Predictive models for clinical decision making: Deep dives in practical machine learning

A Comprehensive Study on Self-Learning Methods and Implications to Autonomous Driving

As artificial intelligence (AI) has already seen numerous successful applications, the upcoming challenge lies in how to realize artificial general intelligence (AGI). Self-learning algorithms can autonomously acquire knowledge and adapt to new, demanding applications, recognized as one of the most effective techniques to overcome this challenge. Although many related studies have been conducted, there is still no comprehensive and systematic review available, nor well-founded recommendations for the application of autonomous intelligent systems, especially autonomous driving. As a result, this article comprehensively analyzes and classifies self-learning algorithms into three categories: broad self-learning, narrow self-learning, and limited self-learning. These categories are used to describe the popular usage, the most promising techniques, and the current status of hybridization with self-supervised learning. Then, the narrow self-learning is divided into three parts based on the self-learning realization path: sample self-learning, model self-learning, and self-learning architecture. For each method, this article discusses in detail its self-learning capacity, challenges, and applications to autonomous driving. Finally, the future research directions of self-learning algorithms are pointed out. It is expected that this study has the potential to eventually contribute to revolutionizing autonomous driving technology.

Simple Clinical Prediction Rules for Identifying Significant Liver Fibrosis: Evaluation of Established Scores and Development of the Aspartate Aminotransferase-Thrombocytopenia-Albumin (ATA) Score

Background: Existing non-invasive tests (NITs) for liver fibrosis offer moderate precision and accessibility but are often limited by complexity, reducing their practicality in routine clinical use. This study aimed to evaluate the diagnostic performance of current fibrosis assessment methods and develop a novel, simplified scoring system-the Aspartate Aminotransferase (AST)-Thrombocytopenia-Albumin (ATA) score-to enhance ease of use and clinical applicability. Methods: This study examined past cases of patients with chronic liver disease (CLD) by using magnetic resonance elastography (MRE) to evaluate fibrosis stages. Serum biomarkers were collected, and common fibrosis scores were calculated. Logistic regression identified potential predictors of significant fibrosis, forming the ATA score. Diagnostic performance was assessed, and internal validation was conducted via bootstrap resampling. Results: Among 70 patients, 31.4% had significant fibrosis. Hepatitis B was the most common cause (60.0%), followed by hepatitis C (18.6%) and nonalcoholic fatty liver disease (NAFLD, 15.7%). The ATA score demonstrated an area under the receiver operating characteristic curve (AUROC) of 0.872, comparable to the AST-to-platelet ratio index (APRI; 0.858) and fibrosis-4 index (FIB-4; 0.847). The recommended cut-offs for identifying high-risk patients were ATA score ≥ 2 (specificity 95.8%, sensitivity 50.0%), APRI ≥ 0.50 (specificity 89.6%, sensitivity 68.2%), and FIB-4 ≥ 1.3 (specificity 58.3%, sensitivity 90.9%). Internal validation confirmed model robustness, with an optimism-corrected AUROC of 0.8551. Conclusions: The ATA score offers a straightforward and efficient method for detecting significant fibrosis, demonstrating comparable diagnostic capability to APRI and FIB-4, while being more user-friendly in clinical practice. A score of 0-1 indicates low risk, suitable for clinical follow-up, whereas a score of ≥2 suggests high risk, warranting further evaluation. Integrating the ATA score into clinical workflows can enhance early detection, optimize resource utilization, and improve patient care.

Bioenergetic biomarkers as predictive indicators and their relationship with cognitive function in newly diagnosed, drug-naïve patients with bipolar disorder

Mitochondrial dysfunction and disrupted bioenergetic processes are critical in the pathogenesis of bipolar disorder (BD), with cognitive impairment being a prominent symptom linked to mitochondrial anomalies. The tricarboxylic acid (TCA) cycle, integral to mitochondrial energy production, may be implicated in this cognitive dysfunction, yet its specific association with BD remains underexplored. In this cross-sectional study, 144 first-episode, drug-naive BD patients and 51 healthy controls were assessed. Using liquid chromatography-tandem mass spectrometry (LC-MS/MS), serum TCA cycle metabolites were quantified, and cognitive function was evaluated through the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) and the Stroop color-word test. The study found that BD patients exhibited significantly elevated serum levels of several TCA metabolites compared to healthy controls, alongside lower cognitive function scores. Correlational analyses revealed that certain bioenergetic metabolites were significantly positively associated with anxiety and negatively correlated with cognitive performance in BD patients. Notably, succinic acid, α-Ketoglutaric acid (α-KG), and malic acid emerged as independent risk factors for BD, with their combined profile demonstrating diagnostic utility. These findings underscore the potential of serum bioenergetic metabolites as biomarkers for BD, providing insights into the mitochondrial dysfunction underlying cognitive impairment and offering a basis for early diagnosis and targeted therapeutic strategies.

Radiomics and clinicoradiological factors as a promising approach for predicting microvascular invasion in hepatitis B-related hepatocellular carcinoma

Microvascular invasion (MVI) is a critical factor in hepatocellular carcinoma (HCC) prognosis, particularly in hepatitis B virus (HBV)-related cases. This editorial examines a recent study by Xu et al who developed models to predict MVI and high-risk (M2) status in HBV-related HCC using contrast-enhanced computed tomography (CECT) radiomics and clinicoradiological factors. The study analyzed 270 patients, creating models that achieved an area under the curve values of 0.841 and 0.768 for MVI prediction, and 0.865 and 0.798 for M2 status prediction in training and validation datasets, respectively. These results are comparable to previous radiomics-based approaches, which reinforces the potential of this method in MVI prediction. The strengths of the study include its focus on HBV-related HCC and the use of widely accessible CECT imaging. However, limitations, such as retrospective design and manual segmentation, highlight areas for improvement. The editorial discusses the implications of the study including the need for standardized radiomics approaches and the potential impact on personalized treatment strategies. It also suggests future research directions, such as exploring mechanistic links between radiomics features and MVI, as well as integrating additional biomarkers or imaging modalities. Overall, this study contributes significantly to HCC management, paving the way for more accurate, personalized treatment approaches in the era of precision oncology.

