Risk Prediction Tools Research Articles

Deep-Learning Automation of Preoperative Radiographic Parameters Associated With Early Periprosthetic Femur Fracture After Total Hip Arthroplasty

BackgroundThe radiographic assessment of bone morphology impacts implant selection and fixation type in total hip arthroplasty (THA) and is important to minimize the risk of periprosthetic femur fracture (PFF). We utilized a deep-learning algorithm to automate femoral radiographic parameters and determined which automated parameters were associated with early PFF. MethodsRadiographs from a publicly available database and from patients undergoing primary cementless THA at a high-volume institution (2016 to 2020) were obtained. A U-Net algorithm was trained to segment femoral landmarks for bone morphology parameter automation. Automated parameters were compared against that of a fellowship-trained surgeon and compared in an independent cohort of 100 patients who underwent THA (50 with early PFF and 50 controls matched by femoral component, age, sex, body mass index, and surgical approach). ResultsOn the independent cohort, the algorithm generated 1,710 unique measurements for 95 images (5% lesser trochanter identification failure) in 22 minutes. Medullary canal width, femoral cortex width, canal flare index, morphological cortical index, canal bone ratio, and canal calcar ratio had good-to-excellent correlation with surgeon measurements (Pearson’s correlation coefficient: 0.76 to 0.96). Canal calcar ratios (0.43 ± 0.08 versus 0.40 ± 0.07) and canal bone ratios (0.39 ± 0.06 versus 0.36 ± 0.06) were higher (P < .05) in the PFF cohort when comparing the automated parameters. ConclusionsDeep-learning automated parameters demonstrated differences in patients who had and did not have early PFF after cementless primary THA. This algorithm has the potential to complement and improve patient-specific PFF risk-prediction tools.

Development and internal validation of machine learning-based models and external validation of existing risk scores for outcome prediction in patients with ischaemic stroke.

We developed new machine learning (ML) models and externally validated existing statistical models [ischaemic stroke predictive risk score (iScore) and totalled health risks in vascular events (THRIVE) scores] for predicting the composite of recurrent stroke or all-cause mortality at 90 days and at 3 years after hospitalization for first acute ischaemic stroke (AIS). In adults hospitalized with AIS from January 2005 to November 2016, with follow-up until November 2019, we developed three ML models [random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBOOST)] and externally validated the iScore and THRIVE scores for predicting the composite outcomes after AIS hospitalization, using data from 721 patients and 90 potential predictor variables. At 90 days and 3 years, 11 and 34% of patients, respectively, reached the composite outcome. For the 90-day prediction, the area under the receiver operating characteristic curve (AUC) was 0.779 for RF, 0.771 for SVM, 0.772 for XGBOOST, 0.720 for iScore, and 0.664 for THRIVE. For 3-year prediction, the AUC was 0.743 for RF, 0.777 for SVM, 0.773 for XGBOOST, 0.710 for iScore, and 0.675 for THRIVE. The study provided three ML-based predictive models that achieved good discrimination and clinical usefulness in outcome prediction after AIS and broadened the application of the iScore and THRIVE scoring system for long-term outcome prediction. Our findings warrant comparative analyses of ML and existing statistical method-based risk prediction tools for outcome prediction after AIS in new data sets.

Prediction for recurrent non-muscle invasive bladder cancer.

Non-muscle invasive bladder cancer (NMIBC) has a high recurrence rate, which places a significant burden on both patients and the healthcare system. Hence, it holds significant importance to predict the recurrence risk following treatment for individuals diagnosed with non-muscle invasive bladder cancer (NMIBC). As new generation technologies continue to emerge, an increasing number of recurrence risk prediction tools are being developed and discovered. This article provides an overview of the primary recurrence risk prediction tools currently available, including the liquid biopsy, tissue biopsy, and risk prediction tables. Each of these tools is described in detail and illustrated with relevant examples. Furthermore, we conduct an analysis of the advantages and disadvantages of these tools. This article aims to enhance the reader's understanding of the current progress in recurrence prediction tools and encourage their practical utilization in the fields of precision medicine and public health.

A Systematic Review and Critical Assessment of Breast Cancer Risk Prediction Tools Incorporating a Polygenic Risk Score for the General Population.

