duration was more than 20 min, no prehospital ROSC, and age was greater than or equal to 80 years) with a more than 99% predictor of 1-month mortality.

Abstract 17881: A Predictive Algorithm to Identify Late Gadolinium Enhancement Using Echocardiography

Background: Late gadolinium enhancement (LGE) on cardiac MRI (CMR) correlates with fibrosis and arrhythmic risk, but cannot be used in several clinical settings compared the availability of echocardiography (TTE). Aim: To develop and evaluate a machine learning (ML) model that uses standard clinical and echocardiography (TTE) variables to identify LGE on paired CMR. Methods: Amongst 725 patients (mean age 56.9 ± 16.5 years, 40.3% female) with TTE and CMR within 3 months at our institution, 2018-2023, we derived and tested several ML models (recursive partitioning, random forest, boosted tree, neural networks, LASSO/Elastic Net) to identify LGE on CMR using age and 24 TTE features. Ten-fold cross validation was used for internal validation. Results: A total of 165 (23%) had the presence of LGE (40% ischemic, 60% nonischemic pattern). Of tested models, a boosted neural network performed best (validation area under the curve [AUC] = 0.98; Figure ). In the validation cohort, the algorithm had a sensitivity and specificity of 93.8% and 94.6% for identifying presence of LGE. Accuracy remained high regardless of gender (M vs. F, AUC, 0.987 vs. 0.988) or race (blacks vs. whites, AUC, 0.987 vs. 0.984) and was not related to time between studies (r = -0.06, p = 0.09). Accuracy was better for ischemic LGE (AUC = 0.94) than nonischemic LGE (AUC = 0.90). Left ventricular size, age, fractional shortening, and wall thickness, were amongst the top predictors of LGE presence in the model. Conclusions: In this single center study, a boosted neural network using TTE features identified LGE on paired CMR with high sensitivity and specificity. If confirmed in external data, this suggests possible utility of ML for improving the diagnostic value of TTE for LGE assessment.

Abstract 13374: The Intersection of Prenatal Clinical and Demographic Factors in Predicting Fetal Brain Volume and Neurodevelopmental Outcome in Single Ventricle CHD

Background: Fetuses with single ventricle (SV) CHD have small total brain volumes (TBV) which has been linked to neurodevelopmental impairment (NDI). Evidence suggests a prenatal origin of NDI for CHD. Our primary aim was to use machine learning methods to identify prenatal variables that predict fetal brain growth and in turn early NDI. Methods: This is a multi-center prospective cohort study of maternal-infant dyads with a fetal diagnosis of SV who underwent fetal brain MRI and in-depth prenatal clinical and sociodemographic data collection (self reported race/ethnicity, income, education, area deprivation index, maternal health, genetics, placental health). Infants had ND testing at 12-30 months with Bayley Scales of Infant and Toddler Development. Machine learning models used recursive partitioning to predict 1) gestational-age (GA) and sex-adjusted fetal TBV and 2) ND outcome while including fetal TBV in the models. Results: A total of 67 maternal-infant dyads were enrolled with a mean GA of 33.4 weeks (SD=3.4) at fetal MRI and male predominance (57%); 48 infants had follow-up at median age of 23.5 months (SD=6.3). Key predictors of fetal brain volume were race/ethnicity, maternal age, and cardiac anatomy (Figure 1A). For NDI, fetal brain volume was the strongest predictor of motor outcomes but was not associated with other domains (Figure 1D). Cardiac anatomy was the strongest predictor of cognitive outcome (Figure 1B), while race/ethnicity was the strongest predictor of language (Figure 1C). Cardiac anatomy, maternal age, and cerebral and placental hemodynamics (e.g., middle cerebral or umbilical artery pulsatility) each predicted multiple domains of neurodevelopment. Conclusions: In regression tree modeling, sociodemographic variables are the strongest predictors of fetal TBV, which in turn predicts motor outcome. Prenatal factors predicting NDI across all domains span sociodemographic, cardiac, and placental characteristics.