Personalized Predictions for Changes in Knee Pain Among Patients With Osteoarthritis Participating in Supervised Exercise and Education: Prognostic Model Study

Background

Knee osteoarthritis (OA) is a common chronic condition that impairs mobility and diminishes quality of life. Despite the proven benefits of exercise therapy and patient education in managing OA pain and functional limitations, these strategies are often underused. To motivate and enhance patient engagement, personalized outcome prediction models can be used. However, the accuracy of existing models in predicting changes in knee pain outcomes remains insufficiently examined.

Objective

This study aims to validate existing models and introduce a concise personalized model predicting changes in knee pain from before to after participating in a supervised patient education and exercise therapy program (GLA:D) among patients with knee OA.

Methods

Our prediction models leverage self-reported patient information and functional measures. To refine the number of variables, we evaluated the variable importance and applied clinical reasoning. We trained random forest regression models and compared the rate of true predictions of our models with those using average values. In supplementary analyses, we additionally considered recently added variables to the GLA:D registry.

Results

We evaluated the performance of a full, continuous, and concise model including all 34 variables, all 11 continuous variables, and the 6 most predictive variables, respectively. All three models performed similarly and were comparable to the existing model, with R2 values of 0.31-0.32 and root-mean-squared errors of 18.65-18.85-despite our increased sample size. Allowing a deviation of 15 (visual analog scale) points from the true change in pain, our concise model correctly estimated the change in pain in 58% of cases, while using average values that resulted in 51% accuracy. Our supplementary analysis led to similar outcomes.

Conclusions

Our concise personalized prediction model provides more often accurate predictions for changes in knee pain after the GLA:D program than using average pain improvement values. Neither the increase in sample size nor the inclusion of additional variables improved previous models. Based on current knowledge and available data, no better predictions are possible. Guidance is needed on when a model's performance is good enough for clinical practice use.

An Explainable Artificial Intelligence Text Classifier for Suicidality Prediction in Youth Crisis Text Line Users: Development and Validation Study

BACKGROUND

Suicide represents a critical public health concern, and machine learning (ML) models offer the potential for identifying at-risk individuals. Recent studies using benchmark datasets and real-world social media data have demonstrated the capability of pretrained large language models in predicting suicidal ideation and behaviors (SIB) in speech and text.

OBJECTIVE

This study aimed to (1) develop and implement ML methods for predicting SIBs in a real-world crisis helpline dataset, using transformer-based pretrained models as a foundation; (2) evaluate, cross-validate, and benchmark the model against traditional text classification approaches; and (3) train an explainable model to highlight relevant risk-associated features.

METHODS

We analyzed chat protocols from adolescents and young adults (aged 14-25 years) seeking assistance from a German crisis helpline. An ML model was developed using a transformer-based language model architecture with pretrained weights and long short-term memory layers. The model predicted suicidal ideation (SI) and advanced suicidal engagement (ASE), as indicated by composite Columbia-Suicide Severity Rating Scale scores. We compared model performance against a classical word-vector-based ML model. We subsequently computed discrimination, calibration, clinical utility, and explainability information using a Shapley Additive Explanations value-based post hoc estimation model.

RESULTS

The dataset comprised 1348 help-seeking encounters (1011 for training and 337 for testing). The transformer-based classifier achieved a macroaveraged area under the curve (AUC) receiver operating characteristic (ROC) of 0.89 (95% CI 0.81-0.91) and an overall accuracy of 0.79 (95% CI 0.73-0.99). This performance surpassed the word-vector-based baseline model (AUC-ROC=0.77, 95% CI 0.64-0.90; accuracy=0.61, 95% CI 0.61-0.80). The transformer model demonstrated excellent prediction for nonsuicidal sessions (AUC-ROC=0.96, 95% CI 0.96-0.99) and good prediction for SI and ASE, with AUC-ROCs of 0.85 (95% CI 0.97-0.86) and 0.87 (95% CI 0.81-0.88), respectively. The Brier Skill Score indicated a 44% improvement in classification performance over the baseline model. The Shapley Additive Explanations model identified language features predictive of SIBs, including self-reference, negation, expressions of low self-esteem, and absolutist language.

CONCLUSIONS

Neural networks using large language model-based transfer learning can accurately identify SI and ASE. The post hoc explainer model revealed language features associated with SI and ASE. Such models may potentially support clinical decision-making in suicide prevention services. Future research should explore multimodal input features and temporal aspects of suicide risk.

Survey and perspective on verification, validation, and uncertainty quantification of digital twins for precision medicine

Digital twins in precision medicine provide tailored health recommendations by simulating patient-specific trajectories and interventions. We examine the critical role of Verification, Validation, and Uncertainty Quantification (VVUQ) for digital twins in ensuring safety and efficacy, with examples in cardiology and oncology. We highlight challenges and opportunities for developing personalized trial methodologies, validation metrics, and standardizing VVUQ processes. VVUQ frameworks are essential for integrating digital twins into clinical practice.

