Background: Chemotherapy-induced nausea and vomiting (CINV) are the two most frightful and unpleasant side effects of chemotherapy. CINV is accountable for poor treatment outcomes, treatment failure, or even death. It can affect patients' overall quality of life, leading to many social, economic, and clinical consequences. Objective: This study compared the performances of different data mining models for predicting the risk of CINV among the patients and developed a smartphone app for clinical decision support to recommend the risk of CINV at the point of care. Methods: Data were collected by retrospective record review from the electronic medical records used at the University of Missouri Ellis Fischel Cancer Center. Patients who received chemotherapy and standard antiemetics at the oncology outpatient service from June 1, 2010, to July 31, 2012, were included in the study. There were six independent data sets of patients based on emetogenicity (low, moderate, and high) and two phases of CINV (acute and delayed). A total of 14 risk factors of CINV were chosen for data mining. For our study, we used five popular data mining algorithms: (1) naive Bayes algorithm, (2) logistic regression classifier, (3) neural network, (4) support vector machine (using sequential minimal optimization), and (5) decision tree. Performance measures, such as accuracy, sensitivity, and specificity with 10-fold cross-validation, were used for model comparisons. A smartphone app called CINV Risk Prediction Application was developed using the ResearchKit in iOS utilizing the decision tree algorithm, which conforms to the criteria of explainable, usable, and actionable artificial intelligence. The app was created using both the bulk questionnaire approach and the adaptive approach. Results: The decision tree performed well in both phases of high emetogenic chemotherapies, with a significant margin compared to the other algorithms. The accuracy measure for the six patient groups ranged from 79.3% to 94.8%. The app was developed using the results from the decision tree because of its consistent performance and simple, explainable nature. The bulk questionnaire approach asks 14 questions in the smartphone app, while the adaptive approach can determine questions based on the previous questions' answers. The adaptive approach saves time and can be beneficial when used at the point of care. Conclusions: This study solved a real clinical problem, and the solution can be used for personalized and precise evidence-based CINV management, leading to a better life quality for patients and reduced health care costs.

The de-identification of clinical reports is essential to protect the confidentiality of patients. The natural-language-processing-based named entity recognition (NER) model is a widely used technique of automatic clinical de-identification. The performance of such a machine learning model relies largely on the proper selection of features. The objective of this study was to investigate the utility of various features in a conditional-random-field (CRF)-based NER model. Natural language processing (NLP) toolkits were used to annotate the protected health information (PHI) from a total of 10,239 radiology reports that were divided into seven types. Multiple features were extracted by the toolkit and the NER models were built using these features and their combinations. A total of 10 features were extracted and the performance of the models was evaluated based on their precision, recall, and F₁-score. The best-performing features were n-gram, prefix-suffix, word embedding, and word shape. These features outperformed others across all types of reports. The dataset we used was large in volume and divided into multiple types of reports. Such a diverse dataset made sure that the results were not subject to a small number of structured texts from where a machine learning model can easily learn the features. The manual de-identification of large-scale clinical reports is impractical. This study helps to identify the best-performing features for building an NER model for automatic de-identification from a wide array of features mentioned in the literature.

Background: Breast cancer poses a significant health challenge in Bangladesh, where limited screening and unique reproductive patterns contribute to delayed diagnoses and subtype-specific disparities. While reproductive risk factors such as age at menarche, parity, and contraceptive use are well studied in high-income countries, their associations with hormone-receptor-positive (HR+) and triple-negative breast cancer (TNBC) remain underexplored in low-resource settings. Methods: A case-control study was conducted at the National Institute of Cancer Research and Hospital (NICRH) including 486 histopathologically confirmed breast cancer cases (246 HR+, 240 TNBC) and 443 cancer-free controls. Socio-demographic and reproductive data were collected through structured interviews. Machine learning models—including Logistic Regression, Lasso, Support Vector Machines, Random Forest, and XGBoost—were trained using stratified five-fold cross-validation. Model performance was evaluated using sensitivity, F1-score, and Area Under Receiver Operating Curve (AUROC). To interpret model predictions and quantify the contribution of individual features, we employed Shapley Additive exPlanation (SHAP) values. Results: XGBoost achieved the highest overall performance (F1-score = 0.750), and SHAP-based interpretability revealed key predictors for each subtype. Rural residence, low education (≤5 years), and undernutrition were significant predictors across subtypes. Cesarean delivery and multiple abortions were more predictive of TNBC, while urban residence, employment, and higher education were more predictive of HR+. Age at menarche and age at first childbirth showed decreasing predictive importance with increasing age for HR+, while larger gaps between marriage and childbirth were more predictive of TNBC. Conclusions: Our findings underscore the value of machine learning coupled with SHAP-based explainability in identifying context-specific risk factors for breast cancer subtypes in resource-limited settings. This approach enhances transparency and supports the development of targeted public health interventions to reduce breast cancer disparities in Bangladesh.

