Risk Prediction Algorithm Research Articles

Incorporating end-user perspectives into the development of a machine learning algorithm for first time perinatal depression prediction.

Incorporating end-user perspectives into the development of a machine learning algorithm for first time perinatal depression prediction.

A Comparative Study of Machine Learning Algorithms for Flood Risk Prediction

Floods are one of the most frequent and devastating natural disasters, significantly affecting human life, infrastructure, and the environment. Accurate and timely flood prediction is essential for disaster preparedness and mitigation. This study presents an intelligent flood risk prediction framework using machine learning and deep learning techniques applied to four Indian cities—Pune, Nashik, Kolhapur, and Satara—each associated with major rivers such as the Mula-Mutha, Godavari, Panchganga, and Krishna, respectively. The dataset incorporates a comprehensive set of hydrometeorological and environmental features including rainfall, temperature, humidity, wind speed, water level, discharge, groundwater level, soil moisture, atmospheric pressure, evaporation rate, and historical flood events. Four algorithms—Random Forest, Support Vector Machine (SVM), XG-Boost, and Artificial Neural Networks (ANN)—were trained and evaluated to predict the flood_risk level. The model performances were compared using accuracy, precision, recall, F1-score, and ROC-AUC. The results demonstrate that the integration of multiple data sources and ensemble techniques significantly improves predictive performance. This research contributes to the development of smart, data-driven flood early warning systems tailored for regional hydrological conditions.

Multicenter Validation of a Machine Learning Model for Surgical Transfusion Risk at 45 US Hospitals

Accurate estimation of surgical transfusion risk is important for perioperative planning and effective resource allocation. Most machine learning models in health care are not validated or perform poorly in external settings. To externally validate a publicly available machine learning algorithm (Surgical Personalized Anticipation of Transfusion Hazard [S-PATH]) to estimate red cell transfusion during surgery within a national sample of hospitals. This retrospective cohort study evaluated all surgical cases performed in 2020 or 2021 at 45 US hospitals participating in the Multicenter Perioperative Outcomes Group. Obstetric and nonoperative cases were excluded. Data analysis was performed from February 2023 to March 2025. At each hospital, S-PATH was used to estimate surgical transfusion risk using patient- and procedure-specific characteristics without local retraining. A baseline model representing the standard-of-care maximum surgical blood ordering schedule (MSBOS) approach, which omits patient factors, was used for comparison. Risk thresholds above which a type and screen would be recommended were set for 96% sensitivity. Performance was evaluated at each hospital separately. The primary outcome was the difference in the percentage of patients with type and screen order recommendations between S-PATH and MSBOS at each hospital. The secondary outcome was area under the receiver operating characteristic curve (AUROC). In this cohort study of 3 275 956 surgical cases (median [IQR] age, 57 [40-69] years; 53.1% female) performed at 45 hospitals (28 of 45 academic [62.2%]), S-PATH recommended type and screen orders for a median (IQR) of 32.5% (25.8%-42.0%) of cases, whereas the MSBOS approach recommended type and screens for a median (IQR) of 51.6% (46.9%-61.1%) of cases for the same sensitivity (median [IQR] difference, 17.9 [14.8-24.9] absolute percentage points). The median (IQR) S-PATH AUROC was 0.929 (0.915-0.946), whereas the median (IQR) MSBOS AUROC was 0.857 (0.822-0.884). In this cohort study of 45 hospitals, a personalized surgical transfusion risk prediction algorithm demonstrated external validity and discrimination. S-PATH was consistently more effective than standard care, suggesting its potential for use as a perioperative clinical decision support tool.

