Outcome Prediction Performance Research Articles

Magnetic Resonance-Guided Cancer Therapy Radiomics and Machine Learning Models for Response Prediction.

Magnetic resonance imaging (MRI) is known for its accurate soft tissue delineation of tumors and normal tissues. This development has significantly impacted the imaging and treatment of cancers. Radiomics is the process of extracting high-dimensional features from medical images. Several studies have shown that these extracted features may be used to build machine-learning models for the prediction of treatment outcomes of cancer patients. Various feature selection techniques and machine models interrogate the relevant radiomics features for predicting cancer treatment outcomes. This study aims to provide an overview of MRI radiomics features used in predicting clinical treatment outcomes with machine learning techniques. The review includes examples from different disease sites. It will also discuss the impact of magnetic field strength, sample size, and other characteristics on outcome prediction performance.

Developing Deep LSTMs With Later Temporal Attention for Predicting COVID-19 Severity, Clinical Outcome, and Antibody Level by Screening Serological Indicators Over Time.

The clinical course of COVID-19, as well as the immunological reaction, is notable for its extreme variability. Identifying the main associated factors might help understand the disease progression and physiological status of COVID-19 patients. The dynamic changes of the antibody against Spike protein are crucial for understanding the immune response. This work explores a temporal attention (TA) mechanism of deep learning to predict COVID-19 disease severity, clinical outcomes, and Spike antibody levels by screening serological indicators over time. We use feature selection techniques to filter feature subsets that are highly correlated with the target. The specific deep Long Short-Term Memory (LSTM) models are employed to capture the dynamic changes of disease severity, clinical outcome, and Spike antibody level. We also propose deep LSTMs with a TA mechanism to emphasize the later blood test records because later records often attract more attention from doctors. Risk factors highly correlated with COVID-19 are revealed. LSTM achieves the highest classification accuracy for disease severity prediction. Temporal Attention Long Short-Term Memory (TA-LSTM) achieves the best performance for clinical outcome prediction. For Spike antibody level prediction, LSTM achieves the best permanence. The experimental results demonstrate the effectiveness of the proposed models. The proposed models can provide a computer-aided medical diagnostics system by simply using time series of serological indicators.

Clinical Validation of Mathematically Derived Early Tumor Dynamics for Solid Tumors in Response to Durvalumab.

Early prediction of response to immunotherapy may help guide patient management by identifying resistance to treatment and allowing adaptation of therapies. This analysis evaluated a mathematical model of response to immunotherapy that provides patient-specific prediction of outcome using the initial change in tumor size/burden from baseline to the first follow-up visit on standard imaging scans. We applied the model to 600 patients with advanced solid tumors who received durvalumab in Study 1108, a phase I/II trial, and compared outcome prediction performance versus size-based criteria with RECIST version 1.1 best overall response (BOR), baseline circulating tumor (ct)DNA level, and other clinical/pathologic predictors of immunotherapy response. In multiple solid tumors, the mathematical parameter representing net tumor growth rate at the first on-treatment computed tomography (CT) scan assessed around 6 weeks after starting durvalumab (α1) had a concordance index to predict overall survival (OS) of 0.66-0.77 on multivariate analyses. This measurement of early tumor dynamics significantly improved multivariate OS models that included standard RECIST v1.1 criteria, baseline ctDNA levels, and other clinical/pathologic factors in predicting OS. Furthermore, α1 was assessed consistently at the first on-treatment CT scan, whereas all traditional RECIST BOR groups were confirmed only after this time. These results support further exploring α1 as an integral biomarker of response to immunotherapy. This biomarker may be predictive of further benefit and can be assessed before RECIST response groups can be assigned, potentially providing an opportunity to personalize oncologic management.

Atrial and ventricular strain using cardiovascular magnetic resonance in the prediction of outcomes of pericarditis patients: a pilot study