Comprehensive learning TLBO with recursive precedence-based solution construction and multilevel local search for the linear ordering problem

The linear ordering problem (LOP) is a difficult permutation-based optimization task with a multitude of practical applications in different areas. The objective of the LOP is to maximize the total of the benefits associated with the pairwise precedence relations between the permuted items. Since it is NP-hard, several approximate methods have been developed to quickly produce acceptable solutions. We aim to design a specialized teaching–learning-based optimization (TLBO) method that exploits the permutation structure of the LOP. TLBO is a modern population-based metaheuristic that mimics the conventional process of transmitting knowledge and skills in a classroom. Here we extend and improve TLBO by employing a comprehensive learning strategy. In this strategy, a selected group of learners from the current class are employed as exemplars to generate new solutions. The key component of our comprehensive learning TLBO (CL-TLBO) is a recursive permutation construction procedure. It is an effective pivot-based partitioning algorithm that focuses on the pairwise precedences between permutation items in exemplar solutions to generate new candidate solutions. For better performance, we investigate different ways of selecting the pivot item at each partitioning step. Furthermore, we offer a multilevel local search technique to refine the solutions of sub-problems that are constructed during the recursive partitioning process. The experimental evaluation demonstrates a very competitive performance of the developed CL-TLBO method when compared with various well-established techniques.

An optimized postsurgery follow-up strategy for patients with esophageal cancer: a cohort study.

After radical surgery, patients with esophageal cancer should undergo long-term surveillance of disease relapse. However, the optimal follow-up strategy remains to be explored. A total of 4688 patients were recruited. Recursive partition analysis was applied to develop recurrence risk stratification for patients. The follow-up strategies of each stratification were developed based on monthly recurrence probability and validated by bootstrap validation and an external dataset. A Markov decision-analytic model was constructed to evaluate the cost-effectiveness of the follow-up strategies. Patients were stratified into four groups according to four pathological features. The authors applied a random survival forest to calculate the monthly recurrence probability of each group. Based on the temporal distribution of recurrences, the authors further established surveillance strategies for four groups. The strategies were validated as optimal protocols by bootstrap resampling and another dataset. Markov cost-effective analysis indicated that our recommended strategies outperformed the mainstream protocols from guidelines. Using less than 12 visits across the first 5 years on average, our follow-up strategies were more efficient than the NCCN recommended strategies (14 visits average). Our results also supported the computerized tomography from the neck to the upper abdomen as a routine examination and PETCT of distant metastasis for some groups with high risks. Our study provided data-driven evidence of personalized and economic follow-up strategies for esophageal cancer patients and shed light on follow-up optimization for other cancer types.

A Day 14 Endpoint for Acute Gvhd Clinical Trials

The overall response rate (ORR) 28 days (D28) after treatment has been adopted as the primary endpoint for clinical trials of acute graft versus host disease (GVHD). ORR combines complete response (CR) and partial response (PR) because CR and PR have very similar non relapse mortality (NRM). D28 ORR has some disadvantages: first, physicians usually decide to modify immunosuppression much earlier than D28, and second, NRM does not always correlate well with ORR at D28, particularly for patients who present with intermediate risk or low risk disease. MAGIC serum biomarkers at day 14 (D14) of GVHD treatment have been shown to correlate better with NRM than clinical symptom response (Srinagesh et al, Blood Advances, 2019). We therefore hypothesized that a combination of clinical symptom severity and biomarkers at D14 of treatment would predict NRM at least as well and possibly better than clinical response at D28. We studied 1144 patients from 23 MAGIC sites who were systemically treated with at least 0.25 mg/kg/day of systemic steroids (prednisone equivalents) and had clinical data and serum samples available in the MAGIC database and biorepository. We divided the patients into a training set (n=764) and a validation set (n=380). We used 12-month NRM from the time of acute GVHD onset as the primary outcome of interest. We first trained a recursive partitioning algorithm in the training set to create a response model (MAGIC D14 clinical response) according to overall GVHD grade at both onset and D14 that predicted 12-month NRM and then applied it to the validation set. Patients that either progressed to grade III/IV GVHD, continued having grade III/IV GVHD, or only improved to grade II at D14 from grade III/IV at D0 had high NRM and were classified as non-responders. Patients with grade I/II GVHD at day 0 who had grade 0-II GVHD at day 14 and patients who improved to grade 0/I at day 14 from grade III/IV at day 0 had low NRM and were classified as responders. Although counterintuitive, responders included patients whose GVHD remained at grade II because the 12 month NRM of this group was only 20%. The new clinical model predicted NRM as well as the D28 standard response model in the validation set. When we integrated the D14 MAP biomarker score (low: ≤ 0.290 vs high: &amp;gt;0.290) with the new D14 clinical response (D14 MAGIC integrated response), we observed three response groups of integrated response with strikingly different 12 month NRM (8%, 35%, 76%, p&amp;lt;0.001), in contrast to the D28 standard response (Figure 1A, 1B). D14 complete response required MAGIC clinical response with low MAP; D14 partial response required either MAGIC clinical response with high MAP or MAGIC clinical non-response with low MAP; and D14 non-response required MAGIC clinical non-response with high MAP. The D14 response groups also displayed significant differences in 12 month OS (82%, 58%, 14% respectively, p&amp;lt;0.001) We then compared the D14 integrated response with the D28 standard response for 12 month NRM prediction. This D14 integrated response model was superior for time-dependent AUC (0.78 vs 0.71, P=0.03), sensitivity (0.71 vs 0.54), positive predictive value (0.52 vs 0.48) and negative predictive value (0.92 vs 0.89), with minimal loss in specificity (0.85 vs 0.87) in the prediction of 12-month NRM compared to the D28 standard response model. Using decision curve analysis, the D14 integrated response displayed higher net benefit (correct identification of patients that will experience NRM by 12 months) over the whole range of threshold probabilities/preferences for changing immunosuppression when 12-month NRM was used as the outcome. We conclude first, that the definition of response that uses onset and D14 grades is as accurate as D28 ORR model; and second, that a model using clinical grades and biomarkers on D14 more accurately predicts long term NRM than the standard D28 definition and may be useful as a clinical trial endpoint.

