Risk Prediction Score Research Articles

Developing the Aesthetic Postoperative Complication Score (APeCS) for Detecting Major Morbidity in Facial Aesthetic Surgery.

Facial aesthetic surgery encompasses a variety of procedures with complication rates that are difficult to estimate due to a lack of published data. We sought to estimate major complication rates in patients undergoing facial aesthetic procedures and develop a risk assessment tool to stratify patients. We utilized the Tracking Operation and Outcomes for Plastic Surgeons (TOPS) database from 2003-2018. The analytic database included major facial aesthetic procedures. Univariate analysis and a backward stepwise multivariate regression model identified risk factors for major complications. Regression coefficients were utilized to create the score. Performance robustness was measured with area under receiver operating characteristic curves and sensitivity analyses. A total of 38,569 patients were identified. The major complication rate was 1.2% (460). The regression model identified risk factors including over 3 concomitant surgeries, BMI ≥25, ASA class ≥2, current or former smoker status, and age ≥45 as the variables fit for risk prediction (n = 13,004; area under curve: 0.68, standard error: 0.013, [0.62-0.67]). Each of the 5 variables counted for 1 point, except over 3 concomitant surgeries counting for 2, giving a score range from 0 to 6. Sensitivity analysis showed the cutoff point of ≥3 to best balance sensitivity and specificity, 58% and 66%, respectively. At this cutoff, 65% of cases were correctly classified as having a major complication. We developed an acceptable risk prediction score with a cutoff value of ≥3 associated with correctly classifying approximately 65% of those at risk for major morbidity when undergoing face and neck aesthetic surgery.

Development and validation of the Baseline Recurrence Risk in Cellulitis (BRRISC) score

Cellulitis is often treated with antibiotics for longer than recommended by guidelines. Prolonged therapy may reduce recurrence in certain patients, but it is not known which patients are at greatest risk. Our objective was to develop and temporally validate a risk prediction score to identify patients attending hospital with cellulitis at highest risk of recurrence. We included UK adult patients with cellulitis attending hospital in an electronic health records (EHR) study to identify demographic, comorbid, physiological, and laboratory factors predicting recurrence (before death) within 90 days, using multivariable logistic regression with backwards elimination in complete cases. A points-based risk score integerised model coefficients for selected predictors. Performance was assessed using the C-index in development and temporal validation samples. The final model included 4938 patients treated for median 8 days (IQR 6-11); 8.8% (n=436) experienced hospitalisation-associated recurrence. A risk score using eight variables (age, heart rate, urea, platelets, albumin, previous cellulitis, venous insufficiency, and liver disease) ranged from 0-15, with C-index=0.65 (95%CI:0.63-0.68). Categorising as low (score 0-1), medium (2-5) and high (6-15) risk, recurrence increased fourfold; 3.2% (95%CI: 2.3-4.4%), 9.7% (8.7-10.8%), and 16.6% (13.3-20.4%). Performance was maintained in the validation sample (C-index=0.63 (95%CI:0.58-0.67)). Among patients at high risk, four distinct clinical phenotypes were identified using hierarchical clustering 1) young, acutely unwell with liver disease; 2) comorbid with previous cellulitis and venous insufficiency; 3) chronic renal disease with severe renal impairment; and 4) acute severe illness, with substantial inflammatory responses. Risk of cellulitis recurrence varies markedly according to individual patient factors captured in the Baseline Recurrence Risk in Cellulitis (BRRISC) score. Further work is needed to optimise the score, considering baseline and treatment response variables not captured in EHR data, and establish the utility of risk-based approaches to guide optimal antibiotic duration.

High Risk of Stroke in Patients With Worsening Heart Failure, Reduced Ejection Fraction, Coronary Heart Disease and Sinus Rhythm: Risk Prediction Score Analysis From the COMMANDER-HF Trial

BackgroundPatients with heart failure with reduced ejection fraction (HFrEF) and sinus rhythm have a heightened risk of stroke. Whether anticoagulation benefits these patients is uncertain. In this post hoc analysis of the A Study to Assess the Effectiveness and Safety of Rivaroxaban in Reducing the Risk of Death, Myocardial Infarction, or Stroke in Participants with Heart Failure and Coronary Artery Disease Following an Episode of Decompensated Heart Failure (COMMANDER-HF) trial we evaluated how a previously validated risk model consisting of 3 variables (history of prior stroke, insulin-treated diabetes, and N-terminal pro-B-type natriuretic peptide level) would perform, compared with plasma d-dimer, for stroke prediction and estimation of the benefit of low-dose rivaroxaban. Methods and ResultsStroke risk and treatment effect were computed across risk score and plasma d-dimer tertiles. Risk score was available in 58% of the COMMANDER-HF population (n = 2928). Over a median follow-up of 512 days (range 342–747 days), 60 patients experienced a stroke (14.6 per 1000 patient-years). The risk model did not identify patients at higher risk of stroke and showed a low overall prognostic performance (C-index = 0.53). The effect of rivaroxaban on stroke was homogeneous across risk score tertiles (P-interaction = .67). Among patients in whom the risk score was estimated, d-dimer was available in 2343 (80%). d-dimer had an acceptable discrimination performance for stroke prediction (C-index = 0.66) and higher plasma d-dimer concentrations were associated with higher rates of stroke (ie, tertile 3 vs tertile 1, hazard ratio 3.65, 95% confidence interval 1.59–8.39, P = .002). Treatment with low-dose rivaroxaban reduced the incidence of stroke in patients at highest risk by d-dimer levels (ie, >515 ng/mL, hazard ratio 0.42, 95% confidence interval 0.18–0.95, P-interaction = .074), without any safety concerns. ConclusionsIn our analysis, plasma d-dimer concentrations performed better than a previously described 3-variable risk score for stroke prediction in patients with heart failure with reduced ejection fraction, a recent clinical worsening and sinus rhythm as enrolled in the COMMANDER-HF trial. In these patients, a raised plasma d-dimer concentration identified patients who might benefit most from rivaroxaban.