ObjectiveOur study aimed to explore with cardiovascular magnetic resonance (CMR) the impact of left atrial (LA) and left ventricular (LV) myocardial strain in patients with acute pericarditis and to investigate their possible prognostic significance in adverse outcomes.MethodThis retrospective study performed CMR scans in 36 consecutive patients with acute pericarditis (24 males, age 52 [23–52]). The primary endpoint was the combination of recurrent pericarditis, constrictive pericarditis, and surgery for pericardial diseases defined as pericardial events. Atrial and ventricular strain function were performed on conventional cine SSFP sequences.ResultsAfter a median follow-up time of 16 months (interquartile range [13–24]), 12 patients with acute pericarditis reached the primary endpoint. In multivariable Cox regression analysis, LA reservoir and LA conduit strain parameters were all independent determinants of adverse pericardial diseases. Conversely, LV myocardial strain parameters did not remain an independent predictor of outcome. With receiving operating characteristics curve analysis, LA conduit and reservoir strain showed excellent predictive performance (area under the curve of 0.914 and 0.895, respectively) for outcome prediction at 12 months.ConclusionLA reservoir and conduit mechanisms on CMR are independently associated with a higher risk of adverse pericardial events. Including atrial strain parameters in the management of acute pericarditis may improve risk stratification.Clinical relevance statementAtrial strain could be a suitable non-invasive and non-contrast cardiovascular magnetic resonance parameter for predicting adverse pericardial complications in patients with acute pericarditis.Key Points• Myocardial strain is a well-validated CMR parameter for risk stratification in cardiovascular diseases.• LA reservoir and conduit functions are significantly associated with adverse pericardial events.• Atrial strain may serve as an additional non-contrast CMR parameter for stratifying patients with acute pericarditis.

Use of Pre-Transplant Minicog Cognitive Impairment Screening and Validation of Frailty and Functionality Assessment Prior to Allogeneic Hematopoietic Stem Cell Transplantation

Background: We demonstrated that impaired functionality and frailty have a negative impact on survival after allogeneic hematopoietic cell transplantation (HCT) (Salas et al., 2019). This impact is independent of other predictive models including HCT comorbidity index (HCT-CI) and disease risk index (DRI). Cognitive changes are frequently reported among survivors of autologous and allogeneic HCT. However, impact of the presence of cognitive impairment prior to allogeneic HCT has not been studied. We hypothesized that the presence of cognitive impairment before transplantation could have a negative effect on the outcomes of allogeneic HCT. Patients and methods: We evaluated transplant outcomes of 277 patients who underwent our updated frailty assessment including the MiniCog scale, incorporated as standard of care in April 2020. The MiniCog assessment included recalling of three unrelated words, for a total of three points and drawing of a clock with the face, number and hands indicating a specific time, for a total of 2 points. We used the cutoff score of <3 as a marker of cognitive impairment and evaluated the effect of cognitive impairment on: overall survival (OS), relapse free survival (RFS), non-relapse mortality (NRM). We further sought to validate our previously described frailty scale in prediction of transplant outcomes, including OS, RFS, and NRM, in the new cohort and compare its performance in outcome prediction to the MiniCog scale. Our frailty evaluation included the following variables: Clinical Frailty Scale (CFS), Lawton's instrumental activities of daily living (IADL), timed to get up and go test (TUGT), grip strength (GS) using a hydraulic hand dynamometer, self-rated health questionnaire (SRH), a question on number of falls in the last year (F), C-reactive protein (CRP) and serum albumin (Alb) levels at the time of the consult. The frailty scale was calculated according to the following formula: 1.5×CFS + 1×IADL +1×GS + 1.5×TUGT +1×SRH+1×F+ 1.5×Alb+2×CRP. The frailty scale score ranged from 0-10.5. With these results we stablished 3 frailty cohorts for the patients being planned for alloHCT, fit (≤1), pre-frail (<1 to ≤5.5), and frail (>5.5), based on this risk model. Results: Median age at transplantation was 59 years and 85 (31%) patients were ≥ 65 years. Of all the patients, 36 (14%) had a Karnofsky Performance Scale Index (KPS) of <90%, 206 patients had a low-intermediate risk disease risk index (DRI). Median time of follow up was 9 months. Majority of the patients (46%) underwent transplantation for acute myeloid leukemia (AML) or myelodysplastic syndrome (MDS). Of all patients, 246 (88.8%) had a normal MiniCog score (>3). There was no difference in 2-y OS between patients with normal and low MiniCog score: 67% (95% confidence interval (CI): 59%-74%) vs 60% (95% CI: 36-78%), p=0.21. NRM at 24 months was 26% (95% CI 10-46%) for patients with an abnormal MiniCog vs 23% (95% CI 17-30%), p=0.80. There was no difference in 2-year RFS between patients with normal and low MiniCog score: 65% (95% CI 57-72%) vs 43.5% (95% CI 23-63%), p=0.09. Using the frailty scale, 144 (56%) patients were considered fit, 30 (12%) pre-frail and 81 (32%) frail. Using our original frailty scale, the 2-y OS among the fit, pre-frail and frail patients was: 78% (95% CI 65%-87%), 64% (95% CI 53%-74%) and 27% (95% CI 5.6%-55.6%) p=0.005. 2-year NRM was 19.2% (95% CI 9-31%), 23% (16-32%), and 41.6% (95% CI 11-70%) for the fit, pre-frail and frail patients respectively, p=0.12. Relapse free survival at 24 months was 72.5% (95% CI 58-83%), 60.5% (95% CI 50-70%), and 29% (95% CI 6-57%), p=0.003. Conclusions: In our cohort the sole presence of cognitive decline, based on the MiniCog assessment failed to be a predictor of outcomes after allogeneic HCT. However, ability of our preciously established frailty scale to predict transplant outcomes in this new cohort of patients further validates the results of our previous findings. The presence of frailty is an independent predictor of adverse outcomes after allogeneic HCT.