Abstract P4-02-19: Development and Validation of a Genetic Risk Prediction Model for Breast Cancer: The Prognostic Role of SNAI1 and CDH1 in Treatment Response and Patient Survival

Abstract Background: Breast cancer is a highly heterogeneous disease characterized by varied molecular profiles, leading to diverse clinical treatment responses. Effective and specific classification and diagnosis are critical for optimizing breast cancer treatment. This study aims to establish a genetic risk prediction model for breast cancer patients, focusing on malignant manifestations and treatment response. Methods: This study extracted and transformed mRNA data from 27 target genes from 486 breast cancer patients into a data matrix. A hierarchical clustering-based risk prediction algorithm was developed to identify the most influential risk molecules and protective factors. The top 10 important genes identified were SNAI1, HDAC1, DNMT3A, WHSC1, PRDM15, PRMT2, CDH1, PRDM7, HDAC3, and PRMT8. Validation was further conducted using immunohistochemistry and scoring on samples from 74 patients. Results: The genetic risk prediction model identified SNAI1 as the most significant risk molecule and CDH1 as the primary protective factor. Pearson's analysis indicated a negative correlation between SNAI1 and CDH1. Kaplan-Meier analysis showed that SNAI1 is a poor prognostic indicator for overall survival in lymphatic invasion (LN+), chemotherapy alone (CT), chemotherapy plus radiotherapy (CT plus RT), and breast-conserving surgery (BCS). Univariate analysis highlighted SNAI1 as a prognostic indicator, while multivariable analysis confirmed that SNAI1 and CDH1 are significant independent prognostic indicators for overall survival in breast cancer after adjusting for clinical parameters. SNAI1 demonstrated excellent discrimination in CT, CT plus RT, and BCS subgroups and acceptable discrimination in overall and radical treatment groups. Conclusion: This study proposes a genetic risk prediction platform to identify potential risk factors. Clinical specimen validation showed that SNAI1 has a significant diagnostic effect on lymphatic invasion, patient survival, and chemotherapy prognosis in breast cancer patients. These findings provide a valuable biomarker prediction model for effective diagnosis, monitoring, and treatment. The ultimate goal is to prolong patient survival while avoiding over-treatment and unnecessary medical expenses. Citation Format: Chieh-Ni Kao, Chi-Wen Luo, Shu-Jyuan Chang, Sin-Hua Moi, Fang-Ming Chen, Ming-Feng Hou. Development and Validation of a Genetic Risk Prediction Model for Breast Cancer: The Prognostic Role of SNAI1 and CDH1 in Treatment Response and Patient Survival [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P4-02-19.

"Let's talk about risk": co-designing a pathway to assess, communicate and act on individual risk of long-term toxicities after breast cancer.

Survivors of breast cancer (BC) may endure significant and persistent post-treatment burdens that negatively impact overall quality of life. We previously developed risk prediction algorithms to identify individual patient profiles at increased risk for long-term toxicities. To prepare for the implementation of these risk algorithms in routine care, we performed a study to assess preferences, catalysts, and barriers concerning communication of individual risk of long-term BC toxicities. The goal was to co-design a pathway for risk assessment, communication, and management starting at diagnosis. A co-design study was performed using a participatory research framework and qualitative methods. Two phases of focus groups (FG) were conducted to assess the perspective of patients and providers through an iterative process of Exploration, Consultation, Prioritization, Integration and Co-design. Discussions were guided by four main questions: Who should communicate the risk? When should the risk be communicated? How should the risk be communicated? What information should be communicated, and care proposed?. FG discussions were recorded, pseudo-anonymized, transcribed and evaluated through a thematic content analysis. Results were reported following the consolidated criteria for reporting qualitative research (COREQ). Six FG were conducted between July 2022 and August 2023, with a total of 28 participants (8 patients and 20 providers). Results revealed a strong willingness to discuss the risk of long-term toxicities, particularly for patients who would present with a higher risk of toxicities. However, this willingness was contingent on the implementation of supportive care pathways that offer personalized communication strategies and risk mitigation approaches tailored to each patient's need. This study found that both patients and providers are interested in, and willing to engage in, the assessment, communication and mitigation of long-term toxicities from the time of diagnosis. To address this need in routine care, a tailored pathway was co-designed and will undergo formal testing in a hybrid Type 3 effectiveness/implementation clinical trial (NCT06479057). This study assessed the needs, preferences and expectations of patients and providers for implementing a care pathway to assess, communicate and mitigate the risk of long-term toxicities after breast cancer treatment using risk prediction algorithms.