Postoperative Cerebrospinal Fluid-Related Complications After Posterior Fossa and Posterolateral Skull Base Surgeries: Development of a Predictive Model and Clinical Risk Score

ObjectivePostoperative pseudomeningocele (PMC) and cerebrospinal fluid (CSF) leak are common complications following posterior fossa and posterolateral skull base surgeries. We sought to (1) determine the rate of CSF-related complications and (2) develop a perioperative model and risk score to identify the highest-risk patients for these events. MethodsThis was a retrospective cohort of 450 patients undergoing posterior fossa and posterolateral skull base procedures from 2016-2020. Logistic regressions were performed for predictor selection on three pre-specified models: (1) a priori variables, (2) predictors selected by large effect sizes, and (3) predictors with p-value ≤ 0.100 on univariable analysis. A final model was created via elimination of nonsignificant predictors, and the integer-based Post-Operative CSF-related Complications (POCC) clinical risk score was derived. Internal validation was done using 10-fold cross-validation (cv) and bootstrapping with uniform shrinkage. Results115 (25.6%) patients developed PMC/CSF leak. Age > 55 (OR 0.560 [95% CI 0.328-0.954]), BMI > 30 (OR 1.88 [95% CI 1.14-3.10]), and postoperative CSF diversion (OR 2.85 [95% CI 1.64-5.00]) were associated with CSF leak/PMC. Model 2 was the most predictive (cvAUROC 0.690). The final risk score was devised using age, BMI class, dural repair technique, use of bone substitute, and duration of postoperative CSF diversion. The POCC Score performed well (cvAUROC 0.761) and was highly specific (96.1%). ConclusionsWe have created the first generalizable and predictive risk score to identify at risk patients for CSF-related complications. The POCC Score may improve surveillance, inform doctor-patient discussions regarding risks of surgery, and assist in perioperative management.

Single-cell analysis extracted CAFs-related genes to established online app to predict clinical outcome and radiotherapy prognosis of prostate cancer.

Cancer-associated fibroblasts (CAFs) play a significant role in regulating the clinical outcome and radiotherapy prognosis of prostate cancer (PCa). The aim of this study is to identify CAFs-related genes (CAFsRGs) using single-cell analysis and evaluate their potential for predicting the prognosis and radiotherapy prognosis in PCa. We acquire transcriptome and single-cell RNA sequencing (scRNA-seq) results of PCa and normal adjacent tissues from The GEO and TCGA databases. The "MCPcounter" and "EPIC" R packages were used to assess the infiltration level of CAFs and examine their correlation with PCa prognosis. ScRNA-seq and differential gene expression analyses were used to extract CAFsRGs. We also applied COX and LASSO analysis to further construct a risk score (CAFsRS) to assess biochemical recurrence-free survival (BRFS) and radiotherapy prognosis of PCa. The predictive efficacy of CAFsRS was evaluated by ROC curves and subgroup analysis. Finally, we integrated the CAFsRS gene signature with relevant clinical features to develop a nomogram, enhancing the predictive accuracy. The abundance of CAFs is associated with a poor prognosis of PCa patients. ScRNA-seq and differential gene expression analysis revealed 323 CAFsRGs. After COX and LASSO analysis, we obtained seven CAFsRGs with prognostic significance (PTGS2, FKBP10, ENG, CDH11, COL5A1, COL5A2, and SRD5A2). Additionally, we established a risk score model based on the training set (n = 257). The ROC curve was used to confirm the performance of CAFsRS (The AUC values for 1, 3 and 5-year survival were determined to be 0.732, 0.773, and 0.775, respectively.). The testing set (n = 129), GSE70770 set (n = 199) and GSE116918 set (n = 248) revealed that the model exhibited exceptional predictive performance. This was also confirmed by clinical subgroup analysis. The violin plot demonstrated a statistically significant disparity in the CAFs infiltrations between the high-risk and low-risk groups of CAFsRS. Further analysis confirmed that both CAFsRS and T stage were independent prognostic factors for PCa. The nomogram was then established and its excellent predictive performance was demonstrated through calibration and ROC curves. Finally, we developed an online prognostic prediction app ( https://sysu-symh-cafsnomogram.streamlit.app/ ) to facilitate the practical application of the nomogram. The prognostic prediction risk score model we constructed could accurately predict BRFS and radiotherapy prognosis PCa, which can provide new ideas for clinicians to develop personalized PCa treatment and follow-up programs.

