Accurate Risk Prediction Research Articles

A novel clinical prediction model for hip fractures: a development and validation study in the total population of Sweden

Low bone density and osteoporosis are indications for bone-specific treatment. However, given the limited availability of bone density data in clinical practice and the fact that most patients with hip fracture do not have osteoporosis, accurate prediction of hip fracture risk in the absence of bone density data would be crucial. This development and validation study included the entire Swedish population aged 50 years or older in 2005 (N=3,340,977) and was conducted by cross-linking data from nationwide registers. Potential predictive variables included diagnoses, prescription medications, familial factors, frailty-related factors, and socioeconomic factors. The primary endpoint was the 5-year risk of hip fracture. Fracture prediction algorithms were developed and validated using multivariable models. Model performance and validation was also examined in a sub cohort restricted to 504,431 individuals with non-Swedish background. During a total follow-up of 15.2 million person-years, 87,089 individuals suffered a hip fracture within 5 years. In the final prediction model, 19 variables were associated with a population attributable fraction of 93.9% (95% CI, 93.7-94.1) in women and 92.7% (95% CI, 92.2-93.0) in men. The strongest predictor, besides old age, was the use of homemaker service, with a 5-year risk of hip fracture of 7.8% in women and 4.7% in men. The diagnoses most strongly predicting the 5-year risk of hip fracture was Parkinson's disease (6.8% in women, 4.6% in men) and dementia (6.1% in women, 3.6% in men). Validation of the prediction model suggested that the optimal threshold for treatment with bone-specific agents was an estimated 5-year hip fracture risk of 3%. Assuming a threshold of 3% and a 30% relative risk reduction from bone-specific treatment, the number needed to treat to prevent one hip fracture was estimated to 36 in women and 52 in men. Similar results were obtained in the sub cohort with non-Swedish background. A clinical prediction model developed and validated in the total Swedish population could predict the risk of hip fractures with high precision even in absence of data on bone density. The model was associated with a population attributable fraction for hip fracture of more than 90%, and the strongest predictor besides old age was the use of homemaker service, which likely reflect frailty. Based on the model, individuals with an estimated 5-year risk of hip fracture of at least 3% could be considered for bone-specific treatment. None.

Blood microRNA testing in participants with suspicious low-dose CT findings: follow-up of the BioMILD lung cancer screening trial

The proper management of suspicious radiologic findings is crucial to optimize the effectiveness of low-dose computed tomography (LDCT) lung cancer screening trials. In the BioMILD study, we evaluated the utility of combining a plasma 24-microRNA signature classifier (MSC) and LDCT to define the individual risk and personalize screening strategies. Here we aim to assess the utility of repeated MSC testing during annual screening rounds in 1024 participants with suspicious LDCT findings. The primary outcome was two-year lung cancer incidence in relation to MSC test results, reported as relative risk (RR) with 95% confidence interval (CI). Lung cancer incidence and mortality were estimated using extended Cox models for time-dependent covariates, yielding the respective hazard ratios (HR). Clinicaltrials.gov ID: NCT02247453. With a median follow-up of 8.5 years, the full study set included 1403 indeterminate LDCT (CTind) and 584 positive LDCT (CT+) results. A lung cancer RR increase in MSC+compared to MSC- participants was observed in both the CTind (RR: 2.5; 95% CI: 1.4-4.32) and CT+ (RR: 2.6; 95% CI: 1.81-3.74) groups and was maintained when considering stage I or resectable tumors only. A 98% negative predictive value in CTind/MSC- and a 30% positive predictive value in CT+/MSC+ lesions were recorded. At seven years' follow-up, MSC+ participants had a cumulative HR of 4.4 (95% CI: 3.0-6.4) for lung cancer incidence and of 8.1 (95% CI: 2.7-24.5) for lung cancer mortality. Our study shows that MSC can be reliably performed during LDCT screening rounds to increase the accuracy of lung cancer risk and mortality prediction and supports its clinical utility in the management of LDCT findings of uncertain malignancy. Italian Association for Cancer Research; Italian Ministry of Health; Horizon2020; National Cancer Institute (NCI); Gensignia LifeScience.