External Validation of Models for Predicting Disability in Community-Dwelling Older People in the Netherlands: A Comparative Study.

Advanced statistical modeling techniques may help predict health outcomes. However, it is not the case that these modeling techniques always outperform traditional techniques such as regression techniques. In this study, external validation was carried out for five modeling strategies for the prediction of the disability of community-dwelling older people in the Netherlands. We analyzed data from five studies consisting of community-dwelling older people in the Netherlands. For the prediction of the total disability score as measured with the Groningen Activity Restriction Scale (GARS), we used fourteen predictors as measured with the Tilburg Frailty Indicator (TFI). Both the TFI and the GARS are self-report questionnaires. For the modeling, five statistical modeling techniques were evaluated: general linear model (GLM), support vector machine (SVM), neural net (NN), recursive partitioning (RP), and random forest (RF). Each model was developed on one of the five data sets and then applied to each of the four remaining data sets. We assessed the performance of the models with calibration characteristics, the correlation coefficient, and the root of the mean squared error. The models GLM, SVM, RP, and RF showed satisfactory performance characteristics when validated on the validation data sets. All models showed poor performance characteristics for the deviating data set both for development and validation due to the deviating baseline characteristics compared to those of the other data sets. The performance of four models (GLM, SVM, RP, RF) on the development data sets was satisfactory. This was also the case for the validation data sets, except when these models were developed on the deviating data set. The NN models showed a much worse performance on the validation data sets than on the development data sets.

Relationship between transcriptional expression of pyruvate dehydrogenase and local control of disease in patients with oral cavity carcinomas

BackgroundThe altered cellular metabolism is one of the hallmarks of the cancer cells, favoring the process of aerobic glycolysis, known as the Warburg effect. The pyruvate dehydrogenase (PDH) complex is one of the elements involved in this metabolic process. The present study aims to evaluate the relationship between the transcriptional expression of PDHB and the risk of local recurrence in patients with oral cavity carcinomas. MethodsWe determined the transcriptional expression of PDHB in biopsies from 41 patients with oral cavity carcinomas treated with surgery. The PDHB expression was categorized according to the local control of the disease with a recursive partitioning analysis. ResultsDuring the follow-up period 13 patients (31.7%) had a local recurrence of the tumor. Considering local disease control as the dependent variable, the recursive partitioning analysis classified the patients in two categories according to high (n=16, 39.0%) or low (n=25, 61.0%) PDHB expression. Five-year local recurrence-free survival for patients with high PDHB expression was 84.8% (95% CI: 65.2-100%), and for patients with low expression it was 54.3% (95% CI: 34.3–74.2 %) (P=0.034). The results of multivariate analysis showed that patients with a low PDHB expression had a 4.90 times higher risk of local recurrence of the tumor (95% CI: 1.02–22.68, P=0.042). ConclusionThere is a relationship between the metabolic characteristics of the tumor and its aggressiveness. According to our results, patients with oral cavity carcinomas with low transcriptional expression levels of PDHB have a significantly higher risk of local tumor recurrence.