Mortality Rates and a Clinical Predictive Model for the Elderly on Maintenance Hemodialysis: A Large Observational Cohort Study of 17,354 Asian Patients

Introduction: Mortality following hemodialysis initiation may influence the decision to initiate hemodialysis in elderly patients. Our objective is to demonstrate mortality following hemodialysis initiation in elderly patients (≥70 years) and to derive a prediction risk score based on clinical and laboratory indicators to determine risk of all-cause mortality in patients aged ≥80 years. Methods: We identified elderly patients (≥70 years) who initiated maintenance hemodialysis between January 2005 and December 2016 using data from the Thai Renal Replacement Therapy (TRT) registry. The mortality rate was determined based on age categories. A predictive risk score for all-cause mortality was created for 4,451 patients aged ≥80 years by using demographics, laboratory values, and interview-based parameters. Using a flexible parametric survival analysis, we predicted mortality 3 months, 6 months, 1 year, 5 years, and 10 years after hemodialysis initiation. Results: 17,354 patients (≥70 years) were included, mean age 76.9 ± 5.1 years, 46.5% male, and 6,309 (36.4%) died. Patients aged <80 years had a median survival time of 110.6 months. A 9-point risk score was developed to predict mortality in patients aged ≥80 years: age >85 years, male, body mass index <18.5 kg/m², hemoglobin <10.0 g/dL, albumin <3.5 g/dL, substantial assistance required in daily living (1 point each), and Karnofsky Performance Status (KPS) score <50 (3 points). C-statistic of 0.797 indicated high model discrimination. Internal validation demonstrated good agreement between observed and anticipated mortalities. Conclusions: Hemodialysis is appropriate for patients aged 70–80 years. A risk score for mortality in patients aged ≥80 years has been developed. The score is based on seven readily obtainable and evaluable clinical characteristics.

Predictors of early morbidity and mortality in newly diagnosed multiple myeloma: data from five randomized, controlled, phase III trials in 3700 patients

Early morbidity and mortality affect patient outcomes in multiple myeloma. Thus, we dissected the incidence and causes of morbidity/mortality during induction therapy (IT) for newly diagnosed multiple myeloma (NDMM), and developed/validated a predictive risk score. We evaluated 3700 transplant-eligible NDMM patients treated in 2005–2020 with novel agent-based triplet/quadruplet IT. Primary endpoints were severe infections, death, or a combination of both. Patients were divided in a training (n = 1333) and three validation cohorts (n = 2367). During IT, 11.8%, 1.8%, and 12.5% of patients in the training cohort experienced severe infections, death, or both, respectively. Four major, baseline risk factors for severe infection/death were identified: low platelet count (<150/nL), ISS III, higher WHO performance status (>1), and age (>60 years). A risk score (1 risk factor=1 point) stratified patients in low (39.5%; 0 points), intermediate (41.9%; 1 point), and high (18.6%; ≥2 points) risk. The risk for severe infection/death increased from 7.7% vs. 11.5% vs. 23.3% in the low- vs. intermediate- vs. high-risk groups (p < 0.001). The risk score was independently validated in three trials incorporating quadruplet IT with an anti-CD38 antibody. Our analyses established a robust and easy-to-use score to identify NDMM patients at risk of severe infection/death, covering the latest quadruplet induction therapies. Trial registrations: HOVON-65/GMMG-HD4: EudraCT No. 2004-000944-26. GMMG-MM5: EudraCT No. 2010-019173-16. GMMG-HD6: NCT02495922. EMN02/HOVON-95: NCT01208766. GMMG-HD7: NCT03617731.

Development and Validation of a Model for Postpancreatectomy Hemorrhage Risk

Postpancreatectomy hemorrhage (PPH) due to postoperative pancreatic fistula (POPF) is a life-threatening complication after pancreatoduodenectomy. However, there is no prediction tool for early identification of patients at high risk of late PPH. To develop and validate a prediction model for PPH. This retrospective prognostic study included consecutive patients with clinically relevant POPF who underwent pancreatoduodenectomy from January 1, 2009, to May 20, 2023, at the University Hospital Mannheim (derivation cohort), and from January 1, 2012, to May 31, 2022, at the University Hospital Dresden (validation cohort). Data analysis was performed from May 30 to July 29, 2023. Clinical and radiologic features of PPH. Accuracy of a predictive risk score of PPH. A multivariate prediction model-the hemorrhage risk score (HRS)-was established in the derivation cohort (n = 139) and validated in the validation cohort (n = 154). A total of 293 patients (187 [64%] men; median age, 69 [IQR, 60-76] years) were included. The HRS comprised 4 variables with associations: sentinel bleeding (odds ratio [OR], 35.10; 95% CI, 5.58-221.00; P < .001), drain fluid culture positive for Candida species (OR, 14.40; 95% CI, 2.24-92.20; P < .001), and radiologic proof of rim enhancement of (OR, 12.00; 95% CI, 2.08-69.50; P = .006) or gas within (OR, 12.10; 95% CI, 2.22-65.50; P = .004) a peripancreatic fluid collection. Two risk categories were identified with patients at low risk (0-1 points) and high risk (≥2 points) to develop PPH. Patients with PPH were predicted accurately in the derivation cohort (C index, 0.97) and validation cohort (C index 0.83). The need for more invasive PPH management (74% vs 34%; P < .001) and severe complications (49% vs 23%; P < .001) were more frequent in high-risk patients compared with low-risk patients. In this retrospective prognostic study, a robust prediction model for PPH was developed and validated. This tool may facilitate early identification of patients at high risk for PPH.