Scalable AI-Enabled Healthcare Systems: Strategic Patient Segmentation and Clinical Intelligence

The growing demand for personalized, efficient, and data-driven healthcare has underscored the need for scalable artificial intelligence (AI) systems capable of supporting strategic decision-making in clinical settings. This study presents an integrated AI-enabled healthcare framework focused on strategic patient segmentation and real-time clinical intelligence. Utilizing electronic health records, streaming data from wearable devices, and clinical notes, the framework combines supervised and unsupervised machine learning algorithms for risk prediction and patient clustering. Gradient Boosting and Random Forest models demonstrated high predictive accuracy (AUC-ROC > 0.91), while K-Means clustering effectively segmented patients into clinically meaningful groups. Principal Component Analysis (PCA) and multivariate statistics confirmed the distinctiveness of patient cohorts in terms of age, comorbidity, readmission, and mortality. Additionally, a real-time clinical intelligence module, supported by Apache Kafka and Spark, delivered timely decision support alerts and was rated highly by clinicians for usability and usefulness. The findings validate the feasibility and impact of deploying scalable AI systems to enhance care precision, optimize resource allocation, and support proactive clinical interventions. This research contributes to the growing body of evidence advocating for responsible AI integration in healthcare, offering a blueprint for future implementations in diverse medical environments.

Machine Learning in Schizophrenia: A Systematic Review and Meta-Analysis of Diagnostic and Predictive Models

Aims: Schizophrenia is a psychiatric disorder characterized by diverse clinical presentations, posing challenges in early diagnosis and prognosis. Machine learning (ML) has emerged as a promising tool to enhance diagnostic accuracy, predict disease progression, and personalize treatment strategies. This systematic review and meta-analysis synthesized current evidence on the application of ML in schizophrenia diagnosis, prognosis, and treatment response prediction.Methods: A search was conducted across databases including PubMed, Embase, Scopus, Web of Science, and IEEE Xplore, adhering to PRISMA guidelines. Studies employing ML algorithms for schizophrenia classification, risk prediction, or treatment response modelling were included. Extracted data encompassed ML model types, sample sizes, data modalities (e.g., neuroimaging, clinical, genetic), and performance metrics such as accuracy, sensitivity, specificity, and area under the curve (AUC). A meta-analysis was performed to estimate pooled diagnostic performance, with heterogeneity assessed using I² statistics and publication bias evaluated via funnel plots and Egger’s test.Results: A total of 31 studies involving task-based functional MRI (t-fMRI) data were included in the meta-analysis. The pooled sensitivity and specificity for ML-based schizophrenia classification were both 0.83 (95% CI: 0.78–0.88), indicating a high level of diagnostic accuracy. Notably, studies focusing on selective attention tasks demonstrated higher specificity (0.86) compared with those assessing working memory tasks (0.79). Significant heterogeneity (I² = 72%) was observed, attributable to variations in neuropsychological domains, participant demographics, and clinical features.Conclusion: Machine learning exhibits substantial potential in improving schizophrenia diagnosis and outcome prediction, particularly when utilizing task-based neuroimaging data. However, challenges related to data heterogeneity, external validation, and clinical implementation persist. Future research should focus on standardizing ML methodologies, integrating multi-modal data, and enhancing model interpretability to facilitate translation into clinical psychiatry.

Development and validation of a model to predict future breast cancer risk after ER-positive and HER2-negative breast cancer.