Noninvasive disease severity assessment in nonalcoholic fatty liver disease

Nonalcoholic fatty liver disease (NAFLD) is the most common liver disease affecting one-fourth population globally. It is also anticipated to be the major cause of liver-related morbidity and transplantation worldwide in the future decades.1 Of note the Asia-Pacific has a tremendous burden because of the rapid Westernization of the region and the racial characteristics.2-4 Currently, NAFLD represents a clinical field possessing several unmet needs. The major one is currently no reliable biomarker for disease severity, disease course, or outcome measurement. Therefore, a reliable and confident biomarker for the purposes deserves exploration. The effort is no doubt more challenging than for viral hepatitis infection. Cheng et al. presented the study results of the correlation of Mac-2 binding protein glycosylation isomer (M2BPGi) serum level with the other documented noninvasive biomarkers or panels in a single-center retrospective NAFLD database.5 M2BPGi has been vigorously studied as a marker for fibrosis in chronic hepatitis B virus infection, chronic hepatitis C virus infection, nonalcoholic steatohepatitis, and even hepatocellular carcinoma.6-9 The cross-sectional study measured the serum levels of the recently-identified glycosylation isoform of Mac-2 binding protein in both NAFLD and healthy controls. The results showed that the serum levels of M2BPGi were significantly correlated with the levels of AST to platelet ratio index, fibrosis 4 score (FIB-4), and NAFLD fibrosis score, respectively. The M2BPGi levels were significantly higher in females and had an incremental increase with age in both NAFLD patients and the healthy controls. In addition, the M2BPGi level was significantly higher in those intermediate or high risk for advanced fibrosis (defined as FIB-4 level ≥ 1.45) than their low-risk counterparts (FIB-4 level < 1.45). The results thus provided a piece of work addressing the noninvasive fibrosis assessment for the complex metabolic liver disorder. Generally, the degree of liver fibrosis is the major determinant and the significant predictor of long-term outcomes in patients with NAFLD. There is a higher risk of mortality associated with a higher stage of fibrosis in NAFLD.10, 11 Moreover, the risk of liver-related mortality increased on an exponential rather than linear scale with an increase in fibrosis stage.12 Liver biopsy remains an expensive and invasive procedure associated with potential complications, sampling error, and interobserver variability. Besides imaging- and elastography-based methods, several serum-based panels have been developed for the noninvasive approach to disease severity. Recently, the easy-to-access FIB-4 method has been validated in providing an accurate prediction of liver fibrosis and liver-related events. The algorithm has been adapted by major societies as a clinically useful tool for advanced fibrosis assessment.13-15 Nonetheless, further validation efforts across disease etiology, ethnicities, and other clinical factors are mandatory. In this direction, the current study provided a real-world validation approach for the performance of the serum biomarker in a clinical setting. Some issues remain under investigation despite the achievements of the study. The transition between fibrosis stages in the disease spectrum of NAFLD remains elusive, which mainly depends on histopathological interpretation. The lack of histopathological data could partly undermine the results and clinical implications. Second, although FIB-4 is widely used to predict the presence of advanced fibrosis, it is a tool for excluding advanced fibrosis rather than diagnosis per se. The cut-off values for FIB-4 in determining different risk levels have been extensively validated in the past decade. The results of the current study could be affected by using different values. In addition, the performance of a single biomarker could be significantly increased by combination with other biomarkers.7 Last but not least, results from a longitudinal follow-up study, preferably with other noninvasive methods, will provide more sufficient and convincing evidence in this regard. In conclusion, M2BPGi is a useful noninvasive biomarker for liver fibrosis assessment in NAFLD patients. Future validation of the performance for outcome prediction in a longitudinal manner will be a promising perspective. The author thanks secretary help from Taiwan Liver Research Foundation (TLRF) and Fatty Liver Special Interest Group of Taiwan Association for the Study of the Liver (TASL). The Foundation and the Association did not influence how the study was conducted or the approval of the manuscript. Jee-Fu Huang: Consultant of Roche, BMS, Gilead, Merck, Sysmex, Pharmaessential, Polaris, Aligos, and Instylla. Speaker for Abbvie, BMS, Gilead, Merck, Sysmex, and Roche.