Unlocking Patient Resistance to AI in Healthcare: A Psychological Exploration

Artificial intelligence (AI) has transformed healthcare, yet patients' acceptance of AI-driven medical services remains constrained. Despite its significant potential, patients exhibit reluctance towards this technology. A notable lack of comprehensive research exists that examines the variables driving patients' resistance to AI. This study explores the variables influencing patients' resistance to adopt AI technology in healthcare by applying an extended Ram and Sheth Model. More specifically, this research examines the roles of the need for personal contact (NPC), perceived technological dependence (PTD), and general skepticism toward AI (GSAI) in shaping patient resistance to AI integration. For this reason, a sequential mixed-method approach was employed, beginning with semi-structured interviews to identify adaptable factors in healthcare. It then followed with a survey to validate the qualitative findings through Structural Equation Modeling (SEM) via AMOS (version 24). The findings confirm that NPC, PTD, and GSAI significantly contribute to patient resistance to AI in healthcare. Precisely, patients who prefer personal interaction, feel dependent on AI, or are skeptical of AI's promises are more likely to resist its adoption. The findings highlight the psychological factors driving patient reluctance toward AI in healthcare, offering valuable insights for healthcare administrators. Strategies to balance AI's efficiency with human interaction, mitigate technological dependence, and foster trust are recommended for successful implementation of AI. This research adds to the theoretical understanding of Innovation Resistance Theory, providing both conceptual insights and practical implications for the effective incorporation of AI in healthcare.

Development of a survey-based stacked ensemble predictive model for autonomy preferences in patients with periodontal disease

OBJECTIVES

This study aimed to develop a model to predict the autonomy preference (AP) and satisfaction after tooth extraction (STE) in patients with periodontal disease. Understanding of individual AP and STE is essential for improving patient satisfaction and promoting informed decision-making in periodontics.

METHODS

A stacked ensemble machine learning model was used to predict patient AP and STE based on the results of a survey that included demographic information, oral health status, AP index, and STE. Data from 421 patients with periodontal disease were collected from two university dental hospitals and evaluated for ensemble modeling in the following predictive models: random forest, naïve Bayes, gradient boost, adaptive boost, and XGBoost.

RESULTS

The models demonstrated good predictive performance, with XGBoost demonstrating the highest accuracy for both AP (0.78) and STE (0.80). The results showed that only 7.6 % of patients had high AP, which tended to decrease with age and varied significantly according to education level and severity of treatment, categorized as supportive periodontal treatment, active periodontal treatment, or extraction and/or dental implant procedures. Additionally, the majority of patients (67.7 %) reported high STE levels, highlighting the effectiveness of the model in accurately predicting AP, which was further supported by the significant correlation between accurately predicted AP levels and high STE outcomes.

CONCLUSIONS

The successful utilization of a stacked ensemble model to predict patient AP and STE demonstrates the potential of machine learning to improve patient-centered care in periodontics. Future research should extend to more diverse patient populations and clinical conditions to validate and refine the predictive abilities of such models in broader healthcare settings.

CLINICAL SIGNIFICANCE

The machine learning-based predictive model effectively enhances personalized decision-making and improves patient satisfaction in periodontal treatment.

Evaluating predictive artificial intelligence approaches used in mobile health platforms to forecast mental health symptoms among youth: a systematic review

The youth mental health crisis is exacerbated by limited access to care and resources. Mobile health (mHealth) platforms using predictive artificial intelligence (AI) can improve access and reduce barriers, enabling real-time responses and precision prevention. This systematic review evaluates predictive AI approaches in mHealth platforms for forecasting mental health symptoms among youth (13-25 years). We searched studies from Embase, PubMed, Web of Science, PsycInfo, and CENTRAL, to identify relevant studies. From 11 studies identified, three studies predicted multiple symptoms, with depression being the most common (63%). Most platforms used smartphones and 25% integrated wearables. Key predictors included smartphone usage (N=5), sleep metrics (N=6), and physical activity (N=5). Nuanced predictors like usage locations and sleep stages improved prediction. Logistic regression was most used (N=6), followed by Support Vector Machines (N=3) and ensemble methods (N=4). F-scores for anxiety and depression ranged from 0.73 to 0.84, and AUCs from 0.50 to 0.74. Stress models had AUCs of 0.68 to 0.83. Bayesian model selection and Shapley values enhanced robustness and interpretability. Barriers included small sample sizes, privacy concerns, missing data, and underrepresentation bias. Rigorous evaluation of predictive performance, generalizability, and user engagement is critical before mHealth platforms are integrated into psychiatric care.

Exploration of simultaneous transients between cerebral hemodynamics and the autonomic nervous system using windowed time-lagged cross-correlation matrices: a CENTER-TBI study

BACKGROUND

Traumatic brain injury (TBI) can significantly disrupt autonomic nervous system (ANS) regulation, increasing the risk for secondary complications, hemodynamic instability, and adverse outcome. This retrospective study evaluated windowed time-lagged cross-correlation (WTLCC) matrices for describing cerebral hemodynamics-ANS interactions to predict outcome, enabling identifying high-risk patients who may benefit from enhanced monitoring to prevent complications.