Urban Exposome and Health

Abstract Background The environment in which we live is thought to explain 70% of the non-communicable disease burden. The aim of EXPANSE project (funded from the European Union Horizon 2020 research and innovation programme under grant agreement No 874627) is to evaluate the optimal strategies for maintaining health in the context of contemporary urban environments. As a component of the project, Urban Labs allow deep phenotyping of the urban exposome and assessing its use and perception. The Urban Labs aim to improve exposure assessment, identify mediating and modifying lifestyle factors, and link the urban exposome to intermediate endpoints relevant for cardiometabolic and pulmonary health. Methods By monitoring the activities of 4,000 individuals in five distinct urban areas, Urban Labs provide a comprehensive understanding of the urban exposome, which can be compared with residential address-based model estimates. The participants complete a baseline questionnaire, repeated bi-monthly questionnaires and exposure-specific questionnaires over a period of 2 years using mobile phone application (ExpanSeeker). Two-week personalized measurements including spatial-temporal activity patterns, sensor-based environmental and behavioral measurements, and minimally invasive biological sample collection are performed covering different seasons. Results The Baseline Questionnaire (BQ) and two-weeks measurement campaigns (MC) were completed in European cities as follows: Athens BQ = 379, MC = 232; Barcelona BQ = 1040, MC = 303; Basel BQ = 731, MC = 655; Lodz BQ = 634, MC = 100; The Netherlands (multicity) BQ = 1033, MC = 326. Data collection process is ongoing. Conclusions The collection of novel data on individuals activities and perceptions within temporal and geographical contexts in Urban Labs will contribute to a more comprehensive understanding of how the urban dwellers are exposed to and interact with their environment, thereby supporting healthier urban engineering and policy making. Key messages • Urban Labs will provide a more comprehensive understanding of the impact of the urban environment on individual’s behavior, their exposome, enabling more accurate prediction of non-communicable risk. • Urban Labs will provide a base to support healthier urban engineering and policy making.

Machine learning for prediction of all-cause mortality in acute coronary syndrome

Abstract Background Acute coronary syndrome (ACS) remains one of the leading causes of death globally. Accurate and reliable mortality risk prediction of ACS patients is essential for developing targeted treatment strategies and improve prognostication. Traditional models for risk stratification such as the Global Registry of Acute Coronary Events (GRACE) and the Thrombolysis In Myocardial Infarction (TIMI) risk scores offer moderate discriminative value, and do not incorporate contemporary predictors of ACS prognosis. Machine learning (ML) models have emerged as an alternate method that may offer improved risk assessment. This study aims to compare machine learning models with traditional risk scores for predicting all-cause mortality in patients with ACS. Methods PubMed, EMBASE, Web of Science and Cochrane databases were searched until 1st November 2023 for studies comparing ML models with traditional statistical methods for event prediction of ACS patients. The primary outcome was comparative discrimination measured by C-statistics with 95% confidence intervals between ML models and traditional methods in estimating the risk of all-cause mortality. Results Ten studies were included (239627 patients). The summary C-statistic of all ML models across all endpoints was 0.89 (95% CI, 0.86-0.92), compared to traditional methods 0.82 (95% CI, 0.79-0.85). The difference in C-statistic between all ML models and traditional methods was 0.07 (p<0.05). Three models undertook external validation, and calibration was inconsistently reported. Conclusion ML models demonstrated superior discrimination of all-cause mortality for ACS patients compared to traditional risk scores. Despite outperforming well-established prognostic tools such as the GRACE and TIMI scores, current clinical applications of ML approaches remain uncertain, particularly in view of the methodological limitations of existing models and the need for greater validation.

Predicting progression from coronary artery disease to heart failure in community cohorts

Abstract Background Approximately 10% of people develop heart failure (HF) within 5 years after an acute coronary event. Once ischaemic damage has occurred, the heart undergoes remodeling, which can lead to subclinical dysfunction and HF. However, remodeling varies considerably between patients and is difficult to predict. Purpose To assess whether adding genetic information to clinical factors improves the accuracy of predicting incident non-fatal/fatal HF within 5 years in people with coronary artery disease (CAD). Methods In the UK Biobank (UKB) [1], we developed a clinical risk model (predictors age, sex, ethnicity, body mass index, social deprivation, smoking, personal medical history, time since most recent CAD admission, lipids, HbA1C, creatinine, medications). Separately, we performed a genome-wide association meta-analysis of 13,360 (6,315 + 7,045) cases and 58,126 (37,245 + 20,881) non-cases from the UKB and deCODE [2] cohorts, relating common genetic variants to non-fatal/fatal HF or cardiomyopathy in people with CAD. For the current analysis of the UKB cohort, we derived a polygenic risk score using summary statistics from deCODE alone. We then compared risk estimates and assessed reclassification of 5-year risk using the clinical risk model with and without polygenic risk score, in UKB. Risk groups were <5%, 5-9.9%, 10-14.9%, ≥15%. Results Among people with CAD in the 5 years prior to assessment in UKB (n=8,880, mean age 61y, 27% women) or deCODE (n=10,113, mean age 66y, 34% women), 377 (4%) and 915 (9%) developed non-fatal/fatal HF over 5 years follow-up, respectively. In the UKB cohort, median 5-year risk with the clinical model was 3.2% (IQR 2-5.4%, C-statistic 0.72, 95% CI 0.70–0.74, Table 1) and plots of predicted versus observed risk showed very good calibration across deciles. The polygenic risk score was predictive of non-fatal/fatal HF in the UKB (hazard ratio 1.03, 95% CI 1.00 to 1.05 per standard deviation) with significance decreasing after adjusting for the prognostic index of the clinical risk model (hazard ratio 1.02, 95% CI 0.99 to 1.05 per standard deviation). Median 5-year risk by the clinical+polygenic model was 3.0% (IQR 1.9-5.0%). Net reclassification with the addition of the polygenic score improved accuracy of risk prediction for 3.6% of non-cases but deteriorated accuracy of risk prediction for 6% of cases. Overall net reclassification was deterioration of -0.025. Conclusion Our data suggest that common genetic variants may be associated with progression from CAD to HF, but that adding genetic data did not improve the accuracy of predicting progression from CAD to HF beyond established clinical predictors in the UKB cohort. Confirmation of these findings in deCODE and other large cohorts is warranted.