Background: Breast cancer poses a significant health challenge in Bangladesh, where limited screening and unique reproductive patterns contribute to delayed diagnoses and subtype-specific disparities. While reproductive risk factors such as age at menarche, parity, and contraceptive use are well-studied in high-income countries, their associations with hormone receptor-positive (HR+) and triple-negative breast cancer (TNBC) remain underexplored in low-resource settings. Methods: A case-control study was conducted at the National Institute of Cancer Research and Hospital (NICRH) including 486 histopathologically confirmed breast cancer cases (246 HR+, 240 TNBC) and 443 cancer-free controls. Socio-demographic and reproductive data were collected through structured interviews. Machine learning models—including Logistic Regression, Lasso, Support Vector Machines, Random Forest, and XGBoost—were trained using stratified 5-fold cross-validation. Model performance was evaluated using sensitivity, F1-score, and AUC. Feature importance was interpreted using Shapley values. Results: XGBoost achieved the highest overall performance (F1-score = 0.750). Rural residence, low education (≤5 years), and undernutrition were significant predictors across subtypes. Cesarean delivery and multiple abortions were more predictive of TNBC, while urban residence, employment, and higher education were more predictive of HR+. Age at menarche and age at first childbirth showed decreasing predictive importance with increasing age for HR+, while larger gaps between marriage and childbirth were more predictive of TNBC.

Purpose: Early and reliable disease prediction from structured clinical data remains challenging when datasets are small, highly imbalanced, and contain limited positive disease cases. Conventional machine learning (ML) and deep learning approaches often struggle to capture clinically meaningful relationships under such low-data representation conditions due to weak statistical associations between features and prediction targets. This study proposes a clinically grounded Distil GPT2-based table-to-text framework for disease prediction using structured healthcare datasets, motivated by the contextual reasoning capability of GPT models to better capture clinically meaningful relationships when statistical learning alone becomes insufficient due to limited data availability. Methods: Structured clinical records were transformed into physician-style textual descriptions and enriched through GPT4-generated medical paraphrasing to improve minority-class representation while preserving clinical meaning. Both the original and generated clinical texts were used to fine-tune a Distil GPT2 model across four public healthcare datasets, including heart disease, heart failure, chronic kidney disease, and thyroid cancer recurrence. Gradient-based explainable AI analysis was additionally incorporated to identify clinically important features influencing prediction outcomes. Results: The proposed framework demonstrated consistently strong predictive performance across four clinical datasets, achieving average precision, specificity, sensitivity, and F1-score of 0.96, 0.97, 0.96, and 0.96, respectively. The model achieved improved sensitivity, stronger generalization, and more stable predictive behavior compared with traditional ML, deep learning, transformer-based, and GAN-augmented approaches. Importantly, the framework consistently emphasized clinically meaningful variables even under severe class imbalance, where conventional ML and neural network models often struggled to identify key clinically relevant relationships. Conclusion: The proposed Distil GPT2-based tableto-text framework provides a practical and clinically interpretable approach for disease prediction from limited structured healthcare data. By integrating contextual clinical reasoning with explainable prediction mechanisms, the framework suggests strong potential for early risk detection, transparent clinical decision support, and reliable deployment in real-world data-scarce healthcare settings.