Theory and Practice of Integrating Machine Learning and Conventional Statistics in Medical Data Analysis

Comparison of spatial prediction models from Machine Learning of cholangiocarcinoma incidence in Thailand

BACKGROUND

Cholangiocarcinoma (CCA) poses a significant public health challenge in Thailand, with notably high incidence rates. This study aimed to compare the performance of spatial prediction models using Machine Learning techniques to analyze the occurrence of CCA across Thailand.

METHODS

This retrospective cohort study analyzed CCA cases from four population-based cancer registries in Thailand, diagnosed between January 1, 2012, and December 31, 2021. The study employed Machine Learning models (Linear Regression, Random Forest, Neural Network, and Extreme Gradient Boosting (XGBoost)) to predict Age-Standardized Rates (ASR) of CCA based on spatial variables. Model performance was evaluated using Root Mean Square Error (RMSE) and R2 with 70:30 train-test validation.

RESULTS

The study included 6,379 CCA cases, with a male predominance (4,075 cases; 63.9%) and a mean age of 66.2 years (standard deviation = 11.1 years). The northeastern region accounted for most of the cases (3,898 cases; 61.1%). The overall ASR of CCA was 8.9 per 100,000 person-years (95% CI: 8.7 to 9.2), with the northeastern region showing the highest incidence (ASR = 13.4 per 100,000 person-years; 95% CI: 12.9 to 13.8). In the overall dataset, the Random Forest model demonstrated better prediction performance in both the training (R2 = 72.07%) and testing datasets (R2 = 71.66%). Regional variations in model performance were observed, with Random Forest performing best in the northern, northeastern regions, while XGBoost excelled in the central and southern regions. The most important spatial predictors for CCA were elevation and distance from water sources.

CONCLUSION

The Random Forest model demonstrated the highest efficiency in predicting CCA incidence rates in Thailand, though predictive performance varied across regions. Spatial factors effectively predicted ASR of CCA, providing valuable insights for national-level disease surveillance and targeted public health interventions. These findings support the development of region-specific approaches for CCA control using spatial epidemiology and machine learning techniques.

An analysis of factors influencing dropout in methadone maintenance treatment program in Dehong Prefecture of China based on Cox regression and decision tree modelling

BACKGROUND

The high dropout rate among Methadone Maintenance Treatment (MMT) patients poses a significant challenge to drug dependence treatment programs, especially in regions with prevalent drug use and HIV transmission risks. This study aimed to analyze factors of dropout in MMT clinics over an 18-year period in Dehong Prefecture, Yunnan Province, China.

METHODS

A retrospective cohort study was conducted using data from China's HIV/AIDS Comprehensive Response Information Management System (CRIMS). Participants included individuals who enrolled in MMT between June 2005 and December 2023 and completed baseline surveys. Cox proportional hazards regression identified independent predictors, while decision tree modeling (CART algorithm) captured variable interactions. The decision tree employed Gini impurity minimization, a 70:30 training-test split, and pruning to prioritize factors like treatment duration and urine test results.

RESULTS

The study included 9,435 MMT participants, with a male-to-female ratio of 26:1 (9,086 males and 349 females). The median duration of treatment was 12.2 months (ranging from 2.7 to 43.9 months), with a minimum of 1 day and a maximum of 217 months. From 2005 to 2023, the cumulative dropout rate among MMT patients in Dehong Prefecture reached 89.6% (8,458/9,435), with an incidence rate of 34.75 dropouts per 100 person-years over 24,354.98 person-years of follow-up. The Cox proportional hazards regression identified that participants with occupations as farmers (AHR = 1.52, 95% CI: 1.41-1.62) or positive urine test results (AHR = 2.47, 95% CI: 2.35-2.59) exhibited significantly higher dropout risks. Protective factors included enrollment age > 35 years (AHR = 0.86), being married (AHR = 0.81), higher education levels (AHR = 0.94), good family relationships (AHR = 0.30), and methadone doses > 60 ml/day (AHR = 0.60). The decision tree model prioritized treatment duration as the root node, followed by urine test results, family relationships, education level, and methadone dosage. Patients with ≤ 12 months of treatment and positive urine tests faced the highest dropout probability (98.9%), while those with > 12 months of treatment but poor family relationships and doses ≤ 60 ml showed intermediate risks (82.3%).