METHODS

The first experiment aimed to predict short-term outcome using WTLCC-based convolution neural network models on the Wroclaw University Hospital (WUH) database (Ptraining = 31 with 1,079 matrices, Pval = 16 with 573 matrices). The second experiment predicted long-term outcome, training on the CENTER-TBI database (Ptraining = 100 with 17,062 matrices) and validating on WUH (Pval = 47 with 6,220 matrices). Cerebral hemodynamics was characterized using intracranial pressure (ICP), cerebral perfusion pressure (CPP), pressure reactivity index (PRx), while ANS metrics included low-to-high-frequency heart rate variability (LF/HF) and baroreflex sensitivity (BRS) over 72 h. Short-term outcome at WUH was assessed using the Glasgow Outcome Scale (GOS) at discharge. Long-term outcome was evaluated at 3 months at WUH and 6 months at CENTER-TBI using GOS and GOS-Extended, respectively. The XGBoost model was used to compare performance of WTLCC-based model and averaged neuromonitoring parameters, adjusted for age, Glasgow Coma Scale, major extracranial injury, and pupil reactivity in outcome prediction.

RESULTS

For short-term outcome prediction, the best-performing WTLCC-based model used ICP-LF/HF matrices. It had an area under the curve (AUC) of 0.80, vs. 0.71 for averages of ANS and cerebral hemodynamics metrics, adjusted for clinical metadata. For long-term outcome prediction, the best-score WTLCC-based model used ICP-LF/HF matrices. It had an AUC of 0.63, vs. 0.66 for adjusted neuromonitoring parameters.

CONCLUSIONS

Among all neuromonitoring parameters, ICP and LF/HF signals were the most effective in generating the WTLCC matrices. WTLCC-based model outperformed adjusted neuromonitoring parameters in short-term but had moderate utility in long-term outcome prediction.

Designing and testing clinical simulations of an early warning system for implementation in acute care settings

Objectives

Conducting simulation testing with end-users is essential for facilitating successful implementation of new health information technologies. This study designed a standardized simulation testing process with a system prototype prior to implementation to help study teams identify the system's interpretability and feasibility from the end-user perspective and to effectively integrate new innovations into real-world clinical settings and workflows.

Materials and Methods

A clinical simulation model was developed to test a new Clinical Decision Support (CDS) system outside of the clinical environment while maintaining high fidelity. A web-based CDS prototype, the "CONCERN Smart Application," which leverages clinical data to measure and express a patient's risk of deterioration on a 3-level scale ("low," "moderate," or "high"), and audiovisual-integrated materials, were used to lead simulation sessions.

Results

A total of 6 simulation sessions with 17 nurses were held to investigate how nurses interact with the CONCERN Smart application and how it influences their critical thinking, and clinical responses. Four themes were extracted from the simulation debriefing sessions and used to inform implementation strategies. The strategies include how the CDS should be improved for practical real-world use.

Discussion and Conclusions

Standardized simulation testing procedures identified and informed the necessary CDS improvements, the enhancements needed for real-world use, and the training requirements to effectively prepare end-users for system go-live.

Predictive modeling of preoperative acute heart failure in older adults with hypertension: a dual perspective of SHAP values and interaction analysis

BACKGROUND

In older adults with hypertension, hip fractures accompanied by preoperative acute heart failure significantly elevate surgical risks and adverse outcomes, necessitating timely identification and management to improve patient outcomes.

RESEARCH OBJECTIVE

This study aims to enhance the early recognition of acute heart failure in older hypertensive adults prior to hip fracture surgery by developing a predictive model using logistic regression (LR) and machine learning methods, optimizing preoperative assessment and management.

METHODS

Employing a retrospective study design, we analyzed hypertensive older adults who underwent hip fracture surgery at Hebei Medical University Third Hospital from January 2018 to December 2022. Predictive models were constructed using LASSO regression and multivariable logistic regression, evaluated via nomogram charts. Five additional machine learning methods were utilized, with variable importance assessed using SHAP values and the impact of key variables evaluated through multivariate correlation analysis and interaction effects.

RESULTS

The study included 1,370 patients. LASSO regression selected 18 key variables, including sex, age, coronary heart disease, pulmonary infection, ventricular arrhythmias, acute myocardial infarction, and anemia. The logistic regression model demonstrated robust performance with an AUC of 0.753. Although other models outperformed it in sensitivity and F1 score, logistic regression's discriminative ability was significant for clinical decision-making. The Gradient Boosting Machine model, notable for a sensitivity of 95.2%, indicated substantial capability in identifying patients at risk, crucial for reducing missed diagnoses.

CONCLUSION

We developed and compared efficacy of predictive models using logistic regression and machine learning, interpreting them with SHAP values and analyzing key variable interactions. This offers a scientific basis for assessing preoperative heart failure risk in older adults with hypertension and hip fractures, providing significant guidance for individualized treatment strategies and underscoring the value of applying machine learning in clinical settings.

Identification of novel hypertension biomarkers using explainable AI and metabolomics

BACKGROUND

The global incidence of hypertension, a condition of elevated blood pressure, is rising alarmingly. According to the World Health Organization's Qatar Hypertension Profile for 2023, around 33% of adults are affected by hypertension. This is a significant public health concern that can lead to serious health complications if left untreated. Metabolic dysfunction is a primary cause of hypertension. By studying key biomarkers, we can discover new treatments to improve the lives of those with high blood pressure.

AIMS

This study aims to use explainable artificial intelligence (XAI) to interpret novel metabolite biosignatures linked to hypertension in Qatari Population.

METHODS

The study utilized liquid chromatography-mass spectrometry (LC/MS) method to profile metabolites from biosamples of Qatari nationals diagnosed with stage 1 hypertension (n = 224) and controls (n = 554). Metabolon platform was used for the annotation of raw metabolite data generated during the process. A comprehensive series of analytical procedures, including data trimming, imputation, undersampling, feature selection, and biomarker discovery through explainable AI (XAI) models, were meticulously executed to ensure the accuracy and reliability of the results.