Predictive Value of Total Metabolic Tumor Burden Prior to Treatment in NSCLC Patients Treated with Immune Checkpoint Inhibition.

We aimed to assess the predictive value of the total metabolic tumor burden prior to treatment in patients with advanced non-small-cell lung cancer (NSCLC) receiving immune checkpoint inhibitors (ICIs). Pre-treatment 2-deoxy-2-[18F]fluoro-D-glucose positron emission tomography/computed tomography (PET/CT) scans performed in two consecutive years for staging in adult patients with confirmed NSCLC were considered. Volume, maximum/mean standardized uptake value (SUVmax/SUVmean), metabolic tumor volume (MTV) and total lesion glycolysis (TLG) were assessed per delineated malignant lesion (including primary tumor, regional lymph nodes and distant metastases) in addition to the morphology of the primary tumor and clinical data. Total metabolic tumor burden was captured by totalMTV and totalTLG. Overall survival (OS), progression-free survival (PFS) and clinical benefit (CB) were used as endpoints for response to treatment. A total of 125 NSCLC patients were included. Osseous metastases were the most frequent distant metastases (n = 17), followed by thoracal distant metastases (pulmonal = 14 and pleural = 13). Total metabolic tumor burden prior to treatment was significantly higher in patients treated with ICIs (mean totalMTV ± standard deviation (SD) 72.2 ± 78.7; mean totalTLG ± SD 462.2 ± 538.9) compared to those without ICI treatment (mean totalMTV ± SD 58.1 ± 233.8; mean totalTLG ± SD 290.0 ± 784.2). Among the patients who received ICIs, a solid morphology of the primary tumor on imaging prior to treatment was the strongest outcome predictor for OS (Hazard ratio HR 28.04, p < 0.01), PFS (HR 30.89, p < 0.01) and CB (parameter estimation PE 3.46, p < 0.01), followed by the metabolic features of the primary tumor. Interestingly, total metabolic tumor burden prior to immunotherapy showed a negligible impact on OS (p = 0.04) and PFS (p = 0.01) after treatment given the hazard ratios of 1.00, but also on CB (p = 0.01) given the PE < 0.01. Overall, biomarkers on pre-treatment PET/CT scans showed greater predictive power in patients receiving ICIs, compared to patients without ICI treatment. Morphological and metabolic properties of the primary tumors prior to treatment in advanced NSCLC patients treated with ICI showed great outcome prediction performances, as opposed to the pre-treatment total metabolic tumor burdens, captured by totalMTV and totalTLG, both with negligible impact on OS, PFS and CB. However, the outcome prediction performance of the total metabolic tumor burden might be influenced by the value itself (e.g., poorer prediction performance at very high or very low values of total metabolic tumor burden). Further studies including subgroup analysis with regards to different values of total metabolic tumor burden and their respective outcome prediction performances might be needed.

Deep versus Handcrafted Tensor Radiomics Features: Prediction of Survival in Head and Neck Cancer Using Machine Learning and Fusion Techniques.

Although handcrafted radiomics features (RF) are commonly extracted via radiomics software, employing deep features (DF) extracted from deep learning (DL) algorithms merits significant investigation. Moreover, a "tensor'' radiomics paradigm where various flavours of a given feature are generated and explored can provide added value. We aimed to employ conventional and tensor DFs, and compare their outcome prediction performance to conventional and tensor RFs. 408 patients with head and neck cancer were selected from TCIA. PET images were first registered to CT, enhanced, normalized, and cropped. We employed 15 image-level fusion techniques (e.g., dual tree complex wavelet transform (DTCWT)) to combine PET and CT images. Subsequently, 215 RFs were extracted from each tumor in 17 images (or flavours) including CT only, PET only, and 15 fused PET-CT images through the standardized-SERA radiomics software. Furthermore, a 3 dimensional autoencoder was used to extract DFs. To predict the binary progression-free-survival-outcome, first, an end-to-end CNN algorithm was employed. Subsequently, we applied conventional and tensor DFs vs. RFs as extracted from each image to three sole classifiers, namely multilayer perceptron (MLP), random-forest, and logistic regression (LR), linked with dimension reduction algorithms. DTCWT fusion linked with CNN resulted in accuracies of 75.6 ± 7.0% and 63.4 ± 6.7% in five-fold cross-validation and external-nested-testing, respectively. For the tensor RF-framework, polynomial transform algorithms + analysis of variance feature selector (ANOVA) + LR enabled 76.67 ± 3.3% and 70.6 ± 6.7% in the mentioned tests. For the tensor DF framework, PCA + ANOVA + MLP arrived at 87.0 ± 3.5% and 85.3 ± 5.2% in both tests. This study showed that tensor DF combined with proper machine learning approaches enhanced survival prediction performance compared to conventional DF, tensor and conventional RF, and end-to-end CNN frameworks.