Discovery of a novel lipid metabolism-related gene signature to predict outcomes and the tumor immune microenvironment in gastric cancer by integrated analysis of single-cell and bulk RNA sequencing

Gastric cancer (GC) is a pressing global clinical issue, with few treatment options and a poor prognosis. The onset and spread of stomach cancer are significantly influenced by changes in lipid metabolism-related pathways. This study aimed to discover a predictive signature for GC using lipid metabolism-related genes (LMRGs) and examine its correlation with the tumor immune microenvironment (TIME). Transcriptome data and clinical information from patients with GC were collected from the TCGA and GEO databases. Data from GC samples were analyzed using both bulk RNA-seq and single-cell sequencing of RNA (scRNA-seq). To identify survival-related differentially expressed LMRGs (DE-LMRGs), differential expression and prognosis studies were carried out. We built a predictive signature using LASSO regression and tested it on the TCGA and GSE84437 datasets. In addition, the correlation of the prognostic signature with the TIME was comprehensively analyzed. In this study, we identified 258 DE-LMRGs in GC and further screened seven survival-related DE-LMRGs. The results of scRNA-seq identified 688 differentially expressed genes (DEGs) between the three branches. Two critical genes (GPX3 and NNMT) were identified using the above two gene groups. In addition, a predictive risk score that relies on GPX3 and NNMT was developed. Survival studies in both the TCGA and GEO datasets revealed that patients categorized to be at low danger had a significantly greater prognosis than those identified to be at high danger. Additionally, by employing calibration plots based on TCGA data, the study demonstrated the substantial predictive capacity of a prognostic nomogram, which incorporated a risk score along with various clinical factors. Within the high-risk group, there was a noticeable abundance of active natural killer (NK) cells, quiescent monocytes, macrophages, mast cells, and activated CD4 + T cells. In summary, a two-gene signature and a predictive nomogram have been developed, offering accurate prognostic predictions for general survival in GC patients. These findings have the potential to assist healthcare professionals in making informed medical decisions and providing personalized treatment approaches.

Automatic Head and Neck Tumor segmentation and outcome prediction relying on FDG-PET/CT images: Findings from the second edition of the HECKTOR challenge.

By focusing on metabolic and morphological tissue properties respectively, FluoroDeoxyGlucose (FDG)-Positron Emission Tomography (PET) and Computed Tomography (CT) modalities include complementary and synergistic information for cancerous lesion delineation and characterization (e.g. for outcome prediction), in addition to usual clinical variables. This is especially true in Head and Neck Cancer (HNC). The goal of the HEad and neCK TumOR segmentation and outcome prediction (HECKTOR) challenge was to develop and compare modern image analysis methods to best extract and leverage this information automatically. We present here the post-analysis of HECKTOR 2nd edition, at the 24th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2021. The scope of the challenge was substantially expanded compared to the first edition, by providing a larger population (adding patients from a new clinical center) and proposing an additional task to the challengers, namely the prediction of Progression-Free Survival (PFS). To this end, the participants were given access to a training set of 224 cases from 5 different centers, each with a pre-treatment FDG-PET/CT scan and clinical variables. Their methods were subsequently evaluated on a held-out test set of 101 cases from two centers. For the segmentation task (Task 1), the ranking was based on a Borda counting of their ranks according to two metrics: mean Dice Similarity Coefficient (DSC) and median Hausdorff Distance at 95th percentile (HD95). For the PFS prediction task, challengers could use the tumor contours provided by experts (Task 3) or rely on their own (Task 2). The ranking was obtained according to the Concordance index (C-index) calculated on the predicted risk scores. A total of 103 teams registered for the challenge, for a total of 448 submissions and 29 papers. The best method in the segmentation task obtained an average DSC of 0.759, and the best predictions of PFS obtained a C-index of 0.717 (without relying on the provided contours) and 0.698 (using the expert contours). An interesting finding was that best PFS predictions were reached by relying on DL approaches (with or without explicit tumor segmentation, 4 out of the 5 best ranked) compared to standard radiomics methods using handcrafted features extracted from delineated tumors, and by exploiting alternative tumor contours (automated and/or larger volumes encompassing surrounding tissues) rather than relying on the expert contours. This second edition of the challenge confirmed the promising performance of fully automated primary tumor delineation in PET/CT images of HNC patients, although there is still a margin for improvement in some difficult cases. For the first time, the prediction of outcome was also addressed and the best methods reached relatively good performance (C-index above 0.7). Both results constitute another step forward toward large-scale outcome prediction studies in HNC.

“A Longitudinal Study of KILLIP Class, Thrombolysis In Myocardial Infarction (TIMI), Global Registry Of Acute Coronary Events (GRACE), Risk Scores For Prediction Of Cardio Vascular Outcomes In Patients Presenting With Acute ST Elevation Myocardial Infarction in a Tertiary Care Centre.”

“A Longitudinal Study of KILLIP Class, Thrombolysis In Myocardial Infarction (TIMI), Global Registry Of Acute Coronary Events (GRACE), Risk Scores For Prediction Of Cardio Vascular Outcomes In Patients Presenting With Acute ST Elevation Myocardial Infarction in a Tertiary Care Centre.”

Risk Stratification of Patients with RUNX1-mutated Acute Myeloid Leukemia

Mutations in RUNX1 ( RUNX1mut) occur in ~15% of intensively treated AML cases. RUNX1mut have no specific hotspot and various types of alteration are observed. The European LeukemiaNet (ELN) risk stratification assigns adverse prognosis to RUNX1mut if they do not co-occur with favorable-risk genotypes. Considering the biological complexity of RUNX1 it seems implausible that all alterations have similar consequences. Using clinical and genetic variables, we developed a prognostic risk stratification model for ELN adverse-risk RUNX1mut AML patients. We combined data from five groups, totaling 609 patients with intensively treated RUNX1mut AML, to develop the model. Our training set included 448 patients treated on trials of the AML Cooperative Group (AMLCG; Herold et al, Leukemia, 2020; (n=178)), AML Study Group (Gerstung et al, NEJM, 2016; (n=116)) and Study Alliance Leukemia (n=154). Patients from the Munich Leukemia Laboratory (MLL; (n=107)) and of the Alliance group (trials NCT00048958, NCT00899223, NCT00900224; Support: U10CA180821, U10CA180882, U24CA196171; https://acknowledgments.alliancefound.org; (n=54)) served as independent validation cohorts. Additionally, 955 patients without RUNX1mut treated on AMLCG trials served as controls. Patients with t(15;17), prior treatment, or RUNX1mut with co-occurring favorable-risk genotypes according to ELN 2017 were excluded. Differences between RUNX1mut patients and controls were investigated using univariate logistic regression. Testing was performed using likelihood ratio tests and adjusted for study group. Univariate analyses were adjusted for multiple testing using the Benjamini-Hochberg procedure. We obtained risk prediction models using multivariate Cox regression. Missing values were imputed using the missForest approach. Model selection was performed using forward selection based on the Bayesian information criterion. Cut-offs were based on the 25 th- 50 th- and 75 th-percentile score values obtained in the training data. Performance was evaluated using Kaplan-Meier curves. For internal validation Harrell's C index was estimated using cross-study validation. For external validation, the final risk prediction models were separately applied to the external datasets. RUNX1mut were more common in older, male patients and sAML (table). White blood cells, lactate dehydrogenase and bone marrow blasts were lower in RUNX1mut patients. Mutations in several myelodysplasia-related genes were enriched in RUNX1mut patients ( ASXL1, BCOR, BCORL1, EZH2, KMT2A, PHF6, and STAG2, SF3B1, SRSF2, and U2AF1), whereas DNMT3A, NPM1 and FLT3 were more frequently altered in controls. A strong association with mutations of the splicing factor complex was identified (49% vs. 13%, p&amp;lt;0.0001). Contradicting previous reports, we found no association with IDH mutations. The risk prediction score we obtained for OS is as follows: 0.03054 x age (y) + 0.74996 x adverse MRC + 0.43779 x FLT3-ITD + 0.00317 x WBC count (10^9/L) - 0.00158 x platelet count (10^9/L) + 0.37401 x NRAS-mutation, where the obtained cut-off values are: &amp;lt;1.592 (low risk); 1.592 to 2.303 (moderate risk) &amp;gt;2.303 (high risk). Binary variables are coded as 0 or 1. Harrell's C index estimated using internal validation was 0.6. Kaplan-Meier curves estimated using the training set suggest strong differences in survival between the risk categories. Median overall survival (OS) was 2.5, 1.0 and 0.6 years for low-, intermediate-, and high-risk. External validation using the Alliance cohort shows similar results (Figure). Results obtained for the MLL cohort show smaller differences. However, we still observe clear separation of risk groups with a significant difference between low- and high-risk groups (adjusted p-value: 0.0266). Scores for relapse-free survival (RFS) as well as for OS and RFS censored for allogeneic transplant show similar results. We analyzed a large collection of intensively treated RUNX1mut AML patients and observed heterogenous outcomes that could be predicted by applying few variables (age, MRC-score, FLT3/NRAS mutation-status, and WBC/platelet count). The OS of RUNX1mut high-risk patients is discouraging, highlighting the unmet need of these patients. In addition, our work demonstrates that ELN risk groups can be further stratified and that integrated approaches using routinely available variables can further advance risk prediction.

Discriminating Risk of Anticoagulant-Related Bleeding in Ambulatory Cancer Patients on Thromboprophylaxis

Introduction: Among patients taking anticoagulants, those with cancer have double the risk of major bleeding (MB). Because they also have a risk of VTE, two large randomized-control trials, AVERT and CASSINI, demonstrated effectiveness of low-dose direct oral anticoagulants (DOACs) for primary prevention of VTE in patients with cancer at high-risk for VTE based on a validated prediction model. Despite demonstrated efficacy, there is limited uptake of primary thromboprophylaxis in clinical practice. This apprehension may, in part, be due to concerns about the risk of bleeding. We aimed to determine the predictive performance of available risk prediction scores for anticoagulant-related bleeding in cancer patients randomized to thromboprophylaxis during participation in the AVERT randomized trial. Methods: Using patients in the modified intent-to-treat analysis in the AVERT randomized controlled trial, we conducted a post-hoc analysis to assess the performance of three available risk prediction scores for anticoagulant-related bleeding in 288 patients randomized to apixaban. We selected scores based on availability of candidate variables: RIETE, VTE Bleed, and Kuijer et al. scores. The primary outcome of interest was development of a MB or clinically relevant non-major bleed (CRNMB). A second analysis was conducted limiting the bleeding event to either a MB or a CRNMB that required a medical intervention. Each bleeding risk score was applied to the cohort with clinical point assignments as in the original papers (Table) . Using a Fine and Gray competing risk model, we tested the association between each score and development of first bleed following anticoagulant prescription. Patients were censored after a MB. The performance of each model was evaluated using time-dependent ROC (model discrimination) and Brier score (model calibration/discrimination). All analyses were conducted using R (4.2.3) and SAS (9.4) statistical software. Results: Between 2014 and 2018, 574 patients were randomized with 563 receiving at least one dose of study medication. A total of 288 patients received apixaban 2.5mg twice daily and 275 received placebo. The mean age was 61 years and mostly women (58.2%). Frequent cancers included: gynecologic (25.8%), lymphoma (25.3%), and pancreatic cancer (13.6%). The median duration of follow-up for the cohort was 183 days with adherence rates of 83.6% and 84.1% with apixaban and placebo respectively. There were 10 MBs and 18 CRNMBs when censoring patients at the time of first bleed. Of the 18 CRNMB events, 6 required a medical intervention. There was no significant association between each 1-point increase in the Kuijer et al. score and risk of MB+CRNMB (subdistribution hazard ratio (sHR) 0.97, p 0.85). However, for each 1-point increase in score in RIETE and VTE BLEED, there was a 72% (sHR 1.72, p &amp;lt;0.0001) and 26% (sHR 1.26, p 0.05) increase in risk of MB+CRNMB respectively. Discrimination of each score at 180 days was 0.69 for RIETE, 0.50 for Kuijer et al., and 0.61 for VTE BLEED. Results of calibration for each score was similar at 180 days was 0.056 for RIETE, 0.059 for Kuijer et al., and 0.058 for VTE BLEED. When limiting analyses to MB+CRNMB that required medical intervention, results remained consistent for the RIETE and Kuijer et al. scores, with a loss of association for the VTE BLEED score. Conclusions: Of the scores analyzed, the RIETE score performed best with moderate discrimination. There is a need to improve the performance of these scores. The ability to quantify risk of anticoagulant-related bleeding in patients who are candidates for primary thromboprophylaxis can allow providers and patients to make informed decisions about primary thromboprophylaxis based on the risk of VTE versus risk of anticoagulant-related bleeding.

Plasma Proteomic Signature Predicts Risk of Myeloid Neoplasm

Introduction The presence of clonally expanded mutant hematopoietic stem cells, clonal hematopoiesis (CH) is known to predispose to hematologic malignancy, particularly myeloid neoplasms (MN). However, the overall risk of developing MNs among individuals with CH is low (&amp;lt;1% per year). Limitations in our ability to predict who with CH will progress to MN is a barrier to prevention strategies. The human proteome reflects the downstream impact of complex interactions between genetic and epigenetic regulation of biologic systems. We hypothesized that plasma proteomic markers might be associated with MN risk and if so could inform our understanding of the mechanisms promoting the progression of CH to MN. Methods Among 46,237 individuals in the UK Biobank (UKB) with whole exome sequencing data and single nucleotide polymorphism (SNP) array profiling of the blood and plasma proteomic profiling using the Olink Explore platform (1,463 unique proteins), we jointly characterized CH and plasma proteomic profile. Cox proportional hazard models were used to assess the association between the level of each plasma protein and risk of MN; while controlling for age, sex, genetic ancestry, smoking status, blood counts, and sample collection year. The relative contribution of clinical, CH, and proteomics features in prediction of incident MN was assessed using lasso regression with 100-fold cross validation. We performed protein quantitative trait loci (pQTL) mapping to identify independent germline autosomal genetic loci that significantly regulate plasma proteomic levels (p &amp;lt;5x10 -8). We calculated genetically predicted risk scores for plasma protein levels as the as the sum of the effect estimates from linear regression analysis with additive genetic effect, multiplied by the number of risk alleles for each locus. As a validation cohort, we used an independent set of 427,614 individuals in the UKB to test for an association between genetically predicted levels of proteomic markers and MN risk. Results A total of 115 participants developed MN at least 3 months after study entry (median 83 months). As expected, participants who developed MN (cases) had a higher frequency of CH at the time of study entry compared to those free of hematologic malignancy (controls); 42 versus 6% for SNV/Indels and 23 versus 3% for mosaic copy number events in cases compared to controls respectively. We identified 115 proteins significantly associated (qFDR &amp;lt;0.05) with incident MN (Figure 1). These include known regulators of the adaptive immune system, innate immune system, leukocyte trafficking, hematopoietic stem cell self-renewal, and platelet activation. Given that CH is a genetic precursor to MN, we assessed the relationship between plasma proteomic markers of MN and CH. Among the 115 plasma proteins that were significantly associated with MN, 35 were significantly (qFDR &amp;lt;0.05) and 59 nominally (p &amp;lt;0.05) associated with CH. After adjusting for CH, the strength of the association between plasma protein markers and MN risk was generally reduced (mean 7%), but remained significant for many (N=44). We identified a total of 79 plasma proteins that were associated with the presence of CH at the time of study entry, 34 of which overlapped with MN risk proteins. Inclusion of proteomics features in a lasso regression model (AUC = 0.85) improved prediction of incident MN beyond clinical factors and CH (AUC = 0.80; Figure 2). There were 1,282 independent genetic loci that were significantly associated with 115 proteomic markers of MN risk. On average, genetic risk scores captured 12% of the total variance of plasma protein levels (range 0.1-64%). In an independent cohort of 427,614 UKB participants (1,099 of whom developed MN), we found that of the 110 plasma protein markers with a genetic risk score R 2 &amp;gt;1%, 12 were significantly associated with MN (p &amp;lt;0.05). Of the remaining, 63 were directionally consistent. Conclusions We show that plasma proteomic markers predict risk of MN. Some, but not all of the impact of plasma proteomic profile on MN risk is mediated by baseline CH. Inclusion of proteomic markers improves MN risk prediction beyond clinical and CH features. Genetic risk score analysis provides orthogonal validation of the plasma proteomic signature for MN risk. These data highlight the promise of protein and genetic biomarker integration for early diagnosis of MN and further support the role of immune cell regulation in early leukemogenesis.

425. Characterization and predictive risk scoring of long COVID in a South Indian cohort after breakthrough COVID infection; a prospective single centre study

Abstract Background The majority of existing literature on long COVID has been focused on hospitalised patients. However, in the real-world setting of the post-vaccination era where breakthrough infections are the norm, it becomes a priority to assess the clinical profile of long COVID symptoms among such individuals. Through our study, we aim to describe the incidence, characterise and stratify the risk of developing long COVID breakthrough infections. Methods This prospective observational study included adult patients with breakthrough COVID-19 infections diagnosed at a tertiary hospital in India. Post-COVID symptoms at weeks 2, 6 and 12 after testing negative were extracted using a questionnaire after which multi-variate analysis was done. Results Out of 414 patients analysed, 164 reported post-COVID symptoms beyond 6 weeks of the infection. Univariate analysis showed that the presence of long COVID was found to be significantly higher among patients above 65 years of age at 31.3% in comparison to the long COVID presence at 29.4% among those below age 65 (p &amp;lt; 0.001). Systemic hypertension was also significantly associated with the presence of long COVID (72.3%), along with bronchial asthma (68.8%) with a p-value of &amp;lt; 0.0001 in both instances. Backwards selection was used leading to a reduced model consisting of age OR 1.053, 95% CI 0.097–1.07), p &amp;lt; 0.001), hypertension (OR 2.59, 95% CI 1.46–4.59, p = 0.001) and bronchial asthma (OR 3.7176, 95% CI 1.24–11.12, p = 0.018) to be significant predictors of long COVID incidence. These were used to develop a propensity score (C statistics – 80.9%). 1A: Long COVID symptomatic profile at weeks 2, 6 and 12 weeks 1B: Organ system wise distribution of long COVID symptoms at 2, 6 and 12 weeks Conclusion A significant presence of long COVID at 12 weeks among non-hospitalised breakthrough infections calls for a series of review check-ups for the early detection of long-term complications. The proposed predictive risk scoring based on significant risk factors may assist clinicians in identifying patients at risk of developing long COVID, leading to appropriate, individualised management. Table 4 Figure 3A &amp; 3B 3A: Logistic regression model for predicting long COVID incidence 3B: Clinical prediction nomogram to depict risk of long COVID incidence Figure 4A &amp; 4B 4A: Association of propensity risk score for Long COVID incidence with symptom burden 4B: Association of propensity risk score for Long COVID incidence with functional status Disclosures All Authors: No reported disclosures

Adding the value of the Charlson Comorbidity Index to the GRACE score for mortality prediction in acute coronary syndromes.