Single nucleotide polymorphisms (SNPs) in the form of a polygenic risk score (PRS) have emerged as a promising factor that could improve the predictive performance of breast cancer (BC) risk prediction tools. This study aims to appraise and critically assess the current evidence on these tools. Studies were identified using Medline, EMBASE and the Cochrane Library up to November 2022 and were included if they described the development and/ or validation of a BC risk prediction model using a PRS for women of the general population and if they reported a measure of predictive performance. We identified 37 articles, of which 29 combined genetic and non-genetic risk factors using seven different risk prediction tools. Most models (55.0%) were developed on populations from European ancestry and performed better than those developed on populations from other ancestry groups. Regardless of the number of SNPs in each PRS, models combining a PRS with genetic and non-genetic risk factors generally had better discriminatory accuracy (AUC from 0.52 to 0.77) than those using a PRS alone (AUC from 0.48 to 0.68). The overall risk of bias was considered low in most studies. BC risk prediction tools combining a PRS with genetic and non-genetic risk factors provided better discriminative accuracy than either used alone. Further studies are needed to cross-compare their clinical utility and readiness for implementation in public health practices.

Evaluating risk-based allocation strategies for PCSK9 inhibitors: results from a simulation study in a contemporary ASCVD cohort

Abstract Background/Aim Targeting PCSK9 inhibitors (PCSK9i) to patients with high LDL-cholesterol (LDL-C) as well as a high cardiovascular event rate yields the highest therapeutic effect and is thus reasonable from a health economic perspective. The current European dyslipidaemia guideline determines PCSK9i eligibility solely by LDL-C, whereas guidelines from the US/the UK additionally use a set of clinical risk factors. Such selection criteria require evaluation. Methods We determined the cardiovascular event rate and calculated the cost/preventable cardiovascular event for PCSK9i initiation in the target populations according to three scenarios: (i) European Society of Cardiology (ESC) 2019 dyslipidaemia guideline (PCSK9i recommended for all patients with residual LDL-C &amp;gt;55 mg/dl), (ii) American Heart Association/American College of Cardiology (AHA/ACC) 2018 cholesterol guideline (use of PCSK9i restricted to patients with residual LDL-C &amp;gt;70 mg/dl and clinical risk factors) and (iii) UK National Institute for Health and Care (NICE) recommendation (use of PCSK9i restricted to patients with residual LDL-C &amp;gt;155 mg/dl or &amp;gt;135 mg/dl with clinical risk factors). A Monte Carlo simulation taking into account statin intolerance was used to simulate uptitration of lipid lowering medications according to the different scenarios. Analyses were based on patients with atherosclerotic cardiovascular disease (ASCVD) from a contemporary all-comers cohort recruited at a large tertiary care center in Germany. Major adverse cardiovascular events (MACE, defined as the composite of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke) were recorded over a median follow-up of 4.2 years. Results We included 1922 patients with ASCVD (mean age 69.3 years, median baseline LDL-C 86.0 mg/dl, 75.0% male). Eligibility for PCSK9i was simulated to be 39.6% (ESC 2019), 6.2% (AHA/ACC 2018) and 1.4% (NICE). The 3 year cardiovascular event rates in the corresponding PCSK9i target populations were 11.4%, 13.8% and 13.8% respectively. Event rate differences were driven by a higher rate of non-fatal myocardial infarction in the risk-based target populations (5.2%, 8.2%, 7.2%). Annual cost/preventable event differed by up to a factor 3 (689,000 €, 515,000 € and 215,000 €), driven mainly by pre-PCSK9i LDL-C levels in the different target populations (68.8 mg/dl, 76.0 mg/dl and 182.7 mg/dl). (Table 1) Conclusion The AHA/ACC and NICE guidelines, based on both residual LDL-C and clinical risk factors, reduced the eligibility for PCSK9i as well as the cost/preventable cardiovascular event compared to the ESC guideline. However, the clinical criteria used for risk classification yielded limited discrimination for the identification of subpopulations at highest cardiovascular risk. Development of accurate risk prediction tools and their integration into risk-based allocation strategies, both requisites for a cost-effective use of PCSK9i, remain a challenge.Table 1

Development and validation of the DOAC Score: a novel bleeding risk prediction tool for patients with atrial fibrillation

Abstract Background Clinicians and patients must balance the ischemic benefits and bleeding risks when deciding to anticoagulate patients with atrial fibrillation (AF). Current clinical decision tools for assessing bleeding risk have limited performance and were developed for individuals anticoagulated with warfarin. Purpose This study develops and validates a clinical risk score to personalize estimates of bleeding risk for individuals with AF taking direct-acting oral anticoagulants (DOACs). Methods Among individuals taking dabigatran 150mg twice per day from 44 countries and 951 centers in this secondary analysis of the RE-LY trial, a risk score was developed to determine the probability for bleeding, based on covariates derived in a Cox proportional hazards model. The risk prediction model was internally validated with bootstrapping. We then refined the model in the GARFIELD-AF registry, with individuals taking dabigatran, edoxaban, rivaroxaban, and apixaban. To determine generalizability in external cohorts and among individuals on different DOACS, the risk prediction model was validated in the COMBINE-AF pooled clinical trial cohort and the RAMQ administrative database. The primary outcome was major bleeding. The risk score, termed the DOAC Score, was compared to the HAS-BLED score. Results Of the 5684 patients in RE-LY, 386 experienced a major bleeding event, within a median follow-up of 1.74-years. The prediction model was well-calibrated (goodness-of-fit P = 0.57) and had an optimism-corrected C statistic of 0.73 after internal validation with bootstrapping. The DOAC Score assigned points for age, creatinine clearance/glomerular filtration rate, underweight status, stroke/transient ischemic attack/embolism history, diabetes, hypertension, antiplatelet use, non-steroidal anti-inflammatory use, liver disease, and bleeding history, with each additional point scored associated with a 48.7% (95% CI: 38.9%-59.3%, P&amp;lt;0.001) increase in major bleeding in RE-LY. The score had superior performance to the HAS-BLED score in RE-LY (C Statistic: 0.73 vs 0.60, P for difference &amp;lt;0.001) and among 12,296 individuals in GARFIELD-AF (C statistic: 0.71 vs 0.66, P for Difference = 0.025). The DOAC Score had stronger predictive performance than the HAS-BLED score in both validation cohorts, including 25,586 individuals in COMBINE-AF (C statistic: 0.67 vs 0.63, P for Difference &amp;lt;0.001) and 11,945 individuals in RAMQ (C statistic: 0.65 vs 0.58, P for Difference &amp;lt;0.001). Conclusion In individuals with AF on DOAC therapy, the DOAC Score can help stratify patients based on expected bleeding risk.

External validation of the new REACT-HF score for heart failure prediction in patients with atrial fibrillation

Abstract Background Heart failure (HF) represents the main cause of hospital admissions and deaths among atrial fibrillation (AF) patients, but precise risk prediction tools are lacking. Recently, the REACT-HF score has been developed to predict incident HF hospitalizations or cardiovascular (CV) death within 2-years based on 12 variables (sex, age, height, creatinine clearance, diabetes, vascular disease, valvular heart disease, heart rate and rhythm, left ventricular hypertrophy and intraventricular conduction delay). Purpose The aim of the present study was to externally validate the REACT-HF score within the Swiss-AF and BEAT-AF cohorts, and to investigate whether this score model can be improved by adding biomarkers (hs-CRP, NT-proBNP, hs-TNT). Methods We included 2’599 AF patients from the Swiss- and BEAT-AF cohorts without prior history of HF, with at least one follow-up and with all baseline variables needed for the REACT-HF score. To validate the score, we estimated a Cox model with the original score as the only predictor. We further explored whether the predictive performance of the REACT-HF score model could be improved by adding any combination of the above listed biomarkers. The primary outcome was incident HF hospitalization or CV death. Secondary outcomes included the individual components of the primary outcome and all-cause mortality. Results Mean age was 70.2 years, 29.1% were female and 54.8% had paroxysmal AF. The mean CHA2DS2-VASc score was 2.7 (+/- 1.6). The most frequent comorbidities were hypertension (65%), diabetes (12%), vascular disease (23%) and coronary artery disease (36%). Most patients (1’883) presented a very low REACT-HF score below 0.05 (72.4%), 530 patients between 0.05 and below 0.1 (20.4%) and 186 had a risk score of 0.1 and above (7.2%). In the validation set, 380 (14.6%), 471 (18.1%), 511 (19.7%) 561 (21.6%) and 676 (26.0%) patients were categorized in the lowest to the highest quintiles based on the original score`s quintiles. Figure 1 shows the Kaplan-Meier survival curves for the primary endpoint of first HF hospitalization or CV death by quintiles of the REACT-HF score, applied in our pooled cohort. And Figure 2 shows the Kaplan-Meier curves for the 1st HF hospitalization. The incidence rates at 2-year follow-up per 100 patient-years of the primary outcome (0.39, 0.59, 1.58, 3.97, 7.53), HF hospitalization (0.39, 0.49, 1.38, 2.75, 6.26), CV death (0.0, 0.10, 0.20, 1.29, 2.12), and all cause-death (0.10, 0.49, 0.78, 2.29, 4.24) all increased gradually among the quintiles of risk score. The estimated c-statistic for the composite primary outcome was 0.764 (approximate 95% CI, 0.723-0.806). The best biomarker enhanced-model within our cohort was the model including hs-CRP and NT-proBNP (c-statistic 0.810). Conclusions The REACT-HF score discriminates well for prediction of HF hospitalization and CV death in a contemporary AF population and can be improved by including NT-pro-BNP and hs-CRP.

Abstract 19050: Polysomnography and Sleep Symptom-Based Patient Profiles Associated With Atrial Fibrillation Development

Introduction: Although sleep disorders are linked to atrial fibrillation (AF), the complex interplay remains unclear. Investigating sleep-based profiles associated with AF can deepen understanding and identify at-risk patients. Hypothesis: We hypothesized discrete phenotypes of symptoms and polysomnography (PSG)-based data differ in relation to incident AF. Methods: This retrospective cohort study of Cleveland Clinic patients (age&gt;18) who underwent PSG 11/27/2004-12/30/2015 identified patient clusters using latent class analysis of 23 sleep-related symptoms, Epworth Sleepiness Scale (ESS) score, and 24 PSG measures of sleep disordered breathing and architecture. Cox proportional hazards models with cluster membership as the independent variable and incident AF as the dependent variable were adjusted for age, sex, race, BMI, cardiopulmonary disease and risk factors, anti-arrhythmic medication, and positive airway pressure with follow up 7.6±3.4 years. Results: Of the sample (n=43,433, age 52±15, 52% male, 74% White) 9% developed AF. Five clusters were identified: 1) Hypoxic + Sleepy (n=3,245): highest % time SaO 2 below 90%, highest ESS, 2) Apneas + Arousals (n=4,592): highest AHI, highest arousal index, 3) Short sleep + Low %REM (n=6,126): shortest sleep duration, longest REM latency, lowest %REM, 4) Hypopneas (n=2,661): most hypopneas, 5) Long sleep + High %REM (n=26,809): longest sleep duration, shortest REM latency, highest %REM (reference). Compared to reference, Hypoxic + Sleepy had 48% (HR=1.48, 95%CI=1.34-1.64), Apneas + Arousals had 22% (HR=1.22, 95%CI=1.11-1.35), and Short Sleep + Low %REM had 11% (HR=1.11, 95%CI=1.01-1.22) higher risk of AF development. Internal validation indicated reasonable predictive accuracy. Sensitivity analysis excluding split night studies confirmed the associations observed. Conclusions: Of 5 patient profiles identified, Hypoxic + Sleepy conferred the strongest AF risk. Consistent with prior evidence of hypoxia as an AF driver and cardiovascular risk of the sleepy phenotype, this constellation of symptoms and physiology illustrates risk in the clinical setting, providing potential value as a risk prediction tool. Future investigation will focus on external validation of these findings.

Abstract 17076: The Relationship Between Impaired Microvascular Function, Cardiovascular Risk Profile and Subclinical Atherosclerotic Burden in a Swedish Middle-Aged Cohort

Introduction: Microvascular function is associated with cardiovascular risk factors and deteriorates in individuals with cardiovascular disease. Hypothesis: The aim of this study was to investigate the relationship between microvascular function, cardiovascular risk profile and subclinical atherosclerotic burden. Methods: The study enrolled 3809 randomly selected individuals, 50-65 years old, participating in the population-based observational cross-sectional Swedish CArdioPulmonary bioImage Study (SCAPIS). Microvascular function was measured in forearm skin using an arterial occlusion and release protocol determining the peak blood oxygen saturation, OxyP. Cardiovascular risk was calculated using the updated Systematic Coronary Risk Evaluation (SCORE2; 10-year risk of fatal and non-fatal CVD events). OxyP was compared with Coronary artery calcification score (CACS) and plaques in the carotid arteries. Results: Individuals with OxyP values in the lowest quartile (Q1; impaired microvascular function), had a mean SCORE2 of 5.8% compared to 3.8% in those with the highest values of OxyP (Q4), a relative risk increase of 53%. In the low-risk SCORE2 level (&lt;5% risk), 19.0% of the individuals have impaired microvascular function as compared to 48.1% in the high-risk SCORE2 level (&gt;10% risk). OxyP was lower in individuals with coronary artery calcification (CACS&gt;0) and those with both carotid plaque and CACS&gt;0, as compared to individuals without subclinical atherosclerotic burden (CACS=0 and no carotid plaque; 87.5 ±5.6 % and 86.9 ±6.0 % vs 88.6 ± 5.8 %, p &lt; 0.01). Conclusion: In a population without cardiovascular disease or diabetes mellitus, impaired microvascular function is associated with higher SCORE2 risk and CACS&gt;0. Hence, assessment of microvascular function may be a useful tool for accurate cardiovascular risk prediction, in addition to imaging assessment of atherosclerosis in the coronary and carotid arteries.

Abstract 17308: Patient and Primary Care Provider Experiences From an Implementation Study of a Risk Prediction Tool Integrating a Polygenic Risk Score and the QRISK2 Cardiovascular Risk Calculator

Introduction: We have previously shown that combining a polygenic risk score (PRS) for cardiovascular disease (CVD), with standard clinical risk calculators such as QRISK®2, results in improved CVD risk prediction via an integrated risk tool (CVD-IRT). Research Question: The objective of this study was to explore the implementation of CVD-IRT within routine practice in the UK National Health Service (NHS), through participant and healthcare provider (HCP) surveys and interviews. Methods: The Healthcare Evaluation of Absolute Risk Testing Study (HEART) (NCT05294419), a prospective, single-arm pragmatic trial, enrolled 836 participants undergoing health checks across 12 NHS general practices. QRISK2 and CVD-IRT scores were returned to participants via HCPs. The primary outcome of the study was feasibility of CVD-IRT implementation. This was assessed by a mixed methods approach (quantitative and qualitative methods). Results: After the results were reported and discussed, 520 surveys were completed by participants and 824 surveys were completed by HCPs. Subsequently, 21 participants were interviewed and 13 HCPs attended focus groups or interviews to explore their experiences. 23 HCPs completed a final questionnaire (34.8% physician, 21.7% research nurse, 43.5% other). For 90.7% of reports, HCPs indicated that the CVD-IRT could be incorporated into routine primary care in a straightforward manner. 80% of HCPs agreed that having these tests could lead to better health outcomes for patients, and 68.4% believed that the CVD-IRT could help them manage their patients through a shared decision making process covering lifestyle and treatment options. Participants found the report personally useful (98.5%) and easy to understand (94.3%); agreed that genetic measures are important to identify the risks of developing CVD (85.2%); thought that the test should be widely made available (interview summaries); and would recommend to friends and family (86.8%). Conclusion: The implementation of the CVD-IRT into routine health-checks was recommendable, feasible and well received by both HCPs and participants, who felt that the information was useful, could support clinical decisions, and easy to understand.

Abstract 15466: Social Determinants of Health Vulnerabilities Among Patients With Substance Use Disorder and Infective Endocarditis

Introduction: Data indicate that the incidence of infective endocarditis (IE) is rising among patients with substance use disorder (SUD) and is linked with increased mortality. However, the social determinants of health (SDOH) vulnerabilities associated with IE in this population has not been well characterized. Aims: To examine the SDOH associated with IE among patients with SUD. Methods: Using our electronic medical records, we identified patients with concurrent diagnosis of IE and SUD from 2017 until May 2023. Socio-demographics and SDOH including unstable housing (identified using diagnosis codes, residential address test, and natural language processing of clinical notes documenting homelessness and shelter referrals) were identified. A multivariable regression model was conducted to identity independent risk factors associated with IE among patients with SUD. Results: Among 26,918 patients admitted with an SUD encounter, 1,422 (5.3%) had an IE encounter. Patients with IE were younger (37 vs 43 years, P&lt;0.001), mostly white (96% vs 87.7%, P&lt;0.001), Medicaid enrollees (84.2% vs 58.6%, P&lt;0.001), and had less than high school education (10.6% vs 8.1%, P=0.005). They were more likely to reside in rural areas (58.1% vs 44.8%) and have unstable housing (28.6% vs 16.3%). Patients with IE had higher prevalence of smoking (83.5% vs 67.3%) as well as cardiac, dental, mental and other health conditions. Certain socio-demographics and SDOH were independently associated with IE on multivariable regression (Figure). Conclusions: Patients with SUD and IE have distinct socio-demographics and SDOH characteristics. Incorporating these factors into future risk prediction tools could better help identifying SUD patients who are at risk to implement targeted techniques which have been proven to lower the risk of infection.

Abstract 16832: Biomarker Enhanced Risk Prediction of Incident Heart Failure in Low-Risk Patients: The Atherosclerosis Risk in Communities (ARIC) Study

Background: Primary prevention strategies to reduce heart failure (HF) burden are urgently warranted. The Pooled Cohort equations to Prevent HF (PCP-HF) risk score is a validated risk tool that integrates readily available parameters obtained in primary care setting to estimate the 10-year risk for incident HF. Although PCP-HF performs well for high-risk individuals, the model’s predictive utility for low-risk individuals may potentially be enhanced by integration of biomarkers. Methods: In this study, the PCP-HF risk score was derived from data collected at visit 5 (2011-2013) of the ARIC cohort study and its 5-year predictive accuracy was investigated using surveillance data up to 2019. Participants with &lt;10% PCP-HF risk probability of incident HF were classified as low-risk. Incremental improvement in predicting incident HF by including laboratory and imaging-based biomarkers in the low-risk population was analyzed by logistic regression, by calculating the area under the curve (AUC) from receiver operating characteristics. Results: Of 4980 study participants with 5 years of follow up, 47% had a low PCP-HF risk score (mean age 73 years, 71% women, 32% Black). Of these, 3% developed HF within 5 years. Compared to those with a high PCP-HF risk score, individuals with low PCP-HF risk score were younger, more often female and Black; had lower left ventricular (LV) mass index and left atrial volume index, higher magnitude of LV global longitudinal strain, and lower levels of NT-proBNP and hs-TnT. On further analysis, the model incorporating laboratory and imaging-based biomarkers exhibited greater predictive accuracy for incident HF compared to the original PCP-HF risk score in this low-risk population (AUC: 0.85 vs 0.62). Conclusion: The integration of laboratory and imaging-based biomarkers with the PCP-HF risk score enhances the predictive accuracy of risk modeling for incident HF, especially in individuals identified as low-risk by the 10-year risk prediction tool.

External Validation of PRISM Score in Multiple Myeloma Patients in a Community Setting: A Retrospective Cohort Study

Background: Newly diagnosed Multiple Myeloma (MM) patients have a 10 to 20 times higher risk of developing venous thromboembolism (VTE) than the general population. The pathogenesis may involve several factors such as activation of procoagulant factors, acquired activated protein C resistance, and inflammation. In addition to the general risk factors for venous thromboembolism such as older age, immobility, and surgery, some MM treatment-related factors such as the use of immunomodulatory drug therapy and abnormal cytogenetics contribute to the increased risk. The PRISM score (Prior VTE, Race, IMiD, Surgery, Metaphase Cytogenetics) is a risk prediction tool for VTE in MM and was published in 2022. The score ranges from 0 to 18 and patients were stratified into three groups: low (0), intermediate (1-6), and high risk (≥ 7). Our primary objective was to assess the external validity of the PRISM score in our community hospital setting. Methods: All consecutive newly diagnosed MM treated at Saint Vincent Hospital from 1/1/2008 to 12/22/2022 were included in our analysis. Patients receiving therapeutic anticoagulation for other indications or a diagnosis of VTE within 6 months preceding the MM diagnosis were excluded. Variables were collected through a retrospective chart review during MM treatment initiation. Model discrimination was assessed by calculating the area under the receiver operating characteristic curve (AUROC). Logistic regression was used to calculate the odds ratio (OR) of VTE occurrence in the risk prediction model. Results: A total of 214 patients with newly diagnosed MM and available data on VTE occurrence were included. The median age of our cohort at treatment initiation was 74 years (range 31- 96). The median BMI was 27.8. 56% were male, and 44% were female. The mean PRISM score was 2.9, and 73% of MM were IgG/IgA subtype. IMiD-based induction regimen was used in 63 % of patients. A total of 24 patients (11.2%) were diagnosed with VTE after treatment initiation, of which 8 (3.7%) patients had a previous history of VTE. Prophylactic aspirin was used in 53.8% and abnormal metaphase cytogenetics was identified in 13% of MM cases. The median time to VTE from induction therapy was 4.8 months. The PRISM score was low in 116 patients, intermediate in 89, and high in 9 patients. The AUROC of the model was 0.759 (95% CI, 0.644-0.874) (1). The cumulative incidence of VTE at 6 months, 12 months, and more than 1 year was 6%, 7%, and 11.2% of the patients respectively. Each 1-point increase in the score was associated with a significantly higher chance of VTE occurrence in the model [ HR= 1.53, 95% CI (1.27-1.85), P&amp;lt;0.001]. Among the parameters, patients with a history of VTE [ HR= 1.55, 95% CI (1.26-1.91), P&amp;lt;0.001] and with abnormal metaphase cytogenetics [HR= 2.35, 95% CI (1.37-4.05), P&amp;lt;0.001] had a higher strength of association with the development of VTE after induction (2). Furthermore, among 9 patients with a high-risk PRISM score (≥ 7), 7 (77%) developed VTE within the first 6 months after treatment initiation. Conclusions: Our findings suggest that PRISM [AUROC=0.75, 95% CI (0.644-0.874), P&amp;lt;0.001] score could statistically predict VTE outcomes in our patient population. Abnormal metaphase cytogenetics was a significant predictor of VTE in the validation cohort. Abnormal cytogenetics has been well established to be an indicator of aggressive tumor biology and proliferation in stroma-independent MM. It has been established to be a predictor of VTE in acute myeloid leukemia as well. Also, high-risk scores were predictive of earlier development of VTE after treatment induction. Our study strengthens the evidence of using PRISM as a predictive tool. Further studies are needed to enhance the sensitivity of our existing tool.

Validation of the International Immunoglobulin A (IgA) Risk Prediction Tool in American Indians and Hispanics

Validation of the International Immunoglobulin A (IgA) Risk Prediction Tool in American Indians and Hispanics

Development and Validation of a Risk Prediction Tool for In-hospital Mortality After Thoracic Endovascular Repair in Patients with Blunt Thoracic Aortic Injury Using the Aortic Trauma Foundation Registry

The objective of our present effort was to use an international blunt thoracic aortic injury (BTAI) registry to create a prediction model identifying important preoperative and intraoperative factors associated with postoperative mortality, and to develop and validate a simple risk prediction tool that could assist with patient selection and risk stratification in this patient population. For the purpose of the present study, all patients undergoing thoracic endovascular aortic repair (TEVAR) for BTAI and registered in the Aortic Trauma Foundation (ATF) database from January 2016 as of June 2022 were identified. Patients undergoing medical management or open repair were excluded. The primary outcome was binary in-hospital all-cause mortality. Two predictive models were generated: a preoperative model (i.e. only including variables before TEVAR or intention-to-treat) and a full model (i.e. also including variables after TEVAR or per-protocol). Out of a total of 944 cases included in the ATF registry until June 2022, 448 underwent TEVAR and were included in the study population. TEVAR for BTAI was associated with an 8.5% in-hospital all-cause mortality in the ATF dataset. These study subjects were subsequently divided using 3:1 random sampling in a derivation cohort (336; 75.0%) and a validation cohort (112; 25.0%). The median age was 38years, and the majority of patients were male (350; 78%). A total of 38 variables were included in the final analysis. Of these, 17 variables were considered in the preoperative model, 9 variables were integrated in the full model, and 12 variables were excluded owing to either extremely low variance or strong correlation with other variables. The calibration graphs showed how both models from the ATF dataset tended to underestimate risk, mainly in intermediate-risk cases. The discriminative capacity was moderate in all models; the best performing model was the full model from the ATF dataset, as evident from both the Receiver Operating Characteristic curve (Area Under the Curve 0.84; 95% CI 0.74-0.91) and from the density graph. In this study, we developed and validated a contemporary risk prediction model, which incorporates several preoperative and postoperative variables and is strongly predictive of early mortality. While this model can reasonably predict in-hospital all-cause mortality, thereby assisting physicians with risk-stratification as well as inform patients and their caregivers, its intrinsic limitations must be taken into account and it should only be considered an adjunctive tool that may complement clinical judgment and shared decision-making.

(Cost-)effectiveness of an individualised risk prediction tool (PERSARC) on patient’s knowledge and decisional conflict among soft-tissue sarcomas patients: protocol for a parallel cluster randomised trial (the VALUE-PERSARC study)

IntroductionCurrent treatment decision-making in high-grade soft-tissue sarcoma (STS) care is not informed by individualised risks for different treatment options and patients’ preferences. Risk prediction tools may provide patients and professionals...