RESULTS

Elevated Vanillylmandelic acid (VMA) levels are markedly associated with stage 1 hypertension compared to controls. Glycerophosphorylcholine (GPC), N-Stearoylsphingosine (d18:1/18:0)*, and glycine are critical metabolites for accurate hypertension prediction. The light gradient boosting model yielded superior results, underscoring the potential of our research in enhancing hypertension diagnosis and treatment. The model's classification metrics: accuracy (78.13%), precision (78.13%), recall (78.13%), F1-score (78.13%), and AUROC (83.88%) affirm its efficacy. SHapley Additive exPlanations (SHAP) further elucidate the metabolite markers, providing a deeper understanding of the disease's pathology.

CONCLUSION

This study identified novel metabolite biomarkers for precise hypertension diagnosis using XAI, enhancing early detection and intervention in the Qatari population.

Classifying High-Risk Patients for Persistent Opioid Use After Major Spine Surgery: A Machine-Learning Approach

BACKGROUND

Persistent opioid use is a common occurrence after surgery and prolonged exposure to opioids may result in escalation and dependence. The objective of this study was to develop machine-learning-based predictive models for persistent opioid use after major spine surgery.

METHODS

Five classification models were evaluated to predict persistent opioid use: logistic regression, random forest, neural network, balanced random forest, and balanced bagging. Synthetic Minority Oversampling Technique was used to improve class balance. The primary outcome was persistent opioid use, defined as patient reporting to use opioids after 3 months postoperatively. The data were split into a training and test set. Performance metrics were evaluated on the test set and included the F1 score and the area under the receiver operating characteristics curve (AUC). Feature importance was ranked based on SHapley Additive exPlanations (SHAP).

RESULTS

After exclusion (patients with missing follow-up data), 2611 patients were included in the analysis, of which 1209 (46.3%) continued to use opioids 3 months after surgery. The balanced random forest classifiers had the highest AUC (0.877, 95% confidence interval [CI], 0.834-0.894) compared to neural networks (0.729, 95% CI, 0.672-0.787), logistic regression (0.709, 95% CI, 0.652-0.767), balanced bagging classifier (0.859, 95% CI, 0.814-0.905), and random forest classifier (0.855, 95% CI, 0.813-0.897). The balanced random forest classifier had the highest F1 (0.758, 95% CI, 0.677-0.839). Furthermore, the specificity, sensitivity, precision, and accuracy were 0.883, 0.700, 0.836, and 0.780, respectively. The features based on SHAP analysis with the highest impact on model performance were age, preoperative opioid use, preoperative pain scores, and body mass index.

CONCLUSIONS

The balanced random forest classifier was found to be the most effective model for identifying persistent opioid use after spine surgery.

Smart variant filtering - A blueprint solution for massively parallel sequencing-based variant analysis

Massively parallel sequencing helps create new knowledge on genes, variants and their association with disease phenotype. This important technological advancement simultaneously makes clinical decision making, using genomic information for cancer patients, more complex. Currently, identifying actionable pathogenic variants with diagnostic, prognostic, or predictive impact requires substantial manual effort. Objective: The purpose is to design a solution for clinical diagnostics of lymphoma, specifically for systematic variant filtering and interpretation. Methods: A scoping review and demonstrations from specialists serve as a basis for a blueprint of a solution for massively parallel sequencing-based genetic diagnostics. Results: The solution uses machine learning methods to facilitate decision making in the diagnostic process. A validation round of interviews with specialists consolidated the blueprint and anchored it across all relevant expert disciplines. The scoping review identified four components of variant filtering solutions: algorithms and Artificial Intelligence (AI) applications, software, bioinformatics pipelines and variant filtering strategies. The blueprint describes the input, the AI model and the interface for dynamic browsing. Conclusion: An AI-augmented system is designed for predicting pathogenic variants. While such a system can be used to classify identified variants, diagnosticians should still evaluate the classification's accuracy, make corrections when necessary, and ultimately decide which variants are truly pathogenic.

AI hybrid survival assessment for advanced heart failure patients with renal dysfunction

Renal dysfunction (RD) often characterizes the worse course of patients with advanced heart failure (AHF). Many prognosis assessments are hindered by researcher biases, redundant predictors, and lack of clinical applicability. In this study, we enroll 1736 AHF/RD patients, including data from Henan Province Clinical Research Center for Cardiovascular Diseases (which encompasses 11 hospital subcenters), and Beth Israel Deaconess Medical Center. We developed an AI hybrid modeling framework, assembling 12 learners with different feature selection paradigms to expand modeling schemes. The optimized strategy is identified from 132 potential schemes to establish an explainable survival assessment system: AIHFLevel. The conditional inference survival tree determines a probability threshold for prognostic stratification. The evaluation confirmed the system's robustness in discrimination, calibration, generalization, and clinical implications. AIHFLevel outperforms existing models, clinical features, and biomarkers. We also launch an open and user-friendly website www.hf-ai-survival.com , empowering healthcare professionals with enhanced tools for continuous risk monitoring and precise risk profiling.