10524 Background: Request for bilateral mastectomy after a unilateral breast cancer (BC) diagnosis is increasing. In many cases the benefit of bilateral mastectomy is likely to be small and offset by substantial risks of morbidity, financial toxicity and overburdening of healthcare systems. It is difficult to accurately determine personal risk of developing a future BC. Existing risk prediction models only predict risk for contralateral BC. Australian consumers identified an unmet need for a model that estimates risk of developing BC in any residual breast tissue (ipsi- or contra-laterally); to help women diagnosed with unilateral BC make informed decisions about bilateral mastectomy. Methods: Data from 1,162 female BC cases participating in two Australian cohort studies were used to develop a model to predict risk of BC for women who developed a 1st invasive ER positive, HER2 negative BC after cohort entry or within 2 years prior to cohort entry. Women with a germline pathogenic variant in a BC predisposition gene, and those who received neoadjuvant systemic therapy were excluded. 187 (88 ipsilateral, 96 contralateral, 3 unknown laterality) BC events (161 invasive and 26 DCIS) occurred over a median follow-up of 13.8 years. Flexible parametric survival analysis was used, with time since diagnosis as the time scale, and death due to any cause considered as a competing event. Potential predictors of future BC risk were investigated, including age at 1st BC, age at 1st birth, parity, breastfeeding duration, menopausal hormone therapy use, BMI, number of 1st-degree relatives with BC, BC polygenic risk score (PRS-313), contralateral mammographic density, surgery (breast conservation vs unilateral mastectomy), tumor grade and size, number of positive axillary nodes, associated LCIS, and use of adjuvant chemotherapy or radiation. Retained in the final risk prediction algorithm (all P < 0.05) were age at diagnosis of 1st BC, surgery type, radiation therapy, family history, and PRS-313. For external validation of the model, data from 3,136 cases (eligibility criteria as per the training set) with 181 subsequent BC events participating in the international Breast Cancer Association Consortium were used. Calibration and a time-dependent area under the curve (AUC) at 10 years were assessed to determine model performance. Sensitivity analysis excluding PRS-313 was also performed (as it is usually not available in clinical practice). Results: Discriminatory ability at 10 years was AUC = 0.66 (95% CI 0.62-0.70) or 0.65 (95% CI 0.61-0.69) if PRS-313 was excluded. The model was well calibrated; expected (176 cases) to observed (181 cases) ratio = 0.97 (95% CI 0.84-1.13). Conclusions: This model provides valid estimates of 10-year BC risk after a 1st ER-positive HER2-negative BC and may be useful in collaborative decision-making between patients and their surgeons when considering bilateral mastectomy.

P-555 The impact of maternal age, BMI, and fetal fraction on false-positive rates in prenatal aneuploidy testing: a nine-year analysis of 38,160 cases