CONCLUSION

Between 2005 and 2023, the dropout rate among MMT patients in Dehong Prefecture was relatively high, driven by modifiable factors (low methadone doses, positive urine tests) and contextual hierarchies (early-phase treatment duration). By integrating Cox regression and decision trees, we advance both epidemiological risk assessment and precision intervention design. Policymakers should prioritize dose optimization and targeted monitoring for high-risk subgroups (e.g., patients ≤ 12 months with concurrent drug use) to improve retention in resource-limited settings.

The relationship between medical students' interest in learning and their ability to solve mathematical problems: the chain-mediating role of teacher-student relationship and self-efficacy

Introduction

Although the impact of learning interest on academic performance has been extensively studied, the chain-mediated mechanism by which medical students'mathematics learning interest influences competence through teacher-student relationships and self-efficacy remains underexplored. Empirical evidence utilizing multi-mediation models to test indirect effects is particularly lacking.

Methods

This study investigated 806 Chinese medical students, assessing problem-solving ability using PISA mathematics items and examining the chain-mediated pathway of teacher-student relationships and mathematics self-efficacy via structural equation modeling (SEM) and bias-corrected bootstrap methods. After controlling for major, grade, and residence.

Results

The results demonstrated: (1) The direct effect of mathematics learning interest on problem-solving ability was non-significant (effect size = 0.0101, 95% CI [-0.0144, 0.0346]); (2) The independent mediating effect of teacher-student relationships was non-significant (effect size = 0.0083, 95% CI [-0.0114, 0.0196]); (3) The independent mediating effect of mathematics self-efficacy was significant (effect size = 0.0140, 95% CI [0.0003, 0.0286], contribution rate = 40.79%); (4) The chain-mediated pathway of teacher-student relationships → self-efficacy reached significance (effect size = 0.0020, 95% CI [0.0003, 0.0048], contribution rate = 5.68%). The total mediation effect accounted for 70.66% of the total effect.

Discussion

These findings confirm that self-efficacy serves as the critical mechanism translating medical students' mathematics interest into competence. We recommend enhancing self-efficacy through clinical scenario-based simulation tasks and stepwise training programs, providing theoretical foundations for reforming medical mathematics curricula.

Pilot study: Initial investigation suggests differences in EMT-associated gene expression in breast tumor regions

Triple negative breast cancer (TNBC) is the most aggressive subtype and disproportionately affects African American women. The development of breast cancer is highly associated with interactions between tumor cells and the extracellular matrix (ECM), and recent research suggests that cellular components of the ECM vary between racial groups. This pilot study aimed to evaluate gene expression in TNBC samples from patients who identified as African American and Caucasian using traditional statistical methods and emerging Machine Learning (ML) approaches. ML enables the analysis of complex datasets and the extraction of useful information from small datasets. We selected four regions of interest from tumor biopsy samples and used laser microdissection to extract tissue for gene expression characterization via RT-qPCR. Both parametric and non-parametric statistical analyses identified genes differentially expressed between the two ethnic groups. Out of 40 genes analyzed, 4 were differentially expressed in the edge of tumor (ET) region and 8 in the ECM adjacent to the tumor (ECMT) region. In addition to statistical approach, ML was used to generate decision trees (DT) for a broader analysis of gene expression and ethnicity. Our DT models achieved 83.33 % accuracy and identified the most significant genes, including CD29 and EGF from the ET region and SNAI1 and CHD2 from the ECMT region. All significant genes were analyzed for pathway enrichment using MSigDB and Gene Ontology databases, most notably the epithelial to mesenchymal transition and cell motility pathways. This pilot study highlights key genes of interest that are differentially expressed in African American and Caucasian TNBC samples.

Development and validation of an explainable machine learning prediction model of hemorrhagic transformation after intravenous thrombolysis in stroke

Objective

To develop and validate an explainable machine learning (ML) model predicting the risk of hemorrhagic transformation (HT) after intravenous thrombolysis.

