Outcome Prediction Models Research Articles

Mapping a Decade (2014-2024) of Research on Extracorporeal Membrane Oxygenation for Acute Respiratory Distress Syndrome: A Visual Analysis with CiteSpace and VOSviewer.

Acute Respiratory Distress Syndrome (ARDS) stands as a primary cause of mortality among critically ill patients. Extracorporeal Membrane Oxygenation (ECMO) is increasingly employed in the rescue therapy of ARDS patients. However, the current status of research in the field of ECMO-assisted ARDS remains unclear. This research aims to categorize and evaluate the literature regarding Extracorporeal Membrane Oxygenation (ECMO) support for Acute Respiratory Distress Syndrome (ARDS), offering a comprehensive analysis of bibliometric properties, research hotspots, and developmental trends within the domain of ECMO-assisted ARDS. A literature search was conducted for ECMO-assisted support for patients with ARDS in the Web of Science Core Collection (WoSCC) database from 2014 to 2024. We employed visualization tools such as CiteSpace and VOSviewer to explore and assess connections among nations, institutions, researchers, and co-cited journals, authors, references, and keywords. This study included 1739 publications. The United States leads in publication volume with Columbia University at the forefront of ECMO research. Intensive Care Medicine has been identified as the most cited journal in this field. Alain Combes from France stands out as a key contributor, particularly in his 2018 publication in the New England Journal of Medicine, which is the most cited work in the discipline. Furthermore, keyword analysis identified three distinct research phases: examining complications associated with ECMO therapy, exploring optimal strategies for mechanical ventilation under ECMO support, and compiling insights into the application of ECMO in treating COVID-19 patients and in the development of predictive models for patient outcomes. Using bibliometric visualization techniques, this study revealed significant progress in the use of ECMO for treating ARDS respiratory support, evaluated the impact of these findings, and outlined potential areas for future studies.

Construction of a machine learning-based prediction model for unfavorable discharge outcomes in patients with ischemic stroke

BackgroundIschemic stroke is a common and serious disease with economic and healthcare burdens. Predicting the unfavorable discharge outcome of patients is essential for formulating appropriate treatment strategies and providing personalized care. Therefore, this study aims to establish and validate a prediction model based on machine learning methods to accurately predict the discharge outcome of ischemic stroke patients, providing valuable information for clinical decision making. MethodsThe derivation data consisted of 964 patients from Guangdong Provincial People's Hospital and was used for training and internal validation. A favourable discharge outcome was defined as a National Institutes of Health Stroke Scale score of ≤1 or a decrease of ≥8 points compared to the admission score. A predictive model was created based on 88 medical characteristics gathered during the patient's initial admission, using nine machine learning algorithms. The model's predictive performance was compared using various evaluation metrics. The final model's feature importance was ranked and explained using the Shapley additive explanation method. FindingsThe random forest model demonstrated the greatest discriminative ability among the nine machine learning models. We created an interpretable random forest model by ranking and reducing the features based on their importance, which included eight features. In internal validations, the final model accurately predicted the discharge outcomes of ischemic stroke with AUC values of 0.903 and has been translated into a convenient tool to facilitate its utility in clinical settings. ConclusionsOur explainable ML model was not only successfully developed to accurately predict discharge outcomes in patients with ischemic stroke and it mitigated the concern of the “black-box” issue with an undirect interpretation of the ML technique.

The Pre-ECPR Score: Developing and Validating a Multivariable Prediction Model for Favorable Neurological Outcomes in Patients Undergoing Extracorporeal Cardiopulmonary Resuscitation

Extracorporeal cardiopulmonary resuscitation (ECPR) can save patients with refractory cardiac arrest; however, according to recent meta-analyses, only 20% of patients achieve favorable outcomes (Modified Rankin Scale 0-3). We aimed to develop and validate an ECPR prediction model to improve patient selection. Prognostic model development and internal validation study. Single-center study. All 120 normothermic ECPR patients treated at Sahlgrenska University Hospital between January 2010 and October 2021. None. Multivariable logistic regression was used to develop the PRognostic Evaluation of ECPR (Pre-ECPR) score. Model performance was assessed through the area under curve (AUC) and compared with the Extracorporeal Life Support Organization (ELSO) "Example of selection criteria for ECPR" for 1-year survival with favorable outcomes. The positive predictive value (PPV) was calculated. Favorable outcomes occurred in 27.5% of the patients. The Pre-ECPR score, incorporating age, no-flow/initial rhythm (a composite variable), total cardiac arrest time, signs of life, pupil dilation, regional cerebral oxygen saturation, arterial pH, and end-tidal CO2, demonstrated an AUC of 0.87 (95% confidence interval [CI] 0.77-0.93). In internal cross-validation, the AUC of 0.79 (95% CI 0.67-0.88) significantly outperformed the ELSO criteria AUC of 0.63 (95% CI 0.54-0.72, p = 0.012). Pre-ECPR score probabilities >6.4% showed 100% sensitivity and a PPV of 40.5% for favorable outcomes. The Pre-ECPR score combines multiple weighted predictors to provide a single balanced probability of favorable outcomes in ECPR patient selection. In cross-validation, it demonstrated significantly more favorable discriminatory performance than that of the ELSO criteria.

Predictive modeling based on functional connectivity of interictal scalp EEG for infantile epileptic spasms syndrome

ObjectiveThis study aims to delineate the electrophysiological variances between patients with infantile epileptic spasms syndrome (IESS) and healthy controls and to devise a predictive model for long-term seizure outcomes. MethodsThe cohort consisted of 30 individuals in the seizure-free group, 23 in the seizure-residual group, and 20 in the control group. We conducted a comprehensive analysis of pretreatment electroencephalography, including the relative power spectrum (rPS), weighted phase-lag index (wPLI), and network metrics. Follow-up EEGs at 2 years of age were also analyzed to elucidate physiological changes among groups. ResultsInfants in the seizure-residual group exhibited increased rPS in theta and alpha bands at IESS onset compared to the other groups (all p < 0.0001). The control group showed higher rPS in fast frequency bands, indicating potentially enhanced cognitive function. The seizure-free group presented increased wPLI across all frequency bands (all p < 0.0001). Our predictive model utilizing wPLI anticipated long-term outcomes at IESS onset (area under the curve 0.75). ConclusionOur findings demonstrated an initial “hypersynchronous state” in the seizure-free group, which was ameliorated following successful treatment. SignificanceThis study provides a predictive model utilizing functional connectivity and insights into the diverse electrophysiology observed among outcome groups of IESS.

204: PET-CT DenseNet outperforms advanced DL models for outcome prediction of oropharyngeal cancer

204: PET-CT DenseNet outperforms advanced DL models for outcome prediction of oropharyngeal cancer

UCN-Centric Prognostic Model for Predicting Overall Survival and Immune Response in Colorectal Cancer.

Colorectal cancer (CRC), a prevalent malignancy, ranks third in global incidence and second in mortality rates. Despite advances in screening methods such as colonoscopy, the accurate diagnosis of CRC remains challenging due to the absence of reliable biomarkers. This study aimed to develop a robust prognostic model for precise CRC outcome prediction. Employing weighted co-expression network analysis (WGCNA) and Cox regression analysis on data from The Cancer Genome Atlas (TCGA), we identified a panel of 12 genes strongly associated with patient survival. This gene panel facilitated accurate CRC outcome predictions, which is also validated via the external validation cohort GSE17536. We conducted further investigations into the key gene, urocortin (UCN), using single-cell transcriptomic data and immune infiltration analysis in CRC patients. Our results revealed a significant correlation between high UCN expression and the reduced prevalence of key immune cells, including B cells, CD4+ cytotoxic T cells, CD8+ T cells, and NKT cells. Functional experiments showed that UCN gene interference in the CRC cell lines significantly decreased cancer cell proliferation, underscoring UCN's role in intestinal immunity modulation. The UCN-centric prognostic model developed enhances prognosis prediction accuracy and offers critical insights for CRC diagnosis and therapeutic interventions.

Patient-reported outcome of lumbar decompression with instrumented fusion for low-grade spondylolisthesis: influence of pathology and baseline symptoms.

Low-grade isthmic and degenerative spondylolisthesis (DS) of the lumbar spine are distinct pathologies but both can be treated with lumbar decompression with fusion. In a very large cohort, we compared patient-reported outcome in relation to the pathology and chief complaint at baseline. This was a retrospective analysis using the EUROSPINE Spine Tango Registry. We included 582 patients (age 60 ± 15years; 65% female), divided into four groups based on two variables: type of spondylolisthesis and chief pain complaint (leg pain (LP) versus back pain). Patients completed the COMI preoperatively and up to 5years follow-up (FU), and rated global treatment outcome (GTO). Regression models were used to predict COMI-scores at FU. Pain scores and satisfaction ratings were analysed. All patients experienced pronounced reductions in COMI scores. Relative to the other groups, the DS-LP group showed between 5%and 11% greater COMI score reduction (p < 0.01 up to 2years'FU). This group also performed best with respect to pain outcomes and satisfaction. Long-term GTO was 93% at the 5year FU, compared with between 82% and 86% in the other groups. Regardless of the type of spondylolisthesis, all groups experienced an improvement in COMI score after surgery. Patients with DS and LP as their chief complaint appear to benefit more than other patients. These results are the first to show that the type of the spondylolisthesis and its chief complaint have an impact on surgical outcome. They will be informative for the consent process prior to surgery and can be used to build predictive models for individual outcome.

Development of a biomarker prediction model for post-trauma multiple organ failure/dysfunction syndrome based on the blood transcriptome

BackgroundMultiple organ failure/dysfunction syndrome (MOF/MODS) is a major cause of mortality and morbidity among severe trauma patients. Current clinical practices entail monitoring physiological measurements and applying clinical score systems to diagnose its onset. Instead, we aimed to develop an early prediction model for MOF outcome evaluated soon after traumatic injury by performing machine learning analysis of genome-wide transcriptome data from blood samples drawn within 24 h of traumatic injury. We then compared its performance to baseline injury severity scores and detection of infections.MethodsBuffy coat transcriptome and linked clinical datasets from blunt trauma patients from the Inflammation and the Host Response to Injury Study (“Glue Grant”) multi-center cohort were used. According to the inclusion/exclusion criteria, 141 adult (age ≥ 16 years old) blunt trauma patients (excluding penetrating) with early buffy coat (≤ 24 h since trauma injury) samples were analyzed, with 58 MOF-cases and 83 non-cases. We applied the Least Absolute Shrinkage and Selection Operator (LASSO) and eXtreme Gradient Boosting (XGBoost) algorithms to select features and develop models for MOF early outcome prediction.ResultsThe LASSO model included 18 transcripts (AUROC [95% CI]: 0.938 [0.890–0.987] (training) and 0.833 [0.699–0.967] (test)), and the XGBoost model included 41 transcripts (0.999 [0.997–1.000] (training) and 0.907 [0.816–0.998] (test)). There were 16 overlapping transcripts comparing the two panels (0.935 [0.884–0.985] (training) and 0.836 [0.703–0.968] (test)). The biomarker models notably outperformed models based on injury severity scores and sex, which we found to be significantly associated with MOF (APACHEII + sex—0.649 [0.537–0.762] (training) and 0.493 [0.301–0.685] (test); ISS + sex—0.630 [0.516–0.744] (training) and 0.482 [0.293–0.670] (test); NISS + sex—0.651 [0.540–0.763] (training) and 0.525 [0.335–0.714] (test)).ConclusionsThe accurate assessment of MOF from blood samples immediately after trauma is expected to aid in improving clinical decision-making and may contribute to reduced morbidity, mortality and healthcare costs. Moreover, understanding the molecular mechanisms involving the transcripts identified as important for MOF prediction may eventually aid in developing novel interventions.

Prognostic prediction models for treatment experienced people living with HIV: a protocol for systematic review and meta-analysis

IntroductionDespite the favourable efficacy of antiretroviral therapy (ART), HIV/AIDS continues to impose significant disease burdens worldwide. This study aims to systematically review published prognostic prediction models for survival outcomes of...

Automated Speech Analysis in Bipolar Disorder: The CALIBER Study Protocol and Preliminary Results.

Background: Bipolar disorder (BD) involves significant mood and energy shifts reflected in speech patterns. Detecting these patterns is crucial for diagnosis and monitoring, currently assessed subjectively. Advances in natural language processing offer opportunities to objectively analyze them. Aims: To (i) correlate speech features with manic-depressive symptom severity in BD, (ii) develop predictive models for diagnostic and treatment outcomes, and (iii) determine the most relevant speech features and tasks for these analyses. Methods: This naturalistic, observational study involved longitudinal audio recordings of BD patients at euthymia, during acute manic/depressive phases, and after-response. Patients participated in clinical evaluations, cognitive tasks, standard text readings, and storytelling. After automatic diarization and transcription, speech features, including acoustics, content, formal aspects, and emotionality, will be extracted. Statistical analyses will (i) correlate speech features with clinical scales, (ii) use lasso logistic regression to develop predictive models, and (iii) identify relevant speech features. Results: Audio recordings from 76 patients (24 manic, 21 depressed, 31 euthymic) were collected. The mean age was 46.0 ± 14.4 years, with 63.2% female. The mean YMRS score for manic patients was 22.9 ± 7.1, reducing to 5.3 ± 5.3 post-response. Depressed patients had a mean HDRS-17 score of 17.1 ± 4.4, decreasing to 3.3 ± 2.8 post-response. Euthymic patients had mean YMRS and HDRS-17 scores of 0.97 ± 1.4 and 3.9 ± 2.9, respectively. Following data pre-processing, including noise reduction and feature extraction, comprehensive statistical analyses will be conducted to explore correlations and develop predictive models. Conclusions: Automated speech analysis in BD could provide objective markers for psychopathological alterations, improving diagnosis, monitoring, and response prediction. This technology could identify subtle alterations, signaling early signs of relapse. Establishing standardized protocols is crucial for creating a global speech cohort, fostering collaboration, and advancing BD understanding.

Thoracic sarcopenia was a poor prognostic predictor in patients receiving immunotherapy retain-->for advanced non-small-cell retain-->lung cancer

Sarcopenia, as measured at the level of the third lumbar (L3) has been shown to predict the survival of cancer patients. However, many patients with advanced non-small cell lung cancer (NSCLC) do not undergo routine abdominal imaging. The objective of this study was to investigate the association of thoracic sarcopenia with survival outcomes among patients who underwent immunotherapy for NSCLC. In this retrospective study, patients who initiated immunotherapy for advanced NSCLC from 2019 to 2022 were enrolled. and detailed patient data were collected. Cross sectional skeletal muscle area was calculated at the fifth thoracic vertebra (T5) on pretreatment chest computed tomography (CT) scan. Gender-specific lowest quartile values was used to define sarcopenia. The risk factors were analyzed using Cox analyses. The log-rank test and the random survival forest (RSF) were used to compare progression free survival (PFS). The model's performance was assessed using calibration curve and the receiver operating characteristic curve (ROC). A total of 242 patients was included (discovery cohort n=194, validation cohort n=48). In the discovery cohort, patients with sarcopenia exhibited significantly poorer PFS (p<0.001) than patients without sarcopenia. Univariate cox regression revealed that sarcopenia, lung cancer stage, body mass index, smoking status, and neutrophil-to-lymphocyte ratio were predictors of poor PFS. A RSF model was constructed based on the aforementioned parameters, to evaluate the model's efficacy, the ROC curve was utilized. with an area under the curve for predicting 6-month PFS of 0.68 and for 12-month PFS of 0.69. The prediction models for survival outcomes built by the discovery cohort showed similar performance in the validation cohort. Sarcopenia at T5 is independent prognostic factors in patients who received immunotherapy for advanced NSCLC.

A prediction model for nonresponsive outcomes in critically ill patients with acute respiratory distress syndrome undergoing prone position ventilation: A retrospective cohort study

ObjectiveThis study aimed to develop a reliable and effective nomogram model to identify high-risk populations with non-response to prone position ventilation (PPV) in acute respiratory distress syndrome (ARDS) patients. MethodsThis retrospective cohort study included 175 patients with ARDS undergoing PPV. An improvement of ≥ 20 mmHg in the PaO2/FiO2 after the first PPV was defined as a 'response'. For the construction of the model, all patients were randomly assigned to the train and validation cohort according to 2:1. Multivariate logistic regression was useed to develop the nomogram. The area under the receiver operating characteristic curve (AUC), decision curve and calibration curve were assessed to evaluate the efficiency, clinical utility and calibration of the model. ResultsThe overall rate of non-response to PPV in ARDS patients was approximately 32.6 %. In the training cohort and validation cohort, the rate are 29.9 % and 34.5 % respectively. Murray score ≥ 2.5 (OR: 4.29), procalcitonin (PCT) ≥ 2 ng/mL (OR: 2.52), N-terminal pro-B-type natriuretic peptide (Nt-proBNP) ≥ 2000 pg/ml (OR: 2.44), and hemoglobin ≤ 90 g/L (OR: 2.39) were independently associated with the rate of non-response to PPV and combined in prediction model. The model demonstrated good predictive value with AUC of 0.817 and 0.828 in the train and validation cohort. Calibration curve showed good calibration and decision curve analysis indicated favorable clinical utility. ConclusionsThis study constructed a risk prediction model for non-response to PPV, which demonstrated good predictive value and clinical utility. Implications for clinical practiceEarly identification of prone position response in ARDS is essential for timely alternative treatments, improving patient prognosis and healthcare efficiency. The predictive model included representative indicators of patients with ARDS, encompassing parameters such as the acute lung injury (Murray score), cardiac function (Nt-proBNP), infectious status (PCT), and hemoglobin levels.

Machine learning models for outcome prediction in thrombectomy for large anterior vessel occlusion.

Predicting long-term functional outcomes shortly after a stroke is challenging, even for experienced neurologists. Therefore, we aimed to evaluate multiple machine learning models and the importance of clinical/radiological parameters to develop a model that balances minimal input data with reliable predictions of long-term functional independency. Our study utilized data from the German Stroke Registry on patients with large anterior vessel occlusion who underwent endovascular treatment. We trained seven machine learning models using 30 parameters from the first day postadmission to predict a modified Ranking Scale of 0-2 at 90 days poststroke. Model performance was assessed using a 20-fold cross-validation and one-sided Wilcoxon rank-sum tests. Key features were identified through backward feature selection. We included 7485 individuals with a median age of 75 years and a median NIHSS score at admission of 14 in our analysis. Our Deep Neural Network model demonstrated the best performance among all models including data from 24 h postadmission. Backward feature selection identified the seven most important features to be NIHSS after 24 h, age, modified Ranking Scale after 24 h, premorbid modified Ranking Scale, intracranial hemorrhage within 24 h, intravenous thrombolysis, and NIHSS at admission. Narrowing the Deep Neural Network model's input data to these features preserved the high performance with an AUC of 0.9 (CI: 0.89-0.91). Our Deep Neural Network model, trained on over 7000 patients, predicts 90-day functional independence using only seven clinical/radiological features from the first day postadmission, demonstrating both high accuracy and practicality for clinical implementation on stroke units.

Visual deterioration outcomes following optic pathway glioma treatment: a 12-year single institution retrospective study.

The surgical treatment of optic pathway gliomas (OPG) remains controversial, with visual outcomes often unpredictable. The present study explored surgical and clinical factors influencing visual acuity (VA) after OPG treatment and developed anatomical subtypes correlated with clinical symptoms. Children with OPG who underwent initial partial tumor resection at Beijing Tiantan Hospital from January 2011 to December 2022 were retrospectively analyzed. Multivariate logistic regression and random forest analyses were performed to identify risk factors for post-treatment VA deterioration and a decision tree model was created based on significant factors. A total of 140 patients were enrolled. Multivariate logistic regression analysis identified surgical approach and initial VA as independent predictors of post-treatment VA deterioration (P < 0.05). Surgical approach, initial VA, and extent of tumor resection were the most significant factors for risk assessment and were included in the decision tree model, with surgical approach as the most important "root" node. The model demonstrated good predictive performance, with area under the curve values of 0.75 and 0.66 for the training and test datasets, respectively. A simple anatomical classification was developed, which revealed clinical characteristic differences among OPG types. Meanwhile, a correlation analysis of post-treatment visual deterioration was performed for each of the three anatomical types. This study offers a predictive model for visual outcomes following initial tumor-reduction surgery in OPG patients, which may help in visual outcomes risk stratification. Additionally, the anatomical classification effectively indicates OPG growth direction, offering potential insights into clinical symptoms.

The strategic use of Big Data - A study protocol for a multicenter clinical trial testing if the use of the Swespine Dialogue Support alter outcomes in degenerative spine surgery

BackgroundPatients surgically treated for lumbar spinal stenosis or cervical radiculopathy report improvement in approximately two out of three cases. Advancements in Machine Learning and the utility of large datasets have enabled the development of prognostic prediction models within spine surgery. This trial investigates if the use of the postoperative outcome prediction model, the Dialogue Support, can alter patient-reported outcome and satisfaction compared to current practice.MethodsThis is a prospective, multicenter clinical trial. Patients referred to a spine clinic with cervical radiculopathy or lumbar spinal stenosis will be screened for eligibility. Participants will be assessed at baseline upon recruitment and at 12 months follow-up. The Dialogue Support will be used on all participants, and they will thereafter be placed into either a surgical or a non-surgical treatment arm, depending on the decision made between patient and surgeon. The surgical treatment group will be studied separately based on diagnosis of either cervical radiculopathy or lumbar spinal stenosis. Both the surgical and the non-surgical group will be compared to a retrospective matched control group retrieved from the Swespine register, on which the Dialogue Support has not been used. The primary outcome measure is global assessment regarding leg/arm pain in the surgical treatment group. Secondary outcome measures include patient satisfaction, Oswestry Disability Index (ODI), EQ-5D, and Numeric Rating Scales (NRS) for pain. In the non-surgical treatment group primary outcome measures are EQ-5D and mortality, as part of a selection bias analysis.DiscussionThe findings of this study may provide evidence on whether the use of an advanced digital decision tool can alter patient-reported outcomes after surgery.Trial registrationThe trial was retrospectively registered at ClinicalTrials.gov on April 17th, 2023, NCT05817747.Protocol version1.Trial designClinical multicenter trial.

Prediction of transforaminal epidural injection success in sciatica (POTEISS): a protocol for the development of a multivariable prediction model for outcome after transforaminal epidural steroid injection in patients with lumbar radicular pain due to disc herniation or stenosis

BackgroundTransforaminal epidural injections (TEI) can alleviate symptoms and help to maintain physical functioning and quality of life in patients with lumbar radicular pain. We aim to develop a prediction model for patient outcome after TEI in patients suffering from unilateral lumbar radicular pain due to lumbar disc herniation (LDH) or single-level spinal stenosis (LSS). The secondary aim is to estimate short-term patient outcome differences between LDH and LSS patients, the association between psychological variables and patient outcome, the rate of additional injections, surgery and complications, and to explore the short-term cost-effectiveness of TEI.MethodsThis study is designed as a multi-centre, observational, prospective cohort study in two large regional hospitals in the Netherlands. Patients diagnosed with unilateral lumbar radicular pain secondary to LDH or LSS and congruent with MRI findings, who are referred for TEI along usual care pathways, are eligible for study participation. A total of 388 patients with LDH or LSS will be included. A pre-defined set of demographic, clinical and radiological variables will be used as the predictors in the model. The primary outcome measure is the Numerical Rating Scale (NRS) for leg pain. Secondary outcome measures include back pain, physical functioning, perceived recovery, pain coping strategies, anxiety and depression and use of analgesics and physical therapy. Patients will be evaluated at baseline, 2 weeks and 6 weeks after treatment. NRS leg pain and Likert perceived recovery data will be used as the dependent variables in a generalized linear mixed model for prediction of TEI outcome, with internal validation of performance (explained variation) by bootstrap resampling. Cost-effectiveness for a period of 6 weeks prior to and after treatment will be performed with decision-analytic modelling.DiscussionPatients with severe lumbar radicular pain often request additional treatment when conservative care is insufficient. TEI can offer relief of symptoms. Currently, it is not possible to predict responsiveness to this treatment for individual patients. This study is designed to explore predictors that can differentiate between patients that will and will not have a positive outcome after TEI. This information may support treatment strategies for this patient group.Trial registrationThis study is registered at ClinicalTrials.gov database under registry number NCT04540068 on September 1, 2020.

Predictive Models of Long-Term Outcome in Patients with Moderate to Severe Traumatic Brain Injury are Biased Toward Mortality Prediction.

The prognostication of long-term functional outcomes remains challenging in patients with traumatic brain injury (TBI). Our aim was to demonstrate that intensive care unit (ICU) variables are not efficient to predict 6-month functional outcome in survivors with moderate to severe TBI (msTBI) but are mostly associated with mortality, which leads to a mortality bias for models predicting a composite outcome of mortality and severe disability. We analyzed the data from the multicenter randomized controlled Continuous Hyperosmolar Therapy in Traumatic Brain-Injured Patients trial and developed predictive models using machine learning methods and baseline characteristics and predictors collected during ICU stay. We compared our models' predictions of 6-month binary Glasgow Outcome Scale extended (GOS-E) score in all patients with msTBI (unfavorable GOS-E 1-4 vs. favorable GOS-E 5-8) with mortality (GOS-E 1 vs. GOS-E 2-8) and binary functional outcome in survivors with msTBI (severe disability GOS-E 2-4 vs. moderate to no disability GOS-E 5-8). We investigated the link between ICU variables and long-term functional outcomes in survivors with msTBI using predictive modeling and factor analysis of mixed data and validated our hypotheses on theInternational Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) model. Based on data from 370 patients with msTBI and classically used ICU variables, the prediction of the 6-month outcome in survivors was inefficient (mean area under the receiver operating characteristic 0.52). Using factor analysis of mixed data graph, we demonstrated that high-variance ICU variables were not associated with outcome in survivors with msTBI (p = 0.15 for dimension 1, p = 0.53 for dimension 2) but mostly with mortality (p < 0.001 for dimension 1), leading to a mortality bias for models predicting a composite outcome of mortality and severe disability. We finally identified this mortality bias in the IMPACT model. We demonstrated using machine learning-based predictive models that classically used ICU variables are strongly associated with mortality but not with 6-month outcome in survivors with msTBI, leading to a mortality bias when predicting a composite outcome of mortality and severe disability.

A large-scale reaction dataset of mechanistic pathways of organic reactions

Understanding organic reaction mechanisms is crucial for interpreting the formation of products at the atomic and electronic level, but still remains as a domain of knowledgeable experts. The lack of a large-scale dataset with chemically reasonable mechanistic sequences also hinders the development of reliable machine learning models to predict organic reactions based on mechanisms as human chemists do. Here, we present a high-quality and the first large-scale reaction dataset, denoted as mech-USPTO-31K, with chemically reasonable arrow-pushing diagrams validated by synthetic chemists, encompassing a wide spectrum of polar organic reaction mechanisms. We envision this dataset curated by applying a simple and flexible method that automatically generates reaction mechanisms using autonomously extracted reaction templates and expert-coded mechanistic templates to become an invaluable tool to develop future reaction outcome prediction models and discover new reactions.

Pitfalls of single-study external validation illustrated with a model predicting functional outcome after aneurysmal subarachnoid hemorrhage

BackgroundPrediction models are often externally validated with data from a single study or cohort. However, the interpretation of performance estimates obtained with single-study external validation is not as straightforward as assumed. We aimed to illustrate this by conducting a large number of external validations of a prediction model for functional outcome in subarachnoid hemorrhage (SAH) patients.MethodsWe used data from the Subarachnoid Hemorrhage International Trialists (SAHIT) data repository (n = 11,931, 14 studies) to refit the SAHIT model for predicting a dichotomous functional outcome (favorable versus unfavorable), with the (extended) Glasgow Outcome Scale or modified Rankin Scale score, at a minimum of three months after discharge. We performed leave-one-cluster-out cross-validation to mimic the process of multiple single-study external validations. Each study represented one cluster. In each of these validations, we assessed discrimination with Harrell’s c-statistic and calibration with calibration plots, the intercepts, and the slopes. We used random effects meta-analysis to obtain the (reference) mean performance estimates and between-study heterogeneity (I2-statistic). The influence of case-mix variation on discriminative performance was assessed with the model-based c-statistic and we fitted a “membership model” to obtain a gross estimate of transportability.ResultsAcross 14 single-study external validations, model performance was highly variable. The mean c-statistic was 0.74 (95%CI 0.70–0.78, range 0.52–0.84, I2 = 0.92), the mean intercept was -0.06 (95%CI -0.37–0.24, range -1.40–0.75, I2 = 0.97), and the mean slope was 0.96 (95%CI 0.78–1.13, range 0.53–1.31, I2 = 0.90). The decrease in discriminative performance was attributable to case-mix variation, between-study heterogeneity, or a combination of both. Incidentally, we observed poor generalizability or transportability of the model.ConclusionsWe demonstrate two potential pitfalls in the interpretation of model performance with single-study external validation. With single-study external validation. (1) model performance is highly variable and depends on the choice of validation data and (2) no insight is provided into generalizability or transportability of the model that is needed to guide local implementation. As such, a single single-study external validation can easily be misinterpreted and lead to a false appreciation of the clinical prediction model. Cross-validation is better equipped to address these pitfalls.

Abstract Mo030: Deep learning based multi-omics model for prediction of outcomes in HFpEF and HFmrEF

Introduction: Heart failure with preserved ejection fraction (HFpEF) presents a unique challenge in cardiology due to its diverse etiology and clinical presentation. Our study introduces a cutting-edge integrated deep learning model that combines multi-dimensional datasets, encompassing genetic variants from GWAS, cardiac structural and functional assessments from MRI, and electrophysiological data from ECGs, along with comorbidity profiles including CKD and LFT. Research Questions: Can a deep learning model that integrates multi-dimensional datasets accurately predict HFpEF risk? How does the inclusion of different types of data (genetic, MRI, comorbidity) affect the model's predictive accuracy? Goals: To develop a deep learning model that accurately predicts HFpEF risk To assess the impact of different types of data on the model's predictive accuracy. Methods: The model employs a graph convolutional multi-model deep learning approach to dissect the intricate interplay between genetic variants from GWAS, cardiac structural and functional assessments from MRI, electrophysiological data from ECGs, and comorbidity profiles including CKD and LFT. Results: The baseline genetic model achieved a PRS of 2.04 and an outcome accuracy of 81%. The incorporation of MRI features improved the outcome accuracy to 85.91%. The inclusion of comorbidities such as CKD and T2D resulted in the highest outcome accuracy of 88.57%. Discussion: The results suggest that the inclusion of MRI features and comorbidities improves the model's predictive accuracy. This model holds significant promise for clinical adoption and the advancement of personalized treatment strategies.