Cardiovascular risk assessment in patients with antiphospholipid syndrome: a cross-sectional performance analysis of nine clinical risk prediction tools

ObjectivesThis study aimed to assess the performance of cardiovascular risk (CVR) prediction models reported by European Alliance of Associations for Rheumatology and European Society of Cardiology recommendations to identify high-atherosclerotic...

In-hospital fall prediction using machine learning algorithms and the Morse fall scale in patients with acute stroke: a nested case-control study

BackgroundFalls are one of the most common accidents in medical institutions, which can threaten the safety of inpatients and negatively affect their prognosis. Herein, we developed a machine learning (ML) model for fall prediction in patients with acute stroke and compared its accuracy with that of the existing fall risk prediction tool, the Morse Fall Scale (MFS).MethodsThis is a retrospective nested case-control study. The initial sample size was 8462 admitted to a single cerebrovascular specialty hospital with acute stroke. A total of 156 fall events occurred, and each fall case was randomly matched with six control cases. Six ML algorithms were used, namely, regularized logistic regression, support vector machine, naïve Bayes (NB), k-nearest neighbors, random forest, and extreme-gradient boosting (XGB).ResultsWe included 156 in the fall group and 934 in the non-fall group. The mean ages of the fall and non-fall groups were 68.3 (± 12.2) and 65.3 (± 12.9) years old, respectively. The MFS total score was significantly higher in the fall group (54.3 ± 18.3) than in the non-fall group (37.7 ± 14.7). The area under the receiver operating curve (AUROC) of the MFS in predicting falls was 0.76 (0.73–0.79). XGB had the highest AUROC of 0.85 (0.78–0.92), and XGB and NB had the highest F1 score of 0.44.ConclusionsThe AUROC values of all of ML algorithms were similar to those of the MFS in predicting fall risk in patients with acute stroke, allowing for accurate and efficient fall screening.

Multiparametric MRI-based 5-year Risk Prediction Model for Biochemical Recurrence of Prostate Cancer after Radical Prostatectomy.

Background Current predictive tools to estimate the risk of biochemical recurrence (BCR) after treatment of prostate cancer do not consider multiparametric MRI (mpMRI) information. Purpose To develop a risk prediction tool that considers mpMRI findings to assess the risk of 5-year BCR after radical prostatectomy. Materials and Methods In this retrospective single-center analysis in 1459 patients with prostate cancer who underwent mpMRI before radical prostatectomy (in 2012-2015), the outcome of interest was 5-year BCR (two consecutive prostate-specific antigen [PSA] levels > 0.2 ng/mL [0.2 µg/L]). Patients were randomly divided into training (70%) and test (30%) sets. Kaplan-Meier plots were applied to the training set to estimate survival probabilities. Multivariable Cox regression models were used to test the relationship between BCR and different sets of exploratory variables. The C-index of the final model was calculated for the training and test sets and was compared with European Association of Urology, University of California San Francisco Cancer of the Prostate Risk Assessment, Memorial Sloan-Kettering Cancer Center, and Partin risk tools using the partial likelihood ratio test. Five risk categories were created. Results The median duration of follow-up in the whole cohort was 59 months (IQR, 32-81 months); 376 of 1459 (25.8%) patients had BCR. A multivariable Cox regression model (referred to as PIPEN, and composed of PSA density, International Society of Urological Pathology grade group, Prostate Imaging Reporting and Data System category, European Society of Urogenital Radiology extraprostatic extension score, nodes) fitted to the training data yielded a C-index of 0.74, superior to that of other predictive tools (C-index 0.70 for all models; P ≤ .01) and a median higher C-index on 500 test set replications (C-index, 0.73). Five PIPEN risk categories were identified with 5-year BCR-free survival rates of 92%, 84%, 71%, 56%, and 26% in very low-, low-, intermediate-, high-, and very high-risk patients, respectively (all P < .001). Conclusion A five-item model for predicting the risk of 5-year BCR after radical prostatectomy for prostate cancer was developed and internally verified, and five risk categories were identified. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Aguirre and Ortegón in this issue.

External Validation of a Risk Prediction Tool in Determining Early Death Among Elderly Patients Initiated on Dialysis

External Validation of a Risk Prediction Tool in Determining Early Death Among Elderly Patients Initiated on Dialysis