Predictive capacity of FAT1 transcriptional expression in patients with head and neck squamous cell carcinomas treated with radiotherapy

ObjectiveTo analyze the predictive capacity at the primary location of the tumor of the FAT1 transcriptional expression in patients with head and neck squamous cell carcinoma treated with radiotherapy. Material and methodsWe conducted a retrospective study from biopsies of the primary location of the tumor in 82 patients with head and neck squamous cell carcinoma treated with radiotherapy. The transcriptional expression of FAT1 was determined by RT-PCR. The level of FAT1 transcriptional expression was categorized according to the local control after radiotherapy using a recursive partitioning analysis. ResultsElevated FAT1 transcriptional expression was associated with an increased risk of local recurrence after radiotherapy. Patients with a high expression level of FAT1 (n=18; 22.0%) had a 5-year local recurrence-free survival of 42.1% (95% CI: 18.6%–65.6%), whereas for patients with a low expression (n=64; 78.0%) it was 72.4% (95% CI: 61.5%–83.3%) (p=0.002). According to the result of a multivariate analysis, patients with a high FAT1 expression category had a 2.3-fold increased risk of local recurrence (95% CI: 1.0–5.2; p=0.043). ConclusionsElevated FAT1 transcriptional expression was associated with a significantly increased risk of local recurrence in patients with head and neck squamous cell carcinoma treated with radiotherapy.

Stratification of risk for emergent intracranial abnormalities in children with headaches: a Pediatric Emergency Care Applied Research Network (PECARN) study protocol

IntroductionHeadache is a common chief complaint of children presenting to emergency departments (EDs). Approximately 0.5%–1% will have emergent intracranial abnormalities (EIAs) such as brain tumours or strokes. However, more than...

Better Existing Water, Sanitation, and Hygiene Can Reduce the Risk of Cholera in an Endemic Setting: Results From a Prospective Cohort Study From Kolkata, India.

Global cholera control efforts rely heavily on effective water, sanitation, and hygiene (WASH) interventions in cholera-endemic settings. Using data from a large, randomized controlled trial of oral cholera vaccine conducted in Kolkata, India, we evaluated whether natural variations in WASH in an urban slum setting were predictive of cholera risk. From the control population (n = 55 086), baseline WASH data from a randomly selected "training subpopulation" (n = 27 634) were analyzed with recursive partitioning to develop a dichotomous ("better" vs "not better") composite household WASH variable from several WASH features collected at baseline, and this composite variable was then evaluated in a mutually exclusive "validation population" (n = 27 452). We then evaluated whether residents of better WASH households in the entire population (n = 55 086) experienced lower cholera risk using Cox regression models. Better WASH was defined by a combination of 4 dichotomized WASH characteristics including safe source of water for daily use, safe source of drinking water, private or shared flush toilet use, and always handwashing with soap after defecation. Residence in better WASH households was associated with a 30% reduction in risk of cholera over a 5-year period (adjusted hazard ratio, 0.70 [95% confidence interval, .49-.99]; P = .048). We also found that the impact of better WASH households on reducing cholera risk was greatest in young children (0-4 years) and this effect progressively declined with age. The evidence suggests that modest improvements in WASH facilities and behaviors significantly modify cholera risk and may be an important component of cholera prevention and elimination strategies in endemic settings. Clinical Trials Registration. NCT00289224.

Dynamic time series clustering with multivariate linkage and automatic dendrogram cutting using a recursive partitioning algorithm

Dynamic time series clustering with multivariate linkage and automatic dendrogram cutting using a recursive partitioning algorithm

Job Satisfaction and the ‘Great Resignation’: An Exploratory Machine Learning Analysis

Labor market dynamics is shaped by various social, psychological and economic drivers. Studies have suggested that job quit and labor market turnover are associated with job satisfaction. This study examines the determinants of job satisfaction using a large survey dataset, namely the LISS Work and Schooling module on an extensive sample of persons from the Netherlands. To handle these big data, machine learning models based on binary recursive partitioning algorithms are employed. Particularly, sequential and randomized tree-based techniques are used for prediction and clustering purposes. In order to interpret the results, the study calculates the sizes and directions of the effects of model features using computations based on the concept of Shapley value in cooperative game theory. The findings suggest that satisfaction with the social atmosphere among colleagues, wage satisfaction, and feeling of being appreciated are major determinants of job satisfaction.