Abstract Study question How do maternal age, BMI, and fetal fraction influence the accuracy of NIPT results, and what clinical implications arise from false-positive rates in specific populations? Summary answer Increased maternal age and obesity correlate with higher false-positive rates, particularly for Monosomy X, while lower fetal fraction significantly reduces the reliability of test results. What is known already Non-invasive prenatal testing (NIPT) based on next-generation sequencing (NGS) is a widely accepted method for detecting chromosomal aneuploidies such as Trisomy 21, Trisomy 18, and Monosomy X. Its high sensitivity and specificity are well-documented, but maternal factors like advanced age and elevated BMI may impact performance. Higher BMI often lowers fetal fraction, increasing false-positive results, while advanced maternal age raises the baseline aneuploidy risk. Although these associations are recognized, large-scale, longitudinal data with clinical outcomes remain scarce. This study examines the interplay of these factors using robust data from 38,160 cases to provide practical insights. Study design, size, duration This retrospective analysis included 38,160 NIPT results collected between 2016 and 2024 using consistent NGS technology. The study evaluated positive, confirmed positive, and false-positive rates stratified by maternal age (<35 vs. ≥35 years), BMI (normal, overweight, obese), and fetal fraction thresholds (<4%, ≥4%). Clinical outcomes of confirmed cases were reviewed to assess the direct impact of NIPT results on pregnancy management decisions. Participants/materials, setting, methods Samples from pregnant women were analyzed for common aneuploidies, including Trisomy 21, Trisomy 18, and Monosomy X, using NGS. Subgroup analyses were conducted based on maternal age, BMI, and fetal fraction. Clinical outcomes for confirmed positive cases were documented. Statistical analyses were performed to identify significant trends in false-positive and false-negative rates across the subgroups, providing insights into the influence of maternal demographic factors on test performance. Main results and the role of chance Among 38,160 NIPT results, the overall false-positive rate was 0.9%, with significant variability across subgroups. Women aged ≥35 years exhibited a higher false-positive rate for Trisomy 21 (1.5%) compared to those under 35 years (0.7%). Similarly, women with a BMI ≥30 showed an elevated false-positive rate for Monosomy X (2.2%) compared to women with normal BMI (0.8%). False-negative results were extremely rare (<0.03%), confirming the high sensitivity of NGS technology. Low fetal fraction (<4%) was identified as a critical factor contributing to false-positive results, particularly in obese women. Among all confirmed positive cases (n = 58), 100% resulted in pregnancy terminations following diagnostic confirmation, highlighting the clinical significance of NIPT results in guiding decision-making. Trisomy 18 demonstrated a 100% confirmation rate, while Monosomy X posed diagnostic challenges with the highest false-positive rate. These findings underscore the importance of integrating demographic factors into risk prediction algorithms to enhance test specificity. Personalized pre-test counseling and improvements in algorithmic sensitivity are recommended to address these disparities and optimize prenatal screening outcomes. Limitations, reasons for caution This study is retrospective and does not include follow-up on live births in negative cases or data on racial/ethnic differences. The findings are based on a single technology (NGS), and further multicenter prospective studies are needed to confirm the results and assess the generalizability to other populations and technologies. Wider implications of the findings Integrating demographic factors such as maternal age, BMI, and fetal fraction into NIPT algorithms could enhance diagnostic accuracy, reduce unnecessary interventions, and improve patient outcomes. Personalized counseling for high-risk groups is crucial to minimizing psychological stress and optimizing the clinical utility of prenatal screening. Trial registration number No

Advancing cardiovascular risk assessment: Real-time SCORES2 calculation through CDSS in primary care patients.

Advancing cardiovascular risk assessment: Real-time SCORES2 calculation through CDSS in primary care patients.

Integrated Care Coordination for Managing Chronic Conditions: Views of Health Staff on the Implementation of a Program Using an Algorithm to Identify People at Higher Risk of Hospitalisation in Sydney, Australia.

Integrated care interventions can improve patient outcomes and reduce the burden on acute health services, but need a strong evidence base to ensure their effectiveness. Understanding the meso and macro context in which care is delivered and determining whether patient needs are met are essential to successful implementation. Care coordination in New South Wales (NSW), Australia has evolved over time to meet the needs of an ageing population with chronic health conditions and multi-morbidity with the aim of reducing potentially preventable hospitalisations. To examine how an integrated care coordination program was understood and implemented at state, district and clinician levels in NSW. The Integrated Care for People with Chronic Conditions (ICPCC) program was implemented statewide, however local implementation varied. Patients who were suitable for integrated care coordination were identified via a hospitalisation risk prediction algorithm and/or referrals from health professionals. Understanding and implementation of ICPCC were assessed via interviews and a focus group with a range of health staff. Qualitative data were analysed using NVivo software and normalisation process theory. There was a strong sense of program coherence from management, clinicians and referrers. They viewed ICPCC as effective in coordinating care for patients at risk of hospitalisation and incorporating self-management at home. All health staff interviewed understood the program purpose and necessity, including the importance of achieving patient and systemic goals. Networking, linking services and program promotion were important, as was reporting on benefits. While the algorithm effectively identified previously hospitalised patients, it did not identify all suitable patients in the community with an increasing risk of requiring acute health care intervention. Referrals from health professionals familiar with patient needs and complexity were an important additional mechanism for patient selection. There was a shared sense of coherence and understanding of the ICPCC program among health staff at the three levels of implementation within NSW. The program played an important role in assisting patients with a range of chronic conditions to access and benefit from integrated care coordination, while increasing their capacity to self-manage at home. Program intake via hospitalisation risk prediction algorithm plus referrals from health professionals familiar with patient needs and complexity can effectively identify those who may benefit from integrated care coordination.