Methods

We retrospectively enrolled patients who received intravenous tissue plasminogen activator (IV-tPA) thrombolysis within 4.5 h after symptom onset to form the original modeling cohort. HT was defined as any hemorrhage on head CT scan completed within 48 h after IV-tPA administration. We utilized the Random Forest (RF), Multilayer Perceptron (MLP), Adaptive Boosting (AdaBoost), and Gaussian Naive Bayes (GauNB) algorithms to develop ML-HT models. The models' predictive performance was evaluated using confusion matrix (including accuracy, precision, recall, and F1 score), and discriminative analysis (area under the receiver-operating-characteristic curve, ROC-AUC) in the original cohort, followed by validation in an independent external cohort. The models' explainability was assessed using SHapley Additive exPlanations (SHAP) global feature plot, SHAP Summary Plot, and Partial Dependence Plot.

Results

A total of 1,007 patients were included in the original modeling cohort, with an HT incidence of 8.94%. The RF-based ML-HT model showed metrics of 0.874 (accuracy), 0.972 (precision), 0.890 (recall), 0.929 (F1 score); with ROC-AUC of 0.7847 in the original cohort and 0.7119 in the external validation cohort. The MLP model showed 0.878, 0.967, 0.989, 0.978, 0.7710, and 0.6768, respectively. The AdaBoost model showed 0.907, 0.967, 0.989, 0.978, 0.7798, and 0.6606, respectively. The GauNB model showed 0.848, 0.983, 0.598, 0.716, 0.6953, and 0.6289, respectively. The explainable analysis of the RF-based ML model indicated that the National Institute of Health Stroke Scale (NIHSS) score, age, platelet count, and atrial fibrillation were the primary determinants for HT following IV-tPA thrombolysis.

Conclusion

The RF-based explainable ML model demonstrated promising predictive ability for estimating the risk of HT after IV-tPA thrombolysis and may have the potential to assist the clinical decision-making in emergency settings.

Advanced statistical methods for hazard modeling in cardiothoracic surgery: a comprehensive review of techniques and approaches

Hazard modeling in cardiothoracic surgery, crucial for understanding patient outcomes, utilizes survival analysis like the Cox proportional hazards model. Kaplan-Meier curves are employed in survival analysis to represent the probability of survival over time. While Cox assumes proportional hazards, the Fine-Gray model deals with competing risks. Parametric models (e.g., Weibull) specify survival distributions, unlike Cox. Bayesian analysis integrates prior knowledge with data. Machine learning, including decision trees and support vector machines, enhances risk prediction by analyzing extensive datasets. However, it is important to note that whatever new approaches one may adopt will enhance the quality of risk assessment and not the risk assessment as such. Preprocessing is vital for data quality in complex cardiovascular datasets, alongside robust validation methods like cross-validation for model reliability across patient cohorts.

Using machine learning methods to predict all-cause somatic hospitalizations in adults: A systematic review

AIM

In this review, we investigated how Machine Learning (ML) was utilized to predict all-cause somatic hospital admissions and readmissions in adults.

METHODS

We searched eight databases (PubMed, Embase, Web of Science, CINAHL, ProQuest, OpenGrey, WorldCat, and MedNar) from their inception date to October 2023, and included records that predicted all-cause somatic hospital admissions and readmissions of adults using ML methodology. We used the CHARMS checklist for data extraction, PROBAST for bias and applicability assessment, and TRIPOD for reporting quality.

RESULTS

We screened 7,543 studies of which 163 full-text records were read and 116 met the review inclusion criteria. Among these, 45 predicted admission, 70 predicted readmission, and one study predicted both. There was a substantial variety in the types of datasets, algorithms, features, data preprocessing steps, evaluation, and validation methods. The most used types of features were demographics, diagnoses, vital signs, and laboratory tests. Area Under the ROC curve (AUC) was the most used evaluation metric. Models trained using boosting tree-based algorithms often performed better compared to others. ML algorithms commonly outperformed traditional regression techniques. Sixteen studies used Natural language processing (NLP) of clinical notes for prediction, all studies yielded good results. The overall adherence to reporting quality was poor in the review studies. Only five percent of models were implemented in clinical practice. The most frequently inadequately addressed methodological aspects were: providing model interpretations on the individual patient level, full code availability, performing external validation, calibrating models, and handling class imbalance.