Serum Creatinine and Amylase in Drain to Predict Pancreatic Fistula Risk after Pancreatoduodenectomy

Introduction: The identification of patients with low risk of clinically relevant postoperative pancreatic fistula (CR-POPF) and postoperative hemorrhage (PPH) can guide drain removal after pancreatoduodenectomy (PD). However, drain fluid amylase (DFA) ≤5,000 U/L on postoperative day (POD) 1 does not robustly predict the absence of CR-POPF. Methods: Consecutive patients undergoing PD at Sun Yat-sen University Cancer Center between July 2018 and October 2021 were analyzed. Recursive partitioning analysis was used to classify patients into groups with different risks of CR-POPF and PPH. Results: Among 288 consecutive patients included, 99 patients (34.38%) developed CR-POPF (86 grade B and 13 grade C). Patients with CR-POPF had increased levels of preoperative creatinine (CRE) and POD1 CRE. The combination of POD1 CRE (>104 μmol/L or not) and POD1 DFA (>5,000 U/L or not) stratified patients into subgroups with the maximum difference in CR-POPF risk. The CR-POPF rates were 17.82% (36/202) in group A (POD1 CRE ≤104 μmol/L and POD1 DFA ≤5,000 U/L), 53.33% (8/15) in group B (POD1 CRE >104 μmol/L and POD1 DFA ≤5,000 U/L), and 77.46% (55/71) in group C (POD1 DFA >5,000 U/L). The PPH rates were 1.98% (4/202), 20.00% (3/15), and 19.72% (14/71) in groups A, B, and C, respectively. Conclusion: Patients with POD1 DFA ≤5,000 U/L and POD1 CRE >104 μmol/L have a high risk of CR-POPF and may not benefit from early drain removal. Patients with POD1 DFA ≤5,000 U/L and POD1 CRE ≤104 μmol/L have low risk of CR-POPF and PPH.

Detecting heterogeneity in the causal direction of dependence: A model-based recursive partitioning approach.

Methods of causal discovery and direction of dependence to evaluate causal properties of variable relations have experienced rapid development. The majority of causal discovery methods, however, relies on the assumption of causal effect homogeneity, that is, the identified causal structure is expected to hold for the entire population. Because causal mechanisms can vary across subpopulations, we propose combining methods of model-based recursive partitioning and non-Gaussian causal discovery to identify such subpopulations. The resulting algorithm can discover subpopulations with potentially varying magnitude and causal direction of effects under mild parameter inequality assumptions. Feasibility conditions are described and results from synthetic data experiments are presented suggesting that large effects and large sample sizes are beneficial for detecting causally competing subgroups with acceptable statistical performance. In a real-world data example, the extraction of meaningful subgroups that differ in the causal mechanism underlying the development of numerical cognition is illustrated. Potential extensions and recommendations for best practice applications are discussed.

Pretreatment Pan-Immune-Inflammation Value (PIV) in Predicting Therapeutic Response and Clinical Outcomes of Neoadjuvant Immunochemotherapy for Esophageal Squamous Cell Carcinoma.

The pan-immune-inflammation value (PIV), which reflects the balance between the host immune and inflammatory status, is a readily available index for evaluating cancer outcomes. Until now, however, no study has demonstrated the clinical response of PIV to neoadjuvant immunochemotherapy (NICT) in esophageal squamous cell carcinoma (ESCC). This retrospective study included 218 patients with ESCC who underwent NICT. The relationship between PIV and therapeutic response (pathological complete response [PCR]) and clinical outcomes (overall survival [OS] and disease-free survival [DFS]) was examined. Cox proportional, hazard-regression analyses and the Kaplan-Meier method were used for survival analyses. Recursive partitioning analysis (RPA) was used to establish a novel risk stratification model. Sixty-six patients (30.3%) achieved PCR after NICT. Using PCR as the endpoint of interest, patients were compared in groups based on the optimal threshold. PIV was closely related to PCR (odds ratio [OR] 0.311, 95% confidence interval [CI] 0.140-0.690, P=0.004). Compared with patients in the low PIV cohort, patients with high PIV had worse 3-year OS (58.7% vs. 83.6%, P<0.001) and DFS (51.9% vs. 79.1%, P<0.001). PIV was an independent predictor of OS (hazard ratio [HR] 2.364, 95% CI 1.183-4.724, P=0.015) and DFS (HR 1.729, 95% CI 1.026-2.913, P=0.040). Three risk groups with varied DFS and OS were staged by using an RPA method, and the prognostication accuracy was considerably improved. Pretreatment PIV can predict the therapeutic efficacy of NICT for ESCC. Because of better prognostic stratification, pretreatment PIV is a novel, sensitive, and effective indicator in ESCC receiving NICT. The prognostic results of PIV need to be verified in additional prospective studies.