Enhancing Basketball Game Outcome Prediction through Fused Graph Convolutional Networks and Random Forest Algorithm.

Basketball is a popular sport worldwide, and many researchers have utilized various machine learning models to predict the outcome of basketball games. However, prior research has primarily focused on traditional machine learning models. Furthermore, models that rely on vector inputs tend to ignore the intricate interactions between teams and the spatial structure of the league. Therefore, this study aimed to apply graph neural networks to basketball game outcome prediction, by transforming structured data into unstructured graphs, to represent the interactions between teams in the 2012-2018 NBA season dataset. Initially, the study used a homogeneous network and undirected graph to build a team representation graph. The constructed graph was fed into a graph convolutional network, which yielded an average success rate of 66.90% in predicting the outcome of games. To improve the prediction success rate, feature extraction based on the random forest algorithm was combined with the model. The fused model yielded the best results, and the prediction accuracy was improved to 71.54%. Additionally, the study compared the results of the developed model with previous studies and the baseline model. Our proposed method considers the spatial structure of teams and the interaction between teams, resulting in superior performance in basketball game outcome prediction. The results of this study provide valuable insights for basketball performance prediction research.

Identification and Validation of Molecular Subtype and Prognostic Signature for Bladder Cancer Based on Neutrophil Extracellular Traps

Neutrophil extracellular traps (NETs) could promote tumor growth and distant metastases. Molecular subtypes of bladder cancer were identified with consensus cluster analysis. A NETs-related prognostic signature was constructed with LASSO cox regression analysis. As a result, we identified three subtypes of bladder cancer, which had a distinct difference in prognosis, immune microenvironment, TIDE score, mRNAsi score and IC50 score. We also developed a prognostic signature based on 5 NETs-related genes, which had a good performance in clinical outcome prediction of bladder cancer. These results may provide more data about the vital role of NETs in bladder cancer.

Identification of an unfolded protein response-related signature for predicting the prognosis of pancreatic ductal adenocarcinoma.

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive lethal malignancy. An effective prognosis prediction model is urgently needed for treatment optimization. The differentially expressed unfolded protein response (UPR)‒related genes between pancreatic tumor and normal tissue were analyzed using the TCGA-PDAC dataset, and these genes that overlapped with UPR‒related prognostic genes from the E-MTAB-6134 dataset were further analyzed. Univariate, LASSO and multivariate Cox regression analyses were applied to establish a prognostic gene signature, which was evaluated by Kaplan‒Meier curve and receiver operating characteristic (ROC) analyses. E‒MTAB‒6134 was set as the training dataset, while TCGA-PDAC, GSE21501 and ICGC-PACA-AU were used for external validation. Subsequently, a nomogram integrating risk scores and clinical parameters was established, and gene set enrichment analysis (GSEA), tumor immunity analysis and drug sensitivity analysis were conducted. A UPR-related signature comprising twelve genes was constructed and divided PDAC patients into high- and low-risk groups based on the median risk score. The UPR-related signature accurately predicted the prognosis and acted as an independent prognostic factor of PDAC patients, and the AUCs of the UPR-related signature in predicting PDAC prognosis at 1, 2 and 3 years were all more than 0.7 in the training and validation datasets. The UPR-related signature showed excellent performance in outcome prediction even in different clinicopathological subgroups, including the female (p<0.0001), male (p<0.0001), grade 1/2 (p<0.0001), grade 3 (p=0.028), N0 (p=0.043), N1 (p<0.001), and R0 (p<0.0001) groups. Furthermore, multiple immune-related pathways were enriched in the low-risk group, and risk scores in the low-risk group were also associated with significantly higher levels of tumor-infiltrating lymphocytes (TILs). In addition, DepMap drug sensitivity analysis and our validation experiment showed that PDAC cell lines with high UPR-related risk scores or UPR activation are more sensitive to floxuridine, which is used as an antineoplastic agent. Herein, we identified a novel UPR-related prognostic signature that showed high value in predicting survival in patients with PDAC. Targeting these UPR-related genes might be an alternative for PDAC therapy. Further experimental studies are required to reveal how these genes mediate ER stress and PDAC progression.

On the Role of Negative Precedent in Legal Outcome Prediction

Abstract Every legal case sets a precedent by developing the law in one of the following two ways. It either expands its scope, in which case it sets positive precedent, or it narrows it, