Clinical Risk Prediction Models Research Articles

Development of transdiagnostic clinical risk prediction models for 12-month onset and course of eating disorders among adolescents in the community.

To develop and internally validate risk prediction models for adolescent onset and persistence of eating disorders. N = 963 Australian adolescents (11-19 years) in the EveryBODY Study cohort completed online surveys in 2018 and 2019. Models were built to predict 12-month risk of (1) onset, and (2) persistence of a DSM-5 eating disorder. Onset Model. Of the n = 687 adolescents without an eating disorder at baseline, 16.9% were identified with an eating disorder after 12 months. The prediction model was based on evidence-based risk factors for eating disorder onset available within the dataset (sex, body mass index percentile, strict weight loss dieting, history of bullying, psychological distress, weight/shape concerns). This model showed fair discriminative performance (mean AUC = .75). The most important factors were psychological distress, weight and shape concerns, and female sex. Diagnostic Persistence Model. Of the n = 276 adolescents with an eating disorder at baseline, 74.6% were identified as continuing to meet criteria for an eating disorder after 12 months. The prediction model for diagnostic persistence was based on available evidence-based risk factors for eating disorder persistence (purging, distress, social impairment). This model showed poor discriminative performance (mean AUC = .65). The most important factors were psychological distress and self-induced vomiting for weight control. We found preliminary evidence for the utility of a parsimonious model for 12-month onset of an eating disorder among adolescents in the community. Future research should include additional evidence-based risk factors and validate models beyond the original sample. This study demonstrated the feasibility of developing parsimonious and accurate models for the prediction of future onset of an eating disorder among adolescents. The most important predictors in this model included psychological distress and weight and shape concerns. This study has laid the ground work for future research to build and test more accurate prediction models in diverse samples, prior to translation into a clinical tool for use in real world settings to aid decisions about referral to early intervention.

Update of a Multivariable Opioid Overdose Risk Prediction Model to Enhance Clinical Care for Long-term Opioid Therapy Patients.

Clinical opioid overdose risk prediction models can be useful tools to reduce the risk of overdose in patients prescribed long-term opioid therapy (LTOT). However, evolving overdose risk environments and clinical practices in addition to potential harmful model misapplications require careful assessment prior to widespread implementation into clinical care. Models may need to be tailored to meet local clinical operational needs and intended applications in practice. To update and validate an existing opioid overdose risk model, the Kaiser Permanente Colorado Opioid Overdose (KPCOOR) Model, in patients prescribed LTOT for implementation in clinical care. The retrospective cohort study consisted of 33, 625 patients prescribed LTOT between January2015 and June2019 at Kaiser Permanente Colorado, with follow-up through June2021. The outcome consisted of fatal opioid overdoses identified from vital records and non-fatal opioid overdoses from emergency department and inpatient settings. Predictors included demographics, medication dispensings, substance use disorder history, mental health history, and medical diagnoses. Cox proportional hazards regressions were used to model 2-year overdose risk. During follow-up, 65 incident opioid overdoses were observed (111.4 overdoses per 100,000 person-years) in the study cohort, of which 11 were fatal. The optimal risk model needed to risk-stratify patients and to be easily interpreted by clinicians. The original 5-variable model re-validated on the new study cohort had a bootstrap-corrected C-statistic of 0.73 (95% CI, 0.64-0.85) compared to a C-statistic of 0.80 (95% CI, 0.70-0.88) in the updated model and 0.77 (95% CI, 0.66-0.87) in the final adapted 7-variable model, which was also well-calibrated. Updating and adapting predictors for opioid overdose in the KPCOOR Model with input from clinical partners resulted in a parsimonious and clinically relevant model that was poised for integration in clinical care.

Clinical characteristics and risk factors of cardiac surgery associated-acute kidney injury progressed to chronic kidney disease in adults: A retrospective, observational cohort study.

To retrospectively investigate the clinical characteristics and risk factors of cardiac surgery associated-acute kidney injury (CS-AKI) progressed to chronic kidney disease (CKD) in adults and to evaluate the performance of clinical risk factor model for predicting CS-AKI to CKD. In this retrospective, observational cohort study, we included patients who were hospitalized for CS-AKI without a prior CKD [estimated glomerular filtration rate (eGFR) < 60 ml · min-1·1.73 m-2] at Central China Fuwai Hospital from January 2018 to December 2020. Survived patients were followed up for 90 days, the endpoint was CS-AKI to CKD, and then divided them into two groups (with or without CS-AKI to CKD). The baseline data including demographics, comorbidities, renal function, and other laboratory parameters were compared between two groups. The logistic regression model was used to analyze the risk factors for CS-AKI to CKD. Finally, receiver operator characteristic (ROC) curve was drawn to evaluate the performance of the clinical risk factor model for predicting CS-AKI to CKD. We included 564 patients with CS-AKI (414 males, 150 females; age: 57.55 ± 11.86 years); 108 (19.1%) patients progressed to new-onset CKD 90 days after CS-AKI. Patients with CS-AKI to CKD had a higher proportion of females, hypertension, diabetes, congestive heart failure, coronary heart disease, low baseline eGFR and hemoglobin level, higher serum creatinine level at discharge (P < 0.05) than those without CS-AKI to CKD. Multivariate logistic regression analysis revealed that female sex(OR = 3.478, 95% CI: 1.844-6.559, P = 0.000), hypertension (OR = 1.835, 95% CI: 1.046-3.220, P = 0.034), coronary heart disease (OR = 1.779, 95% CI: 1.015-3.118, P = 0.044), congestive heart failure (OR = 1.908, 95% CI: 1.124-3.239, P = 0.017), preoperative low baseline eGFR (OR = 0.956, 95% CI: 0.938-0.975, P = 0.000), and higher serum creatinine level at discharge (OR = 1.109, 95% CI: 1.014-1.024, P = 0.000) were independent risk factors for CS-AKI to CKD. The clinical risk prediction model including female sex, hypertension, coronary heart disease, congestive heart failure, preoperative low baseline eGFR, and higher serum creatinine level at discharge produced a moderate performance for predicting CS-AKI to CKD (area under ROC curve = 0.859, 95% CI: 0.823-0.896). Patients with CS-AKI are at high risk for new-onset CKD. Female sex, comorbidities, and eGFR can help identify patients with a high risk for CS-AKI to CKD.

The impact of genomic biomarkers on a validated clinical risk prediction model for upgrading/upstaging among men with low-risk prostate cancer.

262 Background: While active surveillance (AS) is the preferred management strategy for men diagnosed with low-risk prostate cancer (PCa), the inability to distinguish indolent from aggressive tumors in clinically low-risk patients complicates decision-making. Genomic classifiers (GCs) were introduced to improve risk stratification, based on their ability to predict the risk of upgrading or upstaging (UG/US) in this setting. We assessed the impact of GCs on UG/US risk prediction added to a rich clinical model to better guide management decision-making for low-risk patients. Methods: We used multivariate logistic regression and receiver operating characteristic (ROC) curves to develop a prediction model for UG/US in men with low-risk, low-volume intermediate-risk PCa who were potential candidates for AS. The model was developed among 864 men and validated in an independent cohort of 2,267 men with similar risk profiles. After computing areas under the ROC curve (AUC) from these probabilities, we tested the model’s predictive ability with the addition of a series of GCs (OncoType Dx Genomic Prostate Score, Decipher score, and selected Decipher GRID scores) using the logistic model constructed for the development cohort to estimate the predicted probability of UG/US. Results: The prediction model for the development cohort included five diagnostic variables that were significantly associated with risk of UG/US: diagnostic grade (OR 5.83, 95% CI 3.73-9.10), PSA (OR 1.10, 95% CI 1.01-1.20), percent positive cores (OR 1.01, 95% CI 1.01-1.02), TRUS prostate volume (OR 0.98, 95% CI 0.97-0.99), and age (OR 1.05, 95% CI 1.02-1.07). The pooled AUC was 0.72 for 10 iterations of the model. When the addition of GCs was applied to this model, Genomic Prostate Score was independently associated with risk of UG/US (OR 1.42, 95%CI 0.016-0.066, p=0.017). However, the pooled AUC was 0.71, indicating comparable predictive performance to the risk prediction model. Conclusions: The addition of GCs to a validated rich model incorporating detailed clinical variables, when applied to favorable-risk PCa patients, did not substantially improve prediction of UG/US. Our findings suggest that widespread use of biomarkers to guide management or the intensity of follow up may not be supported in men with low-risk, low-volume disease pursuing AS.

Improvement of a clinical colorectal cancer risk prediction model integrating polygenic risk.

81 Background: Improving colorectal cancer risk prediction and stratification is pivotal for implementing better screening and prevention programs in public health and for enabling a personalised approach for assessing patients’ colorectal cancer risk. Methods: In this study, we used the UK Biobank to compare the performance of a risk prediction model incorporating two different polygenic risk scores – one comprising 45 SNPs and the other comprising 140 SNPs. The clinical component of the risk prediction model included a simple measure of first-degree family history. We used age- and sex-specific population incidence rates to calculate full-lifetime risks. Results: The model using the 140-SNP PRS showed an improvement in discrimination, calibration and risk stratification over the model using the 45-SNP PRS for full-lifetime risk: discrimination was 0.706 (95% CI 0.697–0.715) and 0.674 (95% CI 0.664–0.683), respectively, and the P for difference was &lt; 0.001. The 140-SNP model was well calibrated and showed a small overestimation of risk 0.951 (95% CI 0.918–0.986). Standard incidence ratios compared to population incidence rates showed that, for the 140-SNP model, the top quintile of risk shows a 27% improvement compared to the 45-SNP model. Furthermore, there was a 3-fold difference in colorectal cancer incidence between adults identified in the top quintile compared to the bottom quintile of risk using the 140-SNP model versus the 45-SNP model. Conclusions: This updated risk prediction score with a 140–SNP PRS and a simple measure of family history, improves risk prediction and risk stratification in the general population compared with a similar model with a 45-SNP PRS, and will ultimately assist in colorectal cancer disease prevention in the clinic.

A Prognostic Study for the Development of Risk Prediction Model for the Success of Vaginal Birth Following a Cesarean Surgery at Felege Hiwot Comprehensive Specialized Hospital, Northwest Ethiopia.

An attempt at vaginal delivery by a woman who has previously had a cesarean section is known as a trial of labor after cesarean section. The most important issues are how to accurately anticipate successful vaginal birth after cesarean surgery and how to calculate the likelihood of success of vaginal birth after caesarean section that is suitable for women. Consequently, a tailored prediction of vaginal birth after caesarean section may result in a more effective counseling. To create a clinical risk score and prediction model for the success of vaginal birth following a previous caesarean section in women. A prognostic analysis was carried out at Felege Hiwot Comprehensive and Specialized Referral Hospital from 30 February 2017 to 30 March 2021. R statistical programming language version 4.0 was used for analysis once the data had been coded and entered into Epidata, version 3.02. Significant factors (P< 0.05) were kept in the backward multivariable logistic regression model, and variables with (P<0.25) from the bi-variable logistic regression analysis were also added. After a cesarean section, 67% of women were successful in giving birth vaginally. Previous successful vaginal birth after cesarean surgery, rupture of the membranes, and initiation time of ANC, the beginning of labor, parity and time since the previous delivery were remained in the final multivariable prediction model. The AUC of the model was 0.748 (95% CI: 0.714-0.781). Overall, this study demonstrated the likelihood of predicting vaginal birth utilizing the ideal confluence of parity, membrane rupture, and onset of labor, prior history of VBAC, inter-delivery gap, and ANC beginning time. Sixty-seven percent of VBACs were successful. As a result, this model may aid in identifying pregnant women who are candidates for VBAC and who have a better likelihood of success.

104 EFFECTIVENESS OF 2014 ESC HCM RISK-SCD SCORE IN PREDICTION OF APPROPRIATE ICD SHOCK

Abstract Background Since 2014, European Guidelines have recommended using a novel clinical risk prediction model, called the HCM Risk-SCD, to identify which patients with hypertrophic cardiomyopathy (HCM) who are at higher risk for sudden cardiac death (SCD) and would benefit most from having a prophylactic cardioverter defibrillator (ICD) implantable. In recent years, evidence has called into question the effectiveness of this model in patient selection. Material and Methods Data from consecutive patients with HCM and ICD who were followed at our Cardiology Department from January 2000 to January 2022 were retrospectively collected. Results Among 702 HCM patients, 52 (7%) received an ICD (female, 31%; mean age at implantation, 49 ± 20; S-ICD, 13%; single-chamber ICD, 50%). Of them, 7 (13%) patients were implanted for secondary prevention while 45 (87%) for primary prevention. Out of the 45 primary prevention patients, 7 (16%) experienced 8 appropriate shocks (AS) while 10 (22%) had 15 inappropriate shocks (IS), during a mean follow up of 8 ± 5 years. Overall, mean HCM Risk-SCD score was 6.8 ± 4.7. Patients with AS showed a non-significant higher risk score when compared to non-AS subjects (7.1 ± 5.4 vs 4.8 ± 5.1, p=0.27). By stratifying study population according to pre-established and clinically adopted HCM Risk-SCD score cut-off (such as &amp;lt; 4 vs. &amp;gt; or = 4), we failed to highlight any significant difference in term of AS between groups. Indeed, out of 10 (22%) patients with HCM Risk-SCD score &amp;lt; 4, 3 (7%) experienced AS, while among 35 (78%) patients with HCM Risk-SCD score &amp;gt; or = 4, 4 (9%) had AS (30% vs. 11%, p=0.17). Conclusions The HCM Risk-SCD score does not seem to be able to predict the occurrence of life-threatening ventricular arrhythmias and AS in HCM patients. Our data highlight the need to elaborate a new score in order to improve SCD risk stratification and better select patients for ICD implantation.

Registry of Aortic Diseases to Model Adverse Events and Progression (ROADMAP) in Uncomplicated Type B Aortic Dissection: Study Design and Rationale.

To describe the design and methodological approach of a multicenter, retrospective study to externally validate a clinical and imaging-based model for predicting the risk of late adverse events in patients with initially uncomplicated type B aortic dissection (uTBAD). The Registry of Aortic Diseases to Model Adverse Events and Progression (ROADMAP) is a collaboration between 10 academic aortic centers in North America and Europe. Two centers have previously developed and internally validated a recently developed risk prediction model. Clinical and imaging data from eight ROADMAP centers will be used for external validation. Patients with uTBAD who survived the initial hospitalization between January 1, 2001, and December 31, 2013, with follow-up until 2020, will be retrospectively identified. Clinical and imaging data from the index hospitalization and all follow-up encounters will be collected at each center and transferred to the coordinating center for analysis. Baseline and follow-up CT scans will be evaluated by cardiovascular imaging experts using a standardized technique. The primary end point is the occurrence of late adverse events, defined as aneurysm formation (≥6 cm), rapid expansion of the aorta (≥1 cm/y), fatal or nonfatal aortic rupture, new refractory pain, uncontrollable hypertension, and organ or limb malperfusion. The previously derived multivariable model will be externally validated by using Cox proportional hazards regression modeling. This study will show whether a recent clinical and imaging-based risk prediction model for patients with uTBAD can be generalized to a larger population, which is an important step toward individualized risk stratification and therapy.Keywords: CT Angiography, Vascular, Aorta, Dissection, Outcomes Analysis, Aortic Dissection, MRI, TEVAR© RSNA, 2022See also the commentary by Rajiah in this issue.

Active Data Science for Improving Clinical Risk Prediction

Clinical risk prediction models are commonly developed in a post-hoc and passive fashion, capitalizing on convenient data from completed clinical trials or retrospective cohorts. Impacts of the models often end at their publication rather than with the patients. The field of clinical risk prediction is rapidly improving in a progressively more transparent data science era. Based on collective experience over the past decade by the Prostate Biopsy Collaborative Group (PBCG), this paper proposes the following four data science-driven strategies for improving clinical risk prediction to the benefit of clinical practice and research. The first proposed strategy is to actively design prospective data collection, monitoring, analysis and validation of risk tools following the same standards as for clinical trials in order to elevate the quality of training data. The second suggestion is to make risk tools and model formulas available online. User-friendly risk tools will bring quantitative information to patients and their clinicians for improved knowledge-based decision-making. As past experience testifies, online tools expedite independent validation, providing helpful information as to whether the tools are generalizable to new populations. The third proposal is to dynamically update and localize risk tools to adapt to changing demographic and clinical landscapes. The fourth strategy is to accommodate systematic missing data patterns across cohorts in order to maximize the statistical power in model training, as well as to accommodate missing information on the end-user side too, in order to maximize utility for the public.

Clinical characteristics and mortality risk prediction model in children with acute myocarditis

BackgroundAcute myocarditis (AMC) can cause poor outcomes or even death in children. We aimed to identify AMC risk factors and create a mortality prediction model for AMC in children at hospital admission.MethodsThis was a single-center retrospective cohort study of AMC children hospitalized between January 2016 and January 2020. The demographics, clinical examinations, types of AMC, and laboratory results were collected at hospital admission. In-hospital survival or death was documented. Clinical characteristics associated with death were evaluated.ResultsAmong 67 children, 51 survived, and 16 died. The most common symptom was digestive disorder (67.2%). Based on the Bayesian model averaging and Hosmer–Lemeshow test, we created a final best mortality prediction model (acute myocarditis death risk score, AMCDRS) that included ten variables (male sex, fever, congestive heart failure, left-ventricular ejection fraction < 50%, pulmonary edema, ventricular tachycardia, lactic acid value > 4, fulminant myocarditis, abnormal creatine kinase-MB, and hypotension). Despite differences in the characteristics of the validation cohort, the model discrimination was only marginally lower, with an AUC of 0.781 (95% confidence interval = 0.675–0.852) compared with the derivation cohort. Model calibration likewise indicated acceptable fit (Hosmer‒Lemeshow goodness-of-fit, P¼ = 0.10).ConclusionsMultiple factors were associated with increased mortality in children with AMC. The prediction model AMCDRS might be used at hospital admission to accurately identify AMC in children who are at an increased risk of death.

A Clinical and Sociodemographic Risk Prediction Model Evaluating Adherence to Long-Term Care in Survivors of Childhood Leukemia

Background: Regular long-term follow-up care is important for surveillance and timely treatment of late effects after childhood leukemia therapy; however, structural barriers may disproportionately affect at-risk populations. Our objective was to evaluate factors associated with failure to utilize Long-Term Survivor (LTS) services following childhood leukemia treatment. Methods: We electronically extracted and manually curated electronic health records to identify childhood leukemia cases diagnosed between 2011 and 2014 at Texas Children's Cancer and Hematology Center. Individuals treated with a stem cell transplant, or those that transferred to an outside hospital, relapsed, or died within two years of end of treatment (EOT) were excluded. Address at the EOT was geocoded using ArcGIS to estimate travel time to the clinic and Area Deprivation Index, a composite measure of 17 neighborhood-level social determinants of health (SDOH). Multivariable logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (CI) for the association between clinical factors and attending LTS Clinic. LTS Clinic attendance was defined as at least one visit with an LTS provider at any time following EOT. Predictive performance of the model was evaluated using area under the receiver operating characteristic curve (AUC-ROC). Results: We identified 249 survivors of childhood leukemia that were eligible for LTS care in the designated time frame. Most were male (55%), non-Latino white (27%), Latino (57%), or non-Latino Black (10%) and were diagnosed at a median age of 5.7 years (range: 0.6-20). Overall, 41 survivors (16.5%) never attended LTS clinic. These included 15 survivors (6.0%) that were seen for initial disease follow-up after EOT but had no further contact with the hospital system beginning 2 years after EOT. In adjusted models, older age at EOT was associated with a higher likelihood of never attending LTS clinic (p<0.001). For example, survivors age 15 years and older at EOT were 7.4-times more likely to never attend LTS clinic (95% CI: 1.95-28.24). Additionally, compared with privately insured survivors, those with public insurance (OR=4.09, 95% CI: 1.67-9.99) and those that were uninsured (OR=4.68, 95% CI: 0.69-31.52) were more likely to never attend LTS clinic. LTS clinic attendance was largely comparable regardless of diagnosis, treatment exposures, sex, primary household language, and race/ethnicity, although Latinos were less likely to never attend an LTS visit compared with non-Latino whites (OR=0.36, 95% CI: 0.14-0.94). These observed associations were also independent of estimated travel time and neighborhood-level SDOH. Clinical and demographic factors provided reasonable predictive performance (AUC-ROC: 0.733, 95% CI: 0.655-0.821), e.g., categorizing survivors on their predicted probability of never attending LTS clinic based on observed clinical and demographic factors correctly identified 63.4% of survivors who never attend LTS clinic in the upper tertile of the risk prediction distribution. Conclusions: Texas has the highest rates of uninsured children in the United States, comprising 12.7% of all children and 17.5% of Latino children living in the state (2019 American Community Survey data). Moreover, Medicaid eligibility in Texas is significantly limited for adults between the ages of 18 and 65 years. In this study, both older age at EOT and public insurance or uninsured status at EOT were strongly predictive of no documented LTS care among survivors of childhood leukemia, suggesting health insurance is major limiting factor to accessing survivorship services in Texas. Limiting predictive factors to race/ethnicity, age at EOT and payer status at EOT provides reasonable prediction of leukemia survivor that do not engage in LTS care (AUC-ROC: 0.711). Validation of a simple risk prediction model using information readily available in the electronic medical record at EOT may inform targeted interventions to augment access to survivorship services in this vulnerable population.

Clinically Consistent Prostate Cancer Outcome Prediction Models with Machine Learning

<h3>Purpose/Objective(s)</h3> State-of-the-art prediction modeling relies on machine learning (ML). To provide accurate out-of-sample predictions, such as when the number of available covariates is large, these models trade-off variance by adding bias. However, these artificial biases frequently create anomalies that are clinically inconsistent, hindering their use in clinical practice. We illustrate such biases in outcome prediction ML models using the National Cancer Data Base (NCDB) and propose a solution that leverages clinical expertise. We hypothesize that integrating clinical expertise into outcome prediction ML models will correct significant problems present in ML modeling without compromising the model accuracy, enabling models to be more clinically intuitive. <h3>Materials/Methods</h3> We use data from 92,990 prostate cancer patients in the NCDB (death rate: 28%) to build a gradient boosted machine (GBM) model to predict 10-year survival post-treatment. We then investigate how the model output depends on each covariate and identify inconsistencies with clinical knowledge. To correct for these biases, we collect clinical intuition from five oncologists through a survey that asks how increasing a covariate impacts (increase/decrease/variable effect) a patient's survival. We then encode the survey results as clinically informed monotonic constraints into a LASSO model and compare the AUC with and without constraints. <h3>Results</h3> We find that while the GBM model has a relatively high out-of-sample AUC of 0.79 (compared to similar studies, with AUC between 0.65-0.83, although patient cohort, methodology, and outcomes may differ), the way some covariates are used is not clinically consistent. For example, keeping other covariates equal, the probability of survival should consistently be higher for patients with lower prostate specific antigen (PSA) levels compared to patients at higher levels. But the model predicts (with high confidence) a sporadic, non-monotonic pattern. This is because the GBM model biases its predictions for patients that comprise a small fraction of the patient cohort (i.e., patients with very low PSA values) to avoid high variance, but the artificially injected bias is towards lower survival which is not clinically consistent. Covariates named by oncologists to have a monotonic effect on the probability of survival include PSA and age. We find that by adding clinically informed constraints to a LASSO model, the problem of clinical inconsistency is addressed without compromising model performance (AUC is 0.77 with and without constraints). <h3>Conclusion</h3> We illustrate that even in a large-scale regime, state-of-the-art clinical risk prediction models can suffer from biases. Our work shows that innovative integration of clinical intuition can make models clinically consistent and verifies that this method is promising research to resolve critical challenges in ML-based decision support.

Bleeding after percutaneous coronary intervention and selection of candidates for long-term dual antithrombotic treatment

Abstract Background The European Society of Cardiology recommends addition of a second antithrombotic drug (a P2Y12 inhibitor or rivaroxaban 2.5 mg twice daily) on top of aspirin in selected patients with chronic coronary syndrome (CCS) at high residual risk of ischemic events. However, this treatment increases bleeding risk, and identifying subsets of patients with the most favorable trade-off between ischemic and bleeding risk thus is essential. We hypothesized that patients undergoing percutaneous coronary intervention (PCI) who tolerate subsequent dual antiplatelet therapy (aspirin plus a P2Y12 inhibitor) without any bleeding complications selected themselves as candidates for prolonged dual antithrombotic therapy. Methods and results We included 30,531 patients with CCS treated with dual antiplatelet therapy after first-time PCI with a drug-eluting stent in Western Denmark (3.5 million inhabitants) from 1999 to 2018. Of these, 1,220 (4%) were hospitalized for bleeding within one year after PCI (bleeders) and 29,311 (96%) were not (non-bleeders). Patients were followed for maximum nine years (median follow-up 5.4 years). Bleeders had an increased nine-year risk of death (adjusted hazard ratio [aHR] 1.54, 95% CI 1.37–1.73) and hospitalization for bleeding (aHR 2.53, 95% CI 2.20–2.90). These associations were particularly strong for women. Looking at types of bleeding, the strongest predictors of death were gastrointestinal bleeding, cerebral bleeding, and anemia due to bleeding. Risks of myocardial infarction and ischemic stroke did not differ between bleeders and non-bleeders (Table). We then stratified non-bleeders according to their thromboembolic risk using the CHADS-P2A2RC score – a validated clinical risk prediction model developed to estimate thromboembolic risk in patients without atrial fibrillation. Non-bleeders with a high estimated thromboembolic risk (CHADS-P2A2RC score ≥4) had higher nine-year risks of myocardial infarction (hazard ratio [HR] 1.88, 95% CI 1.78–2.07), ischemic stroke (HR 3.02, 95% CI 2.66–3.43), hospitalization for bleeding (HR 1.98, 95% CI 1.81–2.16) and, in particular, death (HR 4.48, 95% CI 4.21–4.77) than non-bleeders with a low-to-moderate predicted risk (CHADS-P2A2RC score &amp;lt;4). Conclusions Patients with CCS experiencing a bleeding event during the first year after first-time PCI had a substantially higher long-term risk of death and recurrent bleeding, but not a higher risk of ischemic events. Therefore, bleeding events during the first year after PCI may guide the preclusion of selected patients from long-term dual antithrombotic therapy. Among non-bleeders, the risk of ischemic events rose proportionately more than the risk of bleeding when comparing high-risk with low-risk patients. This is an important finding for clinicians, for whom accurate identification of patients at highest risk of ischemic events is an essential step in treatment allocation. Funding Acknowledgement Type of funding sources: None.

Assessment of a Large-Scale Unbiased Malignant Pleural Effusion Proteomics Study of a Real-Life Cohort.

Simple SummaryPleural effusion (PE) occurs as a consequence of various pathologies. Malignant effusion due to lung cancer is one of the most frequent causes. A method for accurate differentiation of malignant from benign PE is an unmet clinical need. Proteomics profiling of PE has shown promising results. However, mass spectrometry (MS) analysis typically involves the tedious elimination of abundant proteins before analysis, and clinical annotation of proteomics profiled cohorts is limited. This study compares the proteomes of malignant PE and nonmalignant PE, identifies lung cancer malignant markers in agreement with other studies, and identifies markers strongly associated with patient survival.Background: Pleural effusion (PE) is common in advanced-stage lung cancer patients and is related to poor prognosis. Identification of cancer cells is the standard method for the diagnosis of a malignant PE (MPE). However, it only has moderate sensitivity. Thus, more sensitive diagnostic tools are urgently needed. Methods: The present study aimed to discover potential protein targets to distinguish malignant pleural effusion (MPE) from other non-malignant pathologies. We have collected PE from 97 patients to explore PE proteomes by applying state-of-the-art liquid chromatography-mass spectrometry (LC-MS) to identify potential biomarkers that correlate with immunohistochemistry assessment of tumor biopsy or with survival data. Functional analyses were performed to elucidate functional differences in PE proteins in malignant and benign samples. Results were integrated into a clinical risk prediction model to identify likely malignant cases. Sensitivity, specificity, and negative predictive value were calculated. Results: In total, 1689 individual proteins were identified by MS-based proteomics analysis of the 97 PE samples, of which 35 were diagnosed as malignant. A comparison between MPE and benign PE (BPE) identified 58 differential regulated proteins after correction of the p-values for multiple testing. Furthermore, functional analysis revealed an up-regulation of matrix intermediate filaments and cellular movement-related proteins. Additionally, gene ontology analysis identified the involvement of metabolic pathways such as glycolysis/gluconeogenesis, pyruvate metabolism and cysteine and methionine metabolism. Conclusion: This study demonstrated a partial least squares regression model with an area under the curve of 98 and an accuracy of 0.92 when evaluated on the holdout test data set. Furthermore, highly significant survival markers were identified (e.g., PSME1 with a log-rank of 1.68 × 10−6).

Suicide theory-guided natural language processing of clinical progress notes to improve prediction of veteran suicide risk: protocol for a mixed-method study

IntroductionThe state-of-the-art 3-step Theory of Suicide (3ST) describes why people consider suicide and who will act on their suicidal thoughts and attempt suicide. The central concepts of 3ST—psychological pain, hopelessness,...

Machine Learning-Based Prediction Models for Different Clinical Risks in Different Hospitals: Evaluation of Live Performance.

BackgroundMachine learning algorithms are currently used in a wide array of clinical domains to produce models that can predict clinical risk events. Most models are developed and evaluated with retrospective data, very few are evaluated in a clinical workflow, and even fewer report performances in different hospitals. In this study, we provide detailed evaluations of clinical risk prediction models in live clinical workflows for three different use cases in three different hospitals.ObjectiveThe main objective of this study was to evaluate clinical risk prediction models in live clinical workflows and compare their performance in these setting with their performance when using retrospective data. We also aimed at generalizing the results by applying our investigation to three different use cases in three different hospitals.MethodsWe trained clinical risk prediction models for three use cases (ie, delirium, sepsis, and acute kidney injury) in three different hospitals with retrospective data. We used machine learning and, specifically, deep learning to train models that were based on the Transformer model. The models were trained using a calibration tool that is common for all hospitals and use cases. The models had a common design but were calibrated using each hospital’s specific data. The models were deployed in these three hospitals and used in daily clinical practice. The predictions made by these models were logged and correlated with the diagnosis at discharge. We compared their performance with evaluations on retrospective data and conducted cross-hospital evaluations.ResultsThe performance of the prediction models with data from live clinical workflows was similar to the performance with retrospective data. The average value of the area under the receiver operating characteristic curve (AUROC) decreased slightly by 0.6 percentage points (from 94.8% to 94.2% at discharge). The cross-hospital evaluations exhibited severely reduced performance: the average AUROC decreased by 8 percentage points (from 94.2% to 86.3% at discharge), which indicates the importance of model calibration with data from the deployment hospital.ConclusionsCalibrating the prediction model with data from different deployment hospitals led to good performance in live settings. The performance degradation in the cross-hospital evaluation identified limitations in developing a generic model for different hospitals. Designing a generic process for model development to generate specialized prediction models for each hospital guarantees model performance in different hospitals.