CONCLUSION

This review has identified considerable concerns regarding methodological issues and reporting quality in studies investigating ML to predict hospitalizations. To ensure the acceptability of these models in clinical settings, it is crucial to improve the quality of future studies.

Effect size and inferential statistical techniques coupled with machine learning for assessing the association between prolactin concentration and metabolic homeostasis

BACKGROUND AND OBJECTIVE

Recent guidelines classify low prolactin levels as low as <7 ng/mL and high levels as >25 ng/mL, while the "Homeostatically Functionally Increased Transient Prolactinemia" (HomeoFIT-PRL) range (25-100 ng/mL) suggests that a temporary increase in prolactin could be metabolically beneficial if no related health issues are present. The aim of this study was to investigate the association between mean prolactin concentrations and disturbances in glycidic and lipidic metabolism and to identify the gray zone associated with prolactin inflection points that correlate with these metabolic changes.

METHODS

This cross-sectional study involved 65,795 adults who underwent HOMA-IR, glucose, insulin, total cholesterol, HDL-c, LDL-c, and triglyceride tests. Data was categorized into 106 partitions based on prolactin results. Employing an approach referred to in this study as "Hierarchical Multicriteria Analysis of Differences Between Groups - Statistical and Effect Size Approach" (HiMADiG-SESA) comparing the mean concentrations of metabolic tests across prolactin ranges. A machine learning model was utilized to determine inflection points and their corresponding confidence intervals (CIs). These CIs helped establish gray zones in mean prolactin results related to metabolic changes.

RESULTS

Statistically and clinically, metabolic test means differed for prolactin <7 ng/mL, except insulin. In the HomeoFIT-PRL range, means were lower except for HDL-c. The gray zones of the mean prolactin results associated with changes in glycidic and lipidic metabolism were 9.58-12.87 ng/mL and 13.81-18.73 ng/mL, respectively.

CONCLUSION

A strong correlation was identified between mean prolactin concentrations and the results of metabolism tests below the gray zones associated with inflection points, indicating the potential role of prolactin in the appearance of metabolic disorders. Mean prolactin results can provide deeper insight into metabolic balance.

Improving the Trustworthiness of Interactive Visualization Tools for Healthcare Data through a Medical Fuzzy Expert System

Successful healthcare companies and illness diagnostics require data visualization. Healthcare and medical data analysis are needed to use compound information. Professionals often gather, evaluate, and monitor medical data to gauge risk, performance capability, tiredness, and adaptation to a medical diagnosis. Medical diagnosis data come from EMRs, software systems, hospital administration systems, laboratories, IoT devices, and billing and coding software. Interactive diagnosis data visualization tools enable healthcare professionals to identify trends and interpret data analytics results. Selecting the most trustworthy interactive visualization tool or application is crucial for the reliability of medical diagnosis data. Thus, this study examined the trustworthiness of interactive visualization tools for healthcare data analytics and medical diagnosis. The present study uses a scientific approach for evaluating the trustworthiness of interactive visualization tools for healthcare and medical diagnosis data and provides a novel idea and path for future healthcare experts. Our goal in this research was to make an idealness assessment of the trustworthiness impact of interactive visualization models under fuzzy conditions by using a medical fuzzy expert system based on an analytical network process and technique for ordering preference by similarity to ideal solutions. To eliminate the ambiguities that arose due to the multiple opinions of these experts and to externalize and organize information about the selection context of the interactive visualization models, the study used the proposed hybrid decision model. According to the results achieved through trustworthiness assessments of different visualization tools, BoldBI was found to be the most prioritized and trustworthy visualization tool among other alternatives. The suggested study would aid healthcare and medical professionals in interactive data visualization in identifying, selecting, prioritizing, and evaluating useful and trustworthy visualization-related characteristics, thereby leading to more accurate medical diagnosis profiles.