Scarce and conflicting data still exist about the role of the Charlson Comorbidity Index (CCI) when added to the traditional Global Registry of Acute Coronary Events (GRACE) risk score for outcome prediction in patients with acute coronary syndrome (ACS). All consecutive admissions due to ACS, from 1 January 2018 to 31 December 2020 were retrospectively reviewed from an internal database of a tertiary cardiac center in Salerno (Italy). Logistic and Cox proportional regression analyses were performed in order to assess the contribution of the CCI on 30-day and long-term mortality. The CCI adding value to the GRACE score was analyzed with several measures of improvement in discrimination: increase in the area under the receiver-operating characteristic curve (AUC), the integrated discrimination improvement (IDI), and the categorical and continuous net reclassification improvement (cNRI) more than 0. Robustness of the results was assessed through an internal validation procedure with bootstrapping. One thousand three hundred and ten patients were identified. The median age was 68 (58-78) years. One hundred and twenty (9.2%) and 113 (9.5%) deaths occurred, respectively, during the first 30 days from admission and during long-term follow-up (median follow-up time: 13 months; interquartile range: 9-24). After multivariate regression analysis, the CCI was not associated with short-term mortality, while it was significantly and independently associated with long-term mortality along with the GRACE score (hazard ratio: 1.34, 95% confidence interval: 1.22-1.47; P < 0.001). An additive effect of CCI with the GRACE risk score was observed in predicting long-term mortality: AUC from 0.768 to 0.819 ( P = 0.003), category-based NRI: 0.215, cNRI>0: 0.669 ( P < 0.001), IDI: 0.066 ( P < 0.001). The CCI is a predictor of long-term mortality and improves risk stratification of patients with ACS over the GRACE risk score.

Using Multi-Modal Electronic Health Record Data for the Development and Validation of Risk Prediction Models for Long COVID Using the Super Learner Algorithm

Background: Post-Acute Sequelae of COVID-19 (PASC) have emerged as a global public health and healthcare challenge. This study aimed to uncover predictive factors for PASC from multi-modal data to develop a predictive model for PASC diagnoses. Methods: We analyzed electronic health records from 92,301 COVID-19 patients, covering medical phenotypes, medications, and lab results. We used a Super Learner-based prediction approach to identify predictive factors. We integrated the model outputs into individual and composite risk scores and evaluated their predictive performance. Results: Our analysis identified several factors predictive of diagnoses of PASC, including being overweight/obese and the use of HMG CoA reductase inhibitors prior to COVID-19 infection, and respiratory system symptoms during COVID-19 infection. We developed a composite risk score with a moderate discriminatory ability for PASC (covariate-adjusted AUC (95% confidence interval): 0.66 (0.63, 0.69)) by combining the risk scores based on phenotype and medication records. The combined risk score could identify 10% of individuals with a 2.2-fold increased risk for PASC. Conclusions: We identified several factors predictive of diagnoses of PASC and integrated the information into a composite risk score for PASC prediction, which could contribute to the identification of individuals at higher risk for PASC and inform preventive efforts.

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.

Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder

Biobanks often contain several phenotypes relevant to diseases such as major depressive disorder (MDD), with partly distinct genetic architectures. Researchers face complex tradeoffs between shallow (large sample size, low specificity/sensitivity) and deep (small sample size, high specificity/sensitivity) phenotypes, and the optimal choices are often unclear. Here we propose to integrate these phenotypes to combine the benefits of each. We use phenotype imputation to integrate information across hundreds of MDD-relevant phenotypes, which significantly increases genome-wide association study (GWAS) power and polygenic risk score (PRS) prediction accuracy of the deepest available MDD phenotype in UK Biobank, LifetimeMDD. We demonstrate that imputation preserves specificity in its genetic architecture using a novel PRS-based pleiotropy metric. We further find that integration via summary statistics also enhances GWAS power and PRS predictions, but can introduce nonspecific genetic effects depending on input. Our work provides a simple and scalable approach to improve genetic studies in large biobanks by integrating shallow and deep phenotypes.

Validation of risk scores for prediction of severe pneumonia in kidney transplant recipients hospitalized with community-acquired pneumonia.

Risk scores for community-acquired pneumonia (CAP) are widely used for standardized assessment in immunocompetent patients and to identify patients at risk for severe pneumonia and death. In immunocompromised patients, the prognostic value of pneumonia-specific risk scores seems to be reduced, but evidence is limited. The value of different pneumonia risk scores in kidney transplant recipients (KTR) is not known. Therefore, we retrospectively analyzed 310 first CAP episodes after kidney transplantation in 310 KTR.We assessed clinical outcomes and validated eight different risk scores (CRB-65, CURB-65, DS-CRB-65, qSOFA, SOFA, PSI, IDSA/ATS minor criteria, NEWS-2) for the prognosis of severe pneumonia and in-hospital mortality. Risk scores were assessed up to 48h after admission, but always before an endpoint occurred. Multiple imputation was performed to handle missing values. In total, 16 out of 310 patients (5.2%) died, and 48 (15.5%) developed severe pneumonia. Based on ROC analysis, sequential organ failure assessment (SOFA) and national early warning score 2 (NEWS-2) performed best, predicting severe pneumonia with AUC of 0.823 (0.747-0.880) and 0.784 (0.691-0.855), respectively. SOFA and NEWS-2 are best suited to identify KTR at risk for the development of severe CAP. In contrast to immunocompetent patients, CRB-65 should not be used to guide outpatient treatment in KTR, since there is a 7% risk for the development of severe pneumonia even in patients with a score of zero.