Cardiovascular Disease Risk Modeling for Astronauts: Making the Leap From Earth to Space

NASA has recently completed several long-duration missions to the International Space Station and is solidifying plans to return to the Moon, with an eye toward Mars and beyond. As NASA pushes the boundaries of human space exploration, the hazards of spaceflight, including space radiation, levy an increasing burden on astronaut health and performance. The cardiovascular system may be especially vulnerable due to the combined impacts of space radiation exposure, lack of gravity, and other spaceflight hazards. On Earth, the risk for cardiovascular disease (CVD) following moderate to high radiation doses is well-established from clinical, environmental, and occupational exposures (largely from gamma- and x-rays). Less is known about CVD risks associated with high-energy charged ions found in space and increasingly used in radiotherapy applications on Earth, making this a critical area of investigation for occupational radiation protection. Assessing CVD risk is complicated by its multifactorial nature, where an individual's risk is strongly influenced by factors such as family history, blood pressure, and lipid profiles. These known risk factors provide the basis for development of a variety of clinical risk prediction models (CPMs) that inform the likelihood of medical outcomes over a defined period. These tools improve clinical decision-making, personalize care, and support primary prevention of CVD. They may also be useful for individualizing risk estimates for CVD following radiation exposure both in the clinic and in space. In this review, we summarize unique aspects of radiation risk assessment for astronauts, and we evaluate the most widely used CVD CPMs for their use in NASA radiation risk assessment applications. We describe a comprehensive dual-use risk assessment framework that supports both clinical care and operational management of space radiation health risks using quantitative metrics. This approach is a first step in using personalized medicine for radiation risk assessment to support safe and productive spaceflight and long-term quality of life for NASA astronauts.

Natural Language Processing of Radiology Reports to Detect Complications of Ischemic Stroke.

Abstraction of critical data from unstructured radiologic reports using natural language processing (NLP) is a powerful tool to automate the detection of important clinical features and enhance research efforts. We present a set of NLP approaches to identify critical findings in patients with acute ischemic stroke from radiology reports of computed tomography (CT) and magnetic resonance imaging (MRI). We trained machine learning classifiers to identify categorical outcomes of edema, midline shift (MLS), hemorrhagic transformation, and parenchymal hematoma, as well as rule-based systems (RBS) to identify intraventricular hemorrhage (IVH) and continuous MLS measurements within CT/MRI reports. Using a derivation cohort of 2289 reports from 550 individuals with acute middle cerebral artery territory ischemic strokes, we externally validated our models on reports from a separate institution as well as from patients with ischemic strokes in any vascular territory. In all data sets, a deep neural network with pretrained biomedical word embeddings (BioClinicalBERT) achieved the highest discrimination performance for binary prediction of edema (area under precision recall curve [AUPRC] > 0.94), MLS (AUPRC > 0.98), hemorrhagic conversion (AUPRC > 0.89), and parenchymal hematoma (AUPRC > 0.76). BioClinicalBERT outperformed lasso regression (p < 0.001) for all outcomes except parenchymal hematoma (p = 0.755). Tailored RBS for IVH and continuous MLS outperformed BioClinicalBERT (p < 0.001) and linear regression, respectively (p < 0.001). Our study demonstrates robust performance and external validity of a core NLP tool kit for identifying both categorical and continuous outcomes of ischemic stroke from unstructured radiographic text data. Medically tailored NLP methods have multiple important big data applications, including scalable electronic phenotyping, augmentation of clinical risk prediction models, and facilitation of automatic alert systems in the hospital setting.

Venous thromboembolism and risk stratification in hematological malignancies

Patients with hematologic malignancy have an increased risk of venous thromboembolism (VTE) compared to the general population. This risk is highest during the first months after diagnosis and subsequently decreases over time. The risk of VTE in leukemia ranges from less than 1% to almost 7% within the first 6-months of diagnosis, and is higher in patients with acute leukemia compared to chronic leukemia. The risk of VTE in lymphoma ranges from less than 1% to almost 20% in the first year of diagnosis, varying by lymphoma type. Risk is lowest in patients with indolent lymphoma and highest in those with aggressive lymphoma, including central nervous system (CNS) lymphoma. The risk of VTE in multiple myeloma is highest in the first 6-months of diagnosis and decreases over time. Despite incorporation of thromboprophylaxis strategies in many patients, 6-month incidence of VTE remains greater than 10%. Primary thromboprophylaxis has the potential to decrease risk of VTE in patients at high-risk. Clinical risk prediction models can quantify risk of VTE, thereby identifying those at high-risk. VTE risk prediction models are available for patients with leukemia, lymphoma and multiple myeloma. However, these models either require external validation or have room for improvement in VTE risk discrimination. Future efforts should focus in validation of available models, incorporation of biomarkers as predictors of VTE, and evaluation of the risk/benefit of thromboprophylaxis in high risk patients.

Predicting outcomes in patients with aortic stenosis using machine learning: the Aortic Stenosis Risk (ASteRisk) score

ObjectiveTo use echocardiographic and clinical features to develop an explainable clinical risk prediction model in patients with aortic stenosis (AS), including those with low-gradient AS (LGAS), using machine learning (ML).MethodsIn...