Maximizing Lung Cancer Screening in High-Risk Population Leveraging ML-Developed Risk-Prediction Algorithms: Danish Retrospective Validation of LungFlag.

Maximizing Lung Cancer Screening in High-Risk Population Leveraging ML-Developed Risk-Prediction Algorithms: Danish Retrospective Validation of LungFlag.

A prototype-based rockburst types and risk prediction algorithm considering intra-class variance and inter-class distance of microseismic data

The prediction and classification of rockburst risk based on microseismic data is the premise of preventing rockbursts during deep mine excavation. By reviewing previous studies, this paper finds two problems that hinder the rockburst prediction: 1) there is a lack of research on the distribution features of monitoring data on the main controlling factors of rockbursts; 2) there is no research on the intra-class variance and inter-class gap of microseismic data. Based on the typical rockburst risk events, a quantitative information model of geology and mining is constructed. The relationship between the spatial–temporal distribution characteristics of microseismic data before a rockburst and the main controlling factors of a rockburst is studied. The results show that the distribution features may be different for the same type of microseismic (MS) and rockburst events, and different types of events may show similar distribution features. Therefore, based on the quantitative study of the relationship between the performance of a deep learning prediction algorithm and a rockburst prediction vector, a rockburst risk and type prediction algorithm based on a convolutional neural network (CNN)-gated recurrent unit (GRU) model with prototype-based prediction is proposed. The CNN-GRU model can produce prediction vectors by fusing implicit and explicit information extracted from the original MS data and early warning indicators. Cross-entropy loss, vector-prototype contrastive loss, and vector-prototype contrastive loss are proposed to automatically control the intra-class variance and inter-class gap of prediction vectors belonging to different rockburst risks and types. Many experiments show that the performance of the proposed CNN-GRU model with prototype-based prediction is superior to other algorithms in the prediction of rockburst risks and types based on MS data.

Distribution of age at natural menopause, age at menarche, menstrual cycle length, height and BMI in BRCA1 and BRCA2 pathogenic variant carriers and non-carriers: results from EMBRACE

BackgroundCarriers of germline pathogenic variants (PVs) in the BRCA1 and BRCA2 genes are at higher risk of developing breast and ovarian cancer than the general population. It is unclear if these PVs influence other breast or ovarian cancer risk factors, including age at menopause (ANM), age at menarche (AAM), menstrual cycle length, BMI or height. There is a biological rationale for associations between BRCA1 and BRCA2 PVs and reproductive traits, for example involving DNA damage and repair mechanisms. The evidence for or against such associations is limited.MethodsWe used data on 3,046 BRCA1 and 3,264 BRCA2 PV carriers, and 2,857 non-carrier female relatives of PV carriers from the Epidemiological Study of Familial Breast Cancer (EMBRACE). Associations between ANM and PV carrier status was evaluated using linear regression models allowing for censoring. AAM, menstrual cycle length, BMI, and height in carriers and non-carriers were compared using linear and multinomial logistic regression. Analyses were adjusted for potential confounders, and weighted analyses carried out to account for non-random sampling with respect to cancer status.ResultsNo statistically significant difference in ANM between carriers and non-carriers was observed in analyses accounting for censoring. Linear regression effect sizes for ANM were -0.002 (95%CI: -0.401, 0.397) and -0.172 (95%CI: -0.531, 0.188), for BRCA1 and BRCA2 PV carriers respectively, compared with non-carrier women. The distributions of AAM, menstrual cycle length and BMI were similar between PV carriers and non-carriers, but BRCA1 PV carriers were slightly taller on average than non-carriers (0.5 cm difference, p = 0.003).ConclusionInformation on the distribution of cancer risk factors in PV carriers is needed for incorporating these factors into multifactorial cancer risk prediction algorithms. Contrary to previous reports, we found no evidence that BRCA1 or BRCA2 PV are associated with hormonal or anthropometric factors, except for a weak association with height. We highlight methodological considerations and data limitations inherent in studies aiming to address this question.

Reducing Readmission for Sepsis by Improving Risk Prediction Algorithms.

Reducing Readmission for Sepsis by Improving Risk Prediction Algorithms.

Prevalence of breast arterial calcification and its relationship with cardiovascular disease risk factors: insights from a cross-sectional study in South India

Background: Cardiovascular risk prediction algorithms often underestimate risk in women, resulting in underuse of preventive therapies and lifestyle interventions. Female-specific strategies may more accurately identify those at risk. As many women undergo routine mammography, identifying an association between breast arterial calcification (BAC) and cardiovascular disease (CVD) could improve risk stratification without added cost or radiation exposure. Mammography also offers an opportunity for counseling, given the shared lifestyle risk factors between breast cancer and CVD. Due to limited data on BAC prevalence and its relationship with CVD risk factors in Indian women, this study aims to bridge that gap. Methods: A cross-sectional study was conducted using retrospective hospital data from 286 women over 40 years who underwent routine mammograms at a tertiary care hospital. Data were analyzed using SPSS version 29. Appropriate statistical tests were applied based on data distribution. Results: The median age of participants was 54 years (IQR 48-60). BAC was present in 25% of women. Its prevalence was higher among those with cardiovascular risk factors such as high BMI, elevated triglycerides, low HDL, elevated blood pressure, increased HbA1c, and elevated fasting blood sugar. However, only the association with elevated fasting blood glucose was statistically significant. Conclusion: BAC may serve as a useful surrogate marker for identifying women at increased risk for CVD during routine mammography. Further large-scale, prospective studies are needed to establish its role in early detection and prevention strategies.

Leveraging Machine Learning for Lung Cancer Risk Assessment Based on Survey Insights

Lung cancer is still among the top cancers to cause cancer death in humans around the world. It has a lot to do with lifestyle and smoking- individual factors that contribute to lung cancer development. This research study seeks to analyze the viability of the machines through algorithms for the likely risk prediction of lung cancer through survey data- that is, symptoms, behavioral traits, and demographic data. The dataset consists of information such as smoking habits along with anxiety levels, fatigue, and other symptoms employed. Various machine learning models were trained and evaluated on Logistic Regression, Decision Tree, Random Forest, Naive Bayes, and Support Vector Machines (SVM) algorithms. Among those, Random Forest proved to be the best predictor giving about 96.7% accuracy and strong precision and recall values, indicating its effectiveness in identifying high-risk subjects. This research indicates that machine learning can be applied to healthcare for early diagnosis and screening.

Improving Management of Chest Pain with a High Sensitivity Troponin-Based Protocol.