ASCVD risk scores versus a novel cancer-specific Machine Learning-based calculator among patients with breast, colorectal, lung or prostate cancer

Abstract Background Patients with cancer have up to a 3-fold higher risk for cardiovascular disease (CVD) than the general population. Consequently, traditional risk scores developed to predict CVD in the general population (i.e., Pooled Cohort Equations (PCE), Predicting Risk of cardiovascular disease EVENTs [PREVENT]) may provide less accurate risk prediction for this population. Purpose Compare cancer-specific Machine Learning (ML)-based score to PCE and PREVENT to predict 10-year CVD risk in patients with breast cancer (BC), colorectal cancer (CRC), lung cancer (LC), or prostate cancer (PC). Methods Patients aged ≥18 years, diagnosed with BC, CRC, LC, or PC between 2005-2012 at a large hybrid academic-community practice in Northeast Ohio, United States, were included. An ML Extreme Gradient Boosting (XGBoost) algorithm, adapted for survival modeling, was developed using a training subset of the cohort (55% train + 20% test) and ranked 40 to 50 covariates (including social determinants of health, cancer treatment) for predicting 10-year CVD risk for each cancer type using SHAP (SHapley Additive exPlanations) values. The top 10 ML predictors were utilized to create a predictive equation using logistic regression models. This equation was then tested in the cohort validation subset (25%) and subsequently compared to the PCE and PREVENT (both simple and enhanced versions) using the area under the curve (AUC) of the time-dependent receiver operating characteristic curve. The CVD events were identified using ICD codes corresponding to Acute Coronary Syndrome, Ischemic Stroke, and Heart Failure. CVD risk was assessed through Kaplan-Meier. Results We included 10,240 patients (Table 1). The actual 10-year CVD events were: BC 21% (95% CI 19-23), CRC 10% (95% CI 8-12), LC 27% (95% CI 24-30), and PC 20% (95% CI 17-23). The ACC/AHA PCE predicted mean 10-year CVD risks of 16% (95% CI 15-18) for BC, 19% (95% CI 17-21) for CRC, 28% (95% CI 25-32) for LC, and 19% (95% CI 19-20) for PC, with corresponding AUCs of 0.75, 0.65, 0.76, and 0.61, respectively. The PREVENT simple version predicted mean 10-year CVD risks of 13% (95% CI 13-14) for BC, 13% (95% CI 12-13) for CRC, 15% (95% CI 14-15) for LC, and 16% (95% CI 12-19) for PC, with AUCs of 0.55, 0.55, 0.55, and 0.53, respectively. The PREVENT enhanced equation predicted mean 10-year CVD risks of 17% (95% CI 17-18) for BC, 15% (95% CI 14-15) for CRC, 15% (95% CI 15-16) for LC, and 17% (95% 16-19) for PC, with AUCs of 0.75, 0.62, 0.79, and 0.63, respectively. The predictive equation derived from the top 10 ML predictors (Table 2) achieved AUCs of 0.84 for BC, 0.76 for CRC, 0.83 for LC, and 0.71 for PC. Conclusion(s) Conventional CVD equations inadequately assess the risk in patients with BC, CRC, LC, or PC, often resulting in underestimation or overestimation. Cancer-specific ML-derived equations show good performance and emphasize the importance of integrating cancer-related covariates for precise prediction.Table 1Table 2

The role of coronary artery calcium score and european score2 in primary prevention of cardiovascular disease

Abstract Background There is limited evidence about the role of using coronary artery calcium (CAC) score in conjunction with the European SCORE2 (systematic coronary risk evaluation) for the identification of individuals at higher risk of cardiovascular disease. Purpose To evaluate the relationship between CAC score and the European SCORE2 for cardiovascular risk stratification in an asymptomatic population in primary prevention. Methods This is a preliminary analysis from a single-centre, prospective, randomised 1:1, controlled trial assessing the use of CAC score vs standard preventive strategies in asymptomatic individuals undergoing primary cardiovascular prevention. The 10-year risk of cardiovascular events was assessed with either SCORE2 (ages 45-69) or SCORE2-OP (ages 70-89). Subjects were classified as 1) low to moderate risk, 2) high risk, and 3) very high risk according to SCORE2/SCORE2-OP. CAC score was assessed by ECG gated low-dose computed tomography with no contrast administration. A CAC score of zero was considered to reclassify risk one level downward in high or very-high risk patients. CAC score ≥ 100 was used to reclassify risk one level upward in patients with low to moderate risk and a CAC score > 1000 to reclassify risk upward in subjects with high risk. Results A total of 1,252 participants: 627 in the CAC group and 625 in the control group, median age 55 yrs, 25th-75th 50-61 yrs, with availability of SCORE2 and SCORE2-OP were included in this analysis. SCORE2/SCORE 2-OP was similar in the CAC and control groups. Among the subjects in the CAC group, there was a highly significant difference in SCORE2/SCORE 2-OP between participants with CAC score = zero and those with CAC > zero (median 5.6, 25th-75th 3.6-8.55 vs 2.9, 1.8-4.6, p< 0.001) as well as between those with CAC ≥ 100 and those with CAC score < 100 [median 6.8, 25th-75th (10.3-4.5) vs 3.3 (2.0-5.3), p<0.001]. The prevalence of CAC score = zero was 75% in the low to moderate risk group, 47% in the high risk, and 20% in the very high risk, p<0.001. By contrast the prevalence of CAC score ≥ 100 was 7% in the low to moderate risk group, 24% in the high risk, and 45% in the very high risk, p< 0.001. The proportion of patients classified as low to moderate risk group was 72% with CAC score vs 59% without CAC score, high risk 23% with CAC score vs 35% without CAC score, very high risk 5% vs 6% respectively, p<0.001. Using CAC score risk reclassification occurred in 48% downward and 2% upward in originally classified high risk subjects, 7% upward in originally classified low to moderate risk, and 20% downward in originally classified very high risk subjects. Conclusions CAC score may lead to risk reclassification in a substantial proportion of individuals, in particular among those originally classified as high risk by the SCORE2/SCORE2-OP. The cardiovascular risk prediction accuracy of adding CAC score to SCORE2/SCORE2-OP awaits validation at long-term follow-up.

Neonatal multimorbidity and the phenotype of premature aging in preterm infants.

Multimorbidity is the co-occurrence of multiple chronic health problems, associated with aging, frailty, and poor functioning. Children born preterm experience more multimorbid conditions in early life compared to term-born peers. Though neonatal multimorbidity is linked to poor health-related quality of life, functional outcomes, and peer group participation, gaps in our theoretical understanding and conceptualization remain. Drawing from life course epidemiology and the Developmental Origins of Heath and Disease models, we offer a framework that neonatal multimorbidity reflects maturational vulnerability posed by preterm birth. The impact of such vulnerability on health and development may be further amplified by adverse exposures and interventions within the environment of the neonatal intensive care unit. This can be exacerbated by disadvantaged home or community contexts after discharge. Uncovering the physiologic and social antecedents of multiple morbid conditions in the neonatal period and their biological underpinnings will allow for more accurate risk-prediction, counseling, and care planning for preterm infants and their families. According to this framework, the maturational vulnerability to multimorbidity imparted by preterm birth and its negative effects on health and development are not predetermined or static. Elucidating pathways of early biologic and physical aging will lead to improvements in care and outcomes. IMPACT: Multimorbidity is associated with significant frailty and dysfunction among older adults and is indicative of early physiologic aging. Preterm infants commonly experience multimorbidities in the newborn period, an underrecognized threat to long-term health and development. We offer a novel framework incorporating multimorbidity, early cellular aging, and life course health development to innovate risk-prediction, care-planning, and therapeutics.

Radiomic and clinical model for predicting atypical ductal hyperplasia upgrades and potentially reduce unnecessary surgical treatments

ObjectiveTo identify patients with atypical ductal hyperplasia (ADH) at low risk of upgrading to carcinoma. This study aims to assess the performance of radiomics combined with clinical factors to predict occult breast cancer among women diagnosed with ADH. MethodsThis study retrospectively included microcalcification clusters of patients who underwent Mx and VABB with a diagnosis of ADH at a tertiary center from January 2015 to May 2023. Clinical and radiological data (age, cluster size, BI-RADS classification, mammographic density, breast cancer history, residual microcalcifications) were collected. Surgical outcomes were used to determine upgrade. Four logistic regression models were developed to predict the risk of upgrade. The performance was evaluated using the area under the receiver operating characteristic curve (AUC) and performance scores. ResultsA total of 143 patients with 153 clusters were included. Twelve radiomic features and six clinical factors were selected for model development. The sample was split into 107 training and 46 test cases. Clinical features achieved an AUC of 0.72 (0.60–0.84), radiomic features an AUC of 0.73 (0.61–0.85). Radiomic features with “cluster size” and “age” improved the AUC to 0.79 (0.67–0.91). The best model, incorporating all data, achieved an AUC of 0.82 (0.71–0.92), a specificity of 0.89 (0.75, 0.97), and NPV of 0.92 (0.78–0.98). ConclusionThis study demonstrates the potential of radiomic as a valuable tool for reducing unnecessary treatments for patient classified as “low risk of ADH upgrade”. Combining radiomic information with clinical data improved the accuracy of risk prediction.

Systemic Risks in the Derivatives Market: Origins, Impacts, and Mitigation Strategies

Derivative markets play an indispensable role in the contemporary financial system. The global financial markets have been repeatedly riddled with turbulence instigated by derivative trading in recent years, underscoring the urgency and necessity for in-depth study in this field. The subject matter of this paper is to delve into systemic risk in the derivative markets, including its genesis, impact, and mitigation strategies. This paper aims to proffer valuable insights and recommendations to policymakers and financial practitioners through comprehensive analysis. To achieve this objective, this study utilizes research methodologies such as literature review, analysis of historical cases, and empirical analysis. The findings suggest that the systemic risk in the derivative markets primarily stems from high leverage operations, market information asymmetry, and complex financial product structures. These risks became particularly prominent during the financial crisis, leading to an upsurge in market volatility, liquidity crises, and a domino collapse of financial institutions. Based on these findings, the paper proposes a gamut of strategies to mitigate systemic risk, including enhancing market transparency, optimizing risk models, augmenting capital requirements, and fostering international regulatory cooperation. Additionally, technological innovations such as blockchain and big data analysis are considered to hold significant potential in enhancing market transparency and improving the accuracy of risk prediction.

A comparative analysis of three frailty assessment tools for hospitalized patients with stroke

ObjectiveThe aim of this study is to assess the effectiveness of three frailty assessment tools in determining frailty risk among hospitalized patients with stroke and to offer a reference framework for selecting appropriate clinical frailty assessment tools in stroke management. MethodsA group of 203 hospitalized patients who had stroke were selected through convenience sampling and assessed for frailty using the Frailty Index, Fried Frailty Phenotype, FRAIL Scale, and Tilburg Frailty Scale. The efficacy of the three frailty assessment tools in assessing frailty risk in hospitalized patients with stroke was compared via Bayes discrimination and ROC curve analysis by using the Frailty Index as the diagnostic criterion for stroke-related frailty. ResultsThe incidence of frailty among patients with stroke ranged from 21.2 % to 23.6 %. The Kappa values indicating the agreement between the Frailty Index and Fried’s Frailty Phenotype, FRAIL Scale, and Tilburg Frailty Scale were 0.826, 0.928, and 0.707, respectively (all P < 0.01). The cross-validation accuracy for frailty risk prediction in patients with stroke was 94.1 %, 97.5 %, and 89.7 %, respectively. The areas under the ROC curves for these tools were 0.884, 0.955, and 0.896, respectively. ConclusionThe effectiveness of the three assessment tools in assessing frailty risk in patients with stroke ranked from highest to lowest, was as follows: FRAIL Scale, Fried Frailty Phenotype, and Tilburg Frailty Scale. Considering both assessment efficacy and convenience, the FRAIL Scale is recommended for widespread use in frailty screening among hospitalized patients with stroke.

Construction and validation of a nomogram model for lymph node metastasis of stage II-III gastric cancer based on machine learning algorithms.

Gastric cancer, a pervasive malignancy globally, often presents with regional lymph node metastasis (LNM), profoundly impacting prognosis and treatment options. Existing clinical methods for determining the presence of LNM are not precise enough, necessitating the development of an accurate risk prediction model. Our primary objective was to employ machine learning algorithms to identify risk factors for LNM and establish a precise prediction model for stage II-III gastric cancer. A study was conducted at Renji Hospital Affiliated to Shanghai Jiao Tong University School of Medicine between May 2010 and December 2022. This retrospective study analyzed 1147 surgeries for gastric cancer and explored the clinicopathological differences between LNM and non-LNM cohorts. Utilizing univariate logistic regression and two machine learning methodologies-Least absolute shrinkage and selection operator (LASSO) and random forest (RF)-we identified vascular invasion, maximum tumor diameter, percentage of monocytes, hematocrit (HCT), and lymphocyte-monocyte ratio (LMR) as salient factors and consolidated them into a nomogram model. The area under the receiver operating characteristic (ROC) curve (AUC), calibration curves, and decision curves were used to evaluate the test efficacy of the nomogram. Shapley Additive Explanation (SHAP) values were utilized to illustrate the predictive impact of each feature on the model's output. Significant differences in tumor characteristics were discerned between LNM and non-LNM cohorts through appropriate statistical methods. A nomogram, incorporating vascular invasion, maximum tumor diameter, percentage of monocytes, HCT, and LMR, was developed and exhibited satisfactory predictive capabilities with an AUC of 0.787 (95% CI: 0.749-0.824) in the training set and 0.753 (95% CI: 0.694-0.812) in the validation set. Calibration curves and decision curves affirmed the nomogram's predictive accuracy. In conclusion, leveraging machine learning algorithms, we devised a nomogram for precise LNM risk prognostication in stage II-III gastric cancer, offering a valuable tool for tailored risk assessment in clinical decision-making.

Optimization of Enterprise Financial Risk Management and Crisis Early Warning System Supported by AI

The stability of the financial market plays a key role in the overall health of the economic system, and optimizing corporate financial risk management and crisis warning systems is of great significance. Traditional financial risk management methods often face challenges when processing large-scale, resulting in a decrease in the accuracy of risk assessment. This study combining gated recurrent units (GRU), temporal convolutional networks (TCN) and attention mechanisms. Leverage diverse data sets to improve the accuracy of financial risk assessments and the effectiveness of crisis warnings. Experimental results show that the algorithm outperforms the baseline model in terms of risk prediction accuracy, efficiency and stability. For example, on the LendingClub loan dataset, FLOPs are reduced by more than 46.8%, inference time is improved by more than 48.5%. This method provides new perspectives and solutions for optimizing corporate financial risk management and crisis warning systems. This will have a positive impact on the stable operation of enterprises and other aspects.

Interpretable subgroup learning-based modeling framework: Study of diabetic kidney disease prediction.

Complex diseases, like diabetic kidney disease (DKD), often exhibit heterogeneity, challenging accurate risk prediction with machine learning. Traditional global models ignore patient differences, and subgroup learning lacks interpretability and predictive efficiency. This study introduces the Interpretable Subgroup Learning-based Modeling (iSLIM) framework to address these issues. iSLIM integrates expert knowledge with a tree-based recursive partitioning approach to identify DKD subgroups within an EHR dataset of 11,559 patients. It then constructs separate models for each subgroup, enhancing predictive accuracy while preserving interpretability. Five clinically relevant subgroups are identified, achieving an average sensitivity of 0.8074, outperforming a single global model by 0.1104. Post hoc analyses provide pathological and biological evidence supporting subgroup validity and potential DKD risk factors. The iSLIM surpasses traditional global model in predictive performance and subgroup-specific risk factor interpretation, enhancing the understanding of DKD's heterogeneous mechanisms and potentially increasing the adoption of machine learning models in clinical decision-making.

A Longitudinal Model for a Dynamic Risk Score to Predict Delayed Cerebral Ischemia after Subarachnoid Hemorrhage.

Accurate longitudinal risk prediction for DCI (delayed cerebral ischemia) occurrence after subarachnoid hemorrhage (SAH) is essential for clinicians to administer appropriate and timely diagnostics, thereby improving treatment planning and outcome. This study aimed to develop an improved longitudinal DCI prediction model and evaluate its performance in predicting DCI between day 4 and 14 after aneurysm rupture. Two DCI classification models were trained: (1) a static model based on routinely collected demographics and SAH grading scores and (2) a dynamic model based on results from laboratory and blood gas analysis anchored at the time of DCI. A combined model was derived from these two using a voting approach. Multiple classifiers, including Logistic Regression, Support Vector Machines, Random Forests, Histogram-based Gradient Boosting, and Extremely Randomized Trees, were evaluated through cross-validation using anchored data. A leave-one-out simulation was then performed on the best-performing models to evaluate their longitudinal performance using time-dependent Receiver Operating Characteristic (ROC) analysis. The training dataset included 218 patients, with 89 of them developing DCI (41%). In the anchored ROC analysis, the combined model achieved a ROC AUC of 0.73 ± 0.05 in predicting DCI onset, the static and the dynamic model achieved a ROC AUC of 0.69 ± 0.08 and 0.66 ± 0.08, respectively. In the leave-one-out simulation experiments, the dynamic and voting model showed a highly dynamic risk score (intra-patient score range was 0.25 [0.24, 0.49] and 0.17 [0.12, 0.25] for the dynamic and the voting model, respectively, for DCI occurrence over the course of disease. In the time-dependent ROC analysis, the dynamic model performed best until day 5.4, and afterwards the voting model showed the best performance. A machine learning model for longitudinal DCI risk assessment was developed comprising a static and a dynamic sub-model. The longitudinal performance evaluation highlighted substantial time dependence in model performance, underscoring the need for a longitudinal assessment of prediction models in intensive care settings. Moreover, clinicians need to be aware of these performance variations when performing a risk assessment and weight the different model outputs correspondingly.

Development of decision tree classification algorithms in predicting mortality of COVID-19 patients

IntroductionThe accurate prediction of COVID-19 mortality risk, considering influencing factors, is crucial in guiding effective public policies to alleviate the strain on the healthcare system. As such, this study aimed to assess the efficacy of decision tree algorithms (CART, C5.0, and CHAID) in predicting COVID-19 mortality risk and compare their performance with that of the logistic model.MethodsThis retrospective cohort study examined 5080 cases of COVID-19 in Babol, a city in northern Iran, who tested positive for the virus via PCR from March 2020 to March 2022. In order to check the validity of the findings, the data was randomly divided into an 80% training set and a 20% testing set. The prediction models, such as Logistic regression models and decision tree algorithms, were trained on the 80% training data and tested on the 20% testing data. The accuracy of these methods for the test samples was assessed using measures like ROC curve, sensitivity, specificity, and AUC.ResultsThe findings revealed that the mortality rate for COVID-19 patients who were admitted to hospitals was 7.7%. Through cross validation, it was determined that the CHAID algorithm outperformed other decision tree and logistic regression algorithms in specificity, and precision but not sensitivity in predicting the risk of COVID-19 mortality. The CHAID algorithm demonstrated a specificity, precision, accuracy, and F-score of 0.98, 0.70, 0.95, and 0.52 respectively. All models indicated that factors such as ICU hospitalization, intubation, age, kidney disease, BUN, CRP, WBC, NLR, O2 sat, and hemoglobin were among the factors that influenced the mortality rate of COVID-19 patients.ConclusionsThe CART and C5.0 models had outperformed in sensitivity but CHAID demonstrates a better performance compared to other decision tree algorithms in specificity, precision, accuracy and shows a slight improvement over the logistic regression method in predicting the risk of COVID-19 mortality in the population under study.

Environment-compatible heavy metal risk prediction method created with multilevel ensemble learning

Accurate health risk prediction (HRP) is an effective means of reducing the hazards of heavy metal (HM) exposure. It can address the drawbacks of lag and passivity faced by health risk assessment. This study innovatively proposed an HRP method, MEL-HR, based on multilevel ensemble learning (MEL) technology and environment compatibility. We conducted point and interval prediction experiments on health risks using 490 sets of data covering 17 environment factors. The point prediction results indicated that when the model predicts HI and TCR, the R2 values were 0.707 and 0.619, respectively. For P5, P50, and P95 in interval prediction, the R2 values of the model were 0.706, 0.703, and 0.672 for HI, and that for TCR were 0.620, 0.607, and 0.616, respectively. The analysis of feature importance indicated that, in addition to HM factors, longitude, mining area coefficient, and soil organic matter were key environmental factors affecting the MEL-HR model. Comparative experiments showed that compared to soil HMs-based MEL-HR, environment compatibility-based MEL-HR has improved the accuracy for HI and TCR by 19.83 % and 40.36 % for the point prediction and 22.06 % and 40.01 % for interval prediction. This study can provide technical support for targeted and resilient prevention and control of health risks.

Estimating casualties from urban fires: A focus on building and urban environment information

This study developed two prediction models for urban fire occurrence and related casualties via a fire accident dataset from Seoul, South Korea, from 2017 to 2021. Our models exhibit improved predictive performance by incorporating built environment features, such as building characteristics and the urban context, alongside weather and demographic data. This approach showed improved predictive performance suitable for public health implementation. Compared with the weather- and demographic-only models, our models had an 18.1 % greater fire occurrence prediction accuracy and a 10.4 % greater casualty prediction accuracy. Major variables affecting fire occurrence include building characteristics, e.g., the floor area ratio (FAR), building age, and commercial building number. Important features affecting casualty occurrence include demographic aspects, e.g., income level and weather, and network-based features, e.g., road connectivity and fire station proximity. These findings suggest that fire prevention strategies and fire casualty prevention strategies may need to differ. Furthermore, we identify high-risk zones by conducting spatial analysis and fire risk and casualty prediction on all buildings by applying our models to Seoul's Gangnam District. These contributions can promote safe and healthy urban environments by improving fire risk prediction accuracy and providing important insights into urban planning for appropriate urban fire accident response and prevention.

Abstract A103: Feasibility of using data from the All of Us Research Program to better understand Asian American breast cancer risk

Abstract Asian Americans (AAs) are the only racial/ethnic group in the United States for whom cancer is the leading cause of death. In AA women, the rate of breast cancer (BC) has significantly increased in the past two decades. However, the reasons behind the observed increase in risk are not well understood. The All of Us (AoU) Research Program aims to enroll one million participants nationwide, focusing on including individuals from diverse backgrounds, to advance precision health research. The goal of this pilot study was to evaluate the feasibility of using AoU data to analyze factors potentially associated with breast cancer risk in AAs and to develop enhanced breast cancer prediction models. There were 3,433 self-reported AA women with available genomic data. In addition to the genomic data, data was available from surveys, participants’ electronic health records (EHRs), and physical measurements taken at the time of their study visit. BC cases were identified by self-report on surveys (n=38), EHR diagnosis data (n=104), or both survey and EHR (n=59), totaling 201 cases identified. Of these, only 17 were incident cases. Of the 163 cases detected in the EHR, 96.3% were invasive BCs, and the median age at diagnosis was 54.2 (±12.6) years. Among the 38 survey-only-identified cases, age at diagnosis was unclear due to the large age ranges (e.g.18-64 years) in the categories presented in the survey. For purposes of the case-control analysis, five AA women with no BC diagnosis were matched to each BC case by age group and geographic region. Body mass index, annual household income, marital status, history of alcohol consumption, and history of cigarette smoking and vaping were not statistically significantly different between cases and controls (a=0.05). One third (34.3%) of cases were born in the U.S. and 65.2% were born in another country vs. 25.0% and 74.0% of controls, respectively (p= 0.01). Employment status was also different, with 50.8% of cases being unemployed vs. 42.0% of controls (p=0.02). There appeared to be a difference in first-degree family history of BC (reported in 8% of cases vs. 5% of controls) although the results were not statistically significant (p=0.08). However, family history of BC data was not available for 43.3% and 51.0% of the cases and controls, respectively. Gynecologic factors, such as ages at menarche and menopause, and hormone use, were not asked in the surveys and were not widely captured in the EHR data. In our analysis of nine BC- associated genes (BRCA1, BRCA2, ATM, TP53, PTEN, CHEK2, CDH1, PALB2, STK11), pathogenic variants were identified in 1.0% of cases and 0.7% of controls. In summary, the AoU data set enables the study of a diverse national cohort with rich genomic data. However, the low representation of AAs, small numbers of incident cases, limited data on personal gynecologic history, cancer family history and social determinants of health, and broad categorization of age at diagnosis present challenges to using this data to build accurate and effective risk prediction models for BC risk in AAs. Citation Format: Catalina Haas Garcia, Wen-Pin Chen, Danni Liu, Alice W Lee, Min Zhang, Argyrios Ziogas, Sora Tanjasiri, Hannah Lui Park. Feasibility of using data from the All of Us Research Program to better understand Asian American breast cancer risk [abstract]. In: Proceedings of the 17th AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2024 Sep 21-24; Los Angeles, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2024;33(9 Suppl):Abstract nr A103.

Dysfunctional High-Density Lipoprotein Cholesterol and Coronary Artery Disease: A Narrative Review.

High-density lipoprotein (HDL) cholesterol is traditionally viewed as protective against cardiovascular disease (CVD). However, emerging evidence reveals that dysfunctional HDL, characterized by impaired reverse cholesterol transport (RCT), reduced anti-inflammatory and antioxidant activities and increased endothelial dysfunction, which can contribute to coronary artery disease (CAD). Dysfunctional HDL, resulting from oxidative modifications of Apolipoprotein A-1 (Apo A-1) and enzyme inactivation, fails to effectively remove cholesterol from peripheral tissues and may promote inflammation and atherosclerosis. Genetic mutations affecting HDL metabolism further complicate its role in cardiovascular health. Studies have shown that conventional therapies aimed at raising HDL-C levels do not necessarily reduce cardiovascular events, highlighting the need for new approaches that improve HDL functionality. Therapeutic strategies such as Apo A-1 mimetic peptides, reconstituted HDL infusions, and drugs targeting specific HDL metabolic pathways are being explored. Additionally, weight loss, statin therapy, and niacin have shown potential in enhancing HDL function. The pathophysiology of dysfunctional HDL involves complex mechanisms, including oxidative stress, inflammation, and genetic mutations, which alter its structure and function, diminishing its cardioprotective effects. New functional assays, such as the cholesterol efflux capacity (CEC) and HDL inflammatory index, provide more accurate predictions of cardiovascular risk by assessing HDL quality rather than quantity. As research progresses, the focus is shifting towards therapeutic strategies that enhance HDL function and address the root causes of its dysfunction, offering a more effective approach to reducing cardiovascular risk and preventing CAD.