Advances in AI-assisted biochip technology for biomedicine

The integration of biochips with AI opened up new possibilities and is expected to revolutionize smart healthcare tools within the next five years. The combination of miniaturized, multi-functional, rapid, high-throughput sample processing and sensing capabilities of biochips, with the computational data processing and predictive power of AI, allows medical professionals to collect and analyze vast amounts of data quickly and efficiently, leading to more accurate and timely diagnoses and prognostic evaluations. Biochips, as smart healthcare devices, offer continuous monitoring of patient symptoms. Integrated virtual assistants have the potential to send predictive feedback to users and healthcare practitioners, paving the way for personalized and predictive medicine. This review explores the current state-of-the-art biochip technologies including gene-chips, organ-on-a-chips, and neural implants, and the diagnostic and therapeutic utility of AI-assisted biochips in medical practices such as cancer, diabetes, infectious diseases, and neurological disorders. Choosing the appropriate AI model for a specific biomedical application, and possible solutions to the current challenges are explored. Surveying advances in machine learning models for biochip functionality, this paper offers a review of biochips for the future of biomedicine, an essential guide for keeping up with trends in healthcare, while inspiring cross-disciplinary collaboration among biomedical engineering, medicine, and machine learning fields.

Enhanced deep learning model for precise nodule localization and recurrence risk prediction following curative-intent surgery for lung cancer

PURPOSE

Radical surgery is the primary treatment for early-stage resectable lung cancer, yet recurrence after curative surgery is not uncommon. Identifying patients at high risk of recurrence using preoperative computed tomography (CT) images could enable more aggressive surgical approaches, shorter surveillance intervals, and intensified adjuvant treatments. This study aims to analyze lung cancer sites in CT images to predict potential recurrences in high-risk individuals.

METHODS

We retrieved anonymized imaging and clinical data from an institutional database, focusing on patients who underwent curative pulmonary resections for non-small cell lung cancers. Our study used a deep learning model, the Mask Region-based Convolutional Neural Network (MRCNN), to predict cancer locations and assign recurrence classification scores. To find optimized trained weighted values in the model, we developed preprocessing python codes, adjusted dynamic learning rate, and modifying hyper parameter in the model.

RESULTS

The model training completed; we performed classifications using the validation dataset. The results, including the confusion matrix, demonstrated performance metrics: bounding box (0.390), classification (0.034), mask (0.266), Region Proposal Network (RPN) bounding box (0.341), and RPN classification (0.054). The model successfully identified lung cancer recurrence sites, which were then accurately mapped onto chest CT images to highlight areas of primary concern.

CONCLUSION

The trained model allows clinicians to focus on lung regions where cancer recurrence is more likely, acting as a significant aid in the detection and diagnosis of lung cancer. Serving as a clinical decision support system, it offers substantial support in managing lung cancer patients.

TLE4 downregulation identified by WGCNA and machine learning algorithm promotes papillary thyroid carcinoma progression via activating JAK/STAT pathway

Background: Papillary Thyroid Carcinoma (PTC), a common type of thyroid cancer, has a pathogenesis that is not fully understood. This study utilizes a range of public databases, sophisticated bioinformatics tools, and empirical approaches to explore the key genetic components and pathways implicated in PTC, particularly concentrating on the Transducin-Like Enhancer of Split 4 (TLE4) gene. Methods: Public databases such as TCGA and GEO were utilized to conduct differential gene expression analysis in PTC. Hub genes were identified using Weighted Gene Co-expression Network Analysis (WGCNA), and machine learning techniques, including Random Forest, LASSO regression, and SVM-RFE, were employed for biomarker identification. The clinical impact of the TLE4 gene was assessed in terms of diagnostic accuracy, prognostic value, and its functional enrichment analysis in PTC. Additionally, the study focused on understanding the role of TLE4 in the dynamics of immune cell infiltration, gene function enhancement, and behaviors of PTC cells like growth, migration, and invasion. To complement these analyses, in vivo studies were performed using a xenograft mouse model. Results: 244 genes with significant differential expression across various databases were identified. WGCNA indicated a strong link between specific gene modules and PTC. Machine learning analysis brought the TLE4 gene into focus as a key biomarker. Bioinformatics studies verified that TLE4 expression is lower in PTC, linking it to immune cell infiltration and the JAK-STAT signaling pathways. Experimental data revealed that decreased TLE4 expression in PTC cell lines leads to enhanced cell growth, migration, invasion, and activates the JAK/STAT pathway. In contrast, TLE4 overexpression in these cells inhibited tumor growth and metastasis. Conclusions: This study sheds light on TLE4's crucial role in PTC pathogenesis, positioning it as a potential biomarker and target for therapy. The integration of multi-omics data and advanced analytical methods provides a robust framework for understanding PTC at a molecular level, potentially guiding personalized treatment strategies.

Refining outcome prediction after traumatic brain injury with machine learning algorithms

Outcome after traumatic brain injury (TBI) is typically assessed using the Glasgow outcome scale extended (GOSE) with levels from 1 (death) to 8 (upper good recovery). Outcome prediction has classically been dichotomized into either dead/alive or favorable/unfavorable outcome. Binary outcome prediction models limit the possibility of detecting subtle yet significant improvements. We set out to explore different machine learning methods with the purpose of mapping their predictions to the full 8 grade scale GOSE following TBI. The models were set up using the variables: age, GCS-motor score, pupillary reaction, and Marshall CT score. For model setup and internal validation, a total of 866 patients could be included. For external validation, a cohort of 369 patients were included from Leuven, Belgium, and a cohort of 573 patients from the US multi-center ProTECT III study. Our findings indicate that proportional odds logistic regression (POLR), random forest regression, and a neural network model achieved accuracy values of 0.3-0.35 when applied to internal data, compared to the random baseline which is 0.125 for eight categories. The models demonstrated satisfactory performance during external validation in the data from Leuven, however, their performance were not satisfactory when applied to the ProTECT III dataset.

Machine Learning-Based Survival Prediction Models for Progression-Free and Overall Survival in Advanced-Stage Hodgkin Lymphoma

PURPOSE

Patients diagnosed with advanced-stage Hodgkin lymphoma (aHL) have historically been risk-stratified using the International Prognostic Score (IPS). This study investigated if a machine learning (ML) approach could outperform existing models when it comes to predicting overall survival (OS) and progression-free survival (PFS).

PATIENTS AND METHODS

This study used patient data from the Danish National Lymphoma Register for model development (development cohort). The ML model was developed using stacking, which combines several predictive survival models (Cox proportional hazard, flexible parametric model, IPS, principal component, penalized regression) into a single model, and was compared with two versions of IPS (IPS-3 and IPS-7) and the newly developed aHL international prognostic index (A-HIPI). Internal model validation was performed using nested cross-validation, and external validation was performed using patient data from the Swedish Lymphoma Register and Cancer Registry of Norway (validation cohort).

RESULTS

In total, 707 and 760 patients with aHL were included in the development and validation cohorts, respectively. Examining model performance for OS in the development cohort, the concordance index (C-index) for the ML model, IPS-7, IPS-3, and A-HIPI was found to be 0.789, 0.608, 0.650, and 0.768, respectively. The corresponding estimates in the validation cohort were 0.749, 0.700, 0.663, and 0.741. For PFS, the ML model achieved the highest C-index in both cohorts (0.665 in the development cohort and 0.691 in the validation cohort). The time-varying AUCs for both the ML model and the A-HIPI were consistently higher in both cohorts compared with the IPS models within the first 5 years after diagnosis.

CONCLUSION

The new prognostic model for aHL on the basis of ML techniques demonstrated a substantial improvement compared with the IPS models, but yielded a limited improvement in predictive performance compared with the A-HIPI.

The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century

As healthcare systems around the world face challenges such as escalating costs, limited access, and growing demand for personalized care, artificial intelligence (AI) is emerging as a key force for transformation. This review is motivated by the urgent need to harness AI's potential to mitigate these issues and aims to critically assess AI's integration in different healthcare domains. We explore how AI empowers clinical decision-making, optimizes hospital operation and management, refines medical image analysis, and revolutionizes patient care and monitoring through AI-powered wearables. Through several case studies, we review how AI has transformed specific healthcare domains and discuss the remaining challenges and possible solutions. Additionally, we will discuss methodologies for assessing AI healthcare solutions, ethical challenges of AI deployment, and the importance of data privacy and bias mitigation for responsible technology use. By presenting a critical assessment of AI's transformative potential, this review equips researchers with a deeper understanding of AI's current and future impact on healthcare. It encourages an interdisciplinary dialogue between researchers, clinicians, and technologists to navigate the complexities of AI implementation, fostering the development of AI-driven solutions that prioritize ethical standards, equity, and a patient-centered approach.

Use of Digitalisation and Machine Learning Techniques in Therapeutic Intervention at Early Ages: Supervised and Unsupervised Analysis

Advances in technology and artificial intelligence (smart healthcare) open up a range of possibilities for precision intervention in the field of health sciences. The objectives of this study were to analyse the functionality of using supervised (prediction and classification) and unsupervised (clustering) machine learning techniques to analyse results related to the development of functional skills in patients at developmental ages of 0-6 years. We worked with a sample of 113 patients, of whom 49 were cared for in a specific centre for people with motor impairments (Group 1) and 64 were cared for in a specific early care programme for patients with different impairments (Group 2). The results indicated that in Group 1, chronological age predicted the development of functional skills at 85% and in Group 2 at 65%. The classification variable detected was functional development in the upper extremities. Two clusters were detected within each group that allowed us to determine the patterns of functional development in each patient with respect to functional skills. The use of smart healthcare resources has a promising future in the field of early care. However, data recording in web applications needs to be planned, and the automation of results through machine learning techniques is required.

Investigating the impact of Wnt pathway-related genes on biomarker and diagnostic model development for osteoporosis in postmenopausal females

The Wnt signaling pathway is essential for bone development and maintaining skeletal homeostasis, making it particularly relevant in osteoporosis patients. Our study aimed to identify distinct molecular clusters associated with the Wnt pathway and develop a diagnostic model for osteoporosis in postmenopausal Caucasian women. We downloaded three datasets (GSE56814, GSE56815 and GSE2208) related to osteoporosis from the GEO database. Our analysis identified a total of 371 differentially expressed genes (DEGs) between low and high bone mineral density (BMD) groups, with 12 genes associated with the Wnt signaling pathway, referred to as osteoporosis-associated Wnt pathway-related genes. Employing four independent machine learning models, we established a diagnostic model using the 12 osteoporosis-associated Wnt pathway-related genes in the training set. The XGB model showed the most promising discriminative potential. We further validate the predictive capability of our diagnostic model by applying it to three external datasets specifically related to osteoporosis. Subsequently, we constructed a diagnostic nomogram based on the five crucial genes identified from the XGB model. In addition, through the utilization of DGIdb, we identified a total of 30 molecular compounds or medications that exhibit potential as promising therapeutic targets for osteoporosis. In summary, our comprehensive analysis provides valuable insights into the relationship between the osteoporosis and Wnt signaling pathway.

Finding the Best Match - a Case Study on the (Text-)Feature and Model Choice in Digital Mental Health Interventions

With the need for psychological help long exceeding the supply, finding ways of scaling, and better allocating mental health support is a necessity. This paper contributes by investigating how to best predict intervention dropout and failure to allow for a need-based adaptation of treatment. We systematically compare the predictive power of different text representation methods (metadata, TF-IDF, sentiment and topic analysis, and word embeddings) in combination with supplementary numerical inputs (socio-demographic, evaluation, and closed-question data). Additionally, we address the research gap of which ML model types - ranging from linear to sophisticated deep learning models - are best suited for different features and outcome variables. To this end, we analyze nearly 16.000 open-text answers from 849 German-speaking users in a Digital Mental Health Intervention (DMHI) for stress. Our research proves that - contrary to previous findings - there is great promise in using neural network approaches on DMHI text data. We propose a task-specific LSTM-based model architecture to tackle the challenge of long input sequences and thereby demonstrate the potential of word embeddings (AUC scores of up to 0.7) for predictions in DMHIs. Despite the relatively small data set, sequential deep learning models, on average, outperform simpler features such as metadata and bag-of-words approaches when predicting dropout. The conclusion is that user-generated text of the first two sessions carries predictive power regarding patients' dropout and intervention failure risk. Furthermore, the match between the sophistication of features and models needs to be closely considered to optimize results, and additional non-text features increase prediction results.

Supplementary Information

The online version contains supplementary material available at 10.1007/s41666-023-00148-z.

Predictive modeling of postoperative gastrointestinal dysfunction: the role of serum bilirubin, sodium levels, and surgical duration in gynecological cancer care

OBJECTIVE

To elucidate the role of preoperative serum bilirubin and sodium levels, along with the duration of surgery, in predicting postoperative gastrointestinal dysfunction (POGD) following gynecological cancer surgery, informing tailored perioperative strategies.

METHODS

We conducted a retrospective analysis of 281 patients undergoing gynecological cancer surgery between 2018 and 2023. This analysis focused on preoperative serum bilirubin and sodium levels and intraoperative factors (surgical duration) as potential predictors of POGD. Logistic regression models were utilized for analysis, controlling for relevant confounders.

RESULTS

Elevated preoperative serum bilirubin was associated with a reduced risk of POGD (mean level in non-POGD cases: 14.172 ± 4.0701, vs. POGD cases: 9.6429 ± 3.5351; p <  0.001), suggesting a protective role. Lower preoperative sodium levels were identified in the POGD group (136.26 mEq/L [IQR: 135.2-137.63]) compared to the non-POGD group (139.32 mEq/L [IQR: 137.7-140.75]; p <  0.001), highlighting its predictive value. Additionally, longer surgical duration was associated with increased POGD incidence, with POGD cases experiencing surgeries lasting 6.1547 ± 1.9426 hours compared to 4.5959 ± 1.5475 hours in non-POGD cases (p <  0.001).

CONCLUSION

Our findings underscore the importance of serum bilirubin, sodium levels, and surgical duration as significant predictors of POGD in patients undergoing gynecological cancer surgery. These indicators should be integrated into a predictive model, aiding clinicians in identifying high-risk patients, allowing for personalized perioperative care adjustments, potentially mitigating POGD risks.

Construction and Validation of a Nomogram Clinical Prediction Model for Predicting Osteoporosis in an Asymptomatic Elderly Population in Beijing

BACKGROUND

Based on the high prevalence and occult-onset of osteoporosis, the development of novel early screening tools was imminent. Therefore, this study attempted to construct a nomogram clinical prediction model for predicting osteoporosis.

METHODS

Asymptomatic elderly residents in the training (n = 438) and validation groups (n = 146) were recruited. BMD examinations were performed and clinical data were collected for the participants. Logistic regression analyses were performed. A logistic nomogram clinical prediction model and an online dynamic nomogram clinical prediction model were constructed. The nomogram model was validated by means of ROC curves, calibration curves, DCA curves, and clinical impact curves.

RESULTS

The nomogram clinical prediction model constructed based on gender, education level, and body weight was well generalized and had moderate predictive value (AUC > 0.7), better calibration, and better clinical benefit. An online dynamic nomogram was constructed.

CONCLUSIONS

The nomogram clinical prediction model was easy to generalize, and could help family physicians and primary community healthcare institutions to better screen for osteoporosis in the general elderly population and achieve early detection and diagnosis of the disease.

Identifying Predictors of Psychological Problems Among Adolescents With Congenital Heart Disease for Referral to Psychological Care: A Pilot Study

Background

The lifelong care of patients with congenital heart disease (CHD) typically begins at a young age, giving paediatric cardiologists a unique perspective on the mental health of their patients. Our aim was to describe and predict reported psychological problems among adolescents with CHD.

Methods

A retrospective review was performed on patients aged 12-17 years who presented to the congenital cardiology clinic during a 1-year timeframe. The presence of psychological problems was collected along with CHD class, clinical history, developmental delay, and patient demographics. We described the prevalence of psychological problems and then, using machine learning algorithms, trained and tested optimal predictive models.

Results

Of the 397 patients who met inclusion criteria, the lifetime prevalence of any reported psychological problem was 35.5%. The most prevalent reported problems were attention-deficit/hyperactivity disorder (18.9%), anxiety (17.6%), and depression (16.1%). Contrary to our expectations, we could not predict the presence or absence of any psychological problem using routine clinical data. Instead, we found multivariate models predicting depression and attention-deficit/hyperactivity disorder with promising accuracy. Prediction of anxiety was less successful.

Conclusions

Approximately 1 of 3 adolescents with CHD presented with the lifetime prevalence of 1 or more psychological problems. Congenital cardiac programmes are in a position of influence to respond to these problems and impact their patients' mental health as part of a comprehensive care plan. The discovered models using routine clinical data predicted specific psychological problems with varying accuracy. With further validation, these models could become the tools of routine recommendations for referral to psychological care.