Chest pain is one of the most common presenting complaints to emergency departments in the United States, and management centers on identifying myocardial infarction or other forms of rare but problematic cardiac diagnoses. The high-sensitivity troponin assay can detect abnormal troponin elevations at 10- to 100-fold lower levels compared with traditional troponin assays and thus can allow faster time to disposition and diagnosis, yet adoption has not been universal. Implementing a high-sensitivity troponin protocol with a risk prediction algorithm can decrease the numbers of patients admitted, reduce unnecessary testing, and shorten patient stays in the emergency department. This quality improvement project was undertaken in a community-academic health system lacking a system-wide protocol to workup patients presenting with chest pain to the emergency department. Key stakeholders evaluated multiple barriers and identified measures, planned implementation of the new assay and its associated algorithm, led postimplementation data monitoring and analysis, and delivered progress reports to organizational leaders. Chest pain admissions were managed by hospitalists in the absence of a cardiology inpatient service. The most important barriers were found to be individual provider strategy, electronic medical record design, and the lack of capacity for cardiology evaluations in both inpatient and outpatient settings. Stakeholder buy-in, monthly data reports, team meetings, and widespread education were used to support the changes in ordering patterns and evaluation. Postimplementation, 3293 patients were assessed over a 12-month period. Baseline mean length of stay for chest pain in the emergency department decreased from 297 minutes (SD, 53) to 274 minutes (SD, 33; P = 0.03). Hospital chest pain observation admissions decreased from 23% to 14% of patients presenting with chest pain ( P <0.001). Stress tests ordered for observation patients decreased from 12 per month to 3 ( P <0.001). Similarly, in observation patients, echocardiograms decreased from 61 to 46 per month ( P <0.001), cardiology consultation decreased from 125 per month to 81 ( P <0.001), and cardiac catheterization decreased from 41 per month to 32 following the intervention ( P = 0.003). Developing a standardized management protocol and selecting physician leaders to maintain and revise protocols were high-impact, low to moderate-effort interventions resulting in significant changes in practice. This study demonstrated that a high-sensitivity troponin assay, combined with a chest pain clinical management protocol based on the Heart, EKG, Age, Risk factor, Troponin score, was able to achieve a reduction in emergency department length of stay, a decrease in hospital observation admissions, and reduced cardiac testing in this patient population.

Exploring the extent of post-analytical errors, with a focus on transcription errors- an intervention within the VIPVIZA study.

We examined the magnitude of transcription errors in lipid variables in the VIPVIZA study and assessed whether education among the research personnel reduced the error frequency at follow-up. We also examined how the errors affected the SCORE2 risk prediction algorithm for cardiovascular disease, which includes lipid parameters, as this could lead to an incorrect treatment decision. The VIPVIZA study includes assessment of lipid parameters, where results for total cholesterol, triglycerides, HDL cholesterol, and calculated LDL cholesterol are transcribed into the research database by research nurses. Transcription errors were identified by recalculating LDL cholesterol, and a difference>0.15 indicated a transcription error in any of the four lipid parameters. To assess the presence of risk category misclassification, we compared the individual's SCORE2 risk category based on incorrect lipid levels to the SCORE2 categories based on the correct lipid levels. The transcription error frequency was 0.55 % in the 2019 VIPVIZA research database and halved after theeducational intervention to 0.25 % in 2023. Of the 39 individuals who had a transcription error in total or HDL cholesterol (with the possibility of affecting the SCORE2 risk category based on non-HDL cholesterol), six individuals (15 %) received an incorrect risk category due to the error. Transcription errors persist despite digitalisation improvements. It is essential to minimise transcriptions in fields outside the laboratory environment, as weobserved that critical decisions also rely on accurate information such as the SCORE2-risk algorithm, which is dependent on lab results but not necessarily reported by thelaboratory.

Application of K-Means and C4.5 Algorithms for dropout risk prediction in vocational high schools

The issue of dropout remains a serious challenge at the vocational high school (SMK) level, including at SMK Islam Nusantara. Over the past five years, the school has experienced a dropout rate of 20% of all enrolled students. This study aims to identify students at high risk of dropping out by applying the K-Means and C4.5 algorithms. The K-Means algorithm is used to cluster students based on factors such as academic performance, socioeconomic conditions, and the distance from home to school. Subsequently, the C4.5 algorithm is used to predict dropout risk based on the clustering results. The data used in this study were obtained directly from the Dapodik Data Center of SMK Islam Nusantara and include student data from the 2019 to 2024 academic years. The results indicate that socioeconomic factors, the distance from home to school, and academic performance significantly influence dropout risk. Students from low-income families and with poor academic performance were found to be at the highest risk of dropping out. This study makes a significant contribution to SMK Islam Nusantara by developing an early warning system that can help identify students at risk of dropping out. With more targeted interventions, such as academic counseling and socioeconomic support, it is expected that the dropout rate can be significantly reduced. Additionally, this predictive model can be applied to other vocational schools with similar conditions to improve the overall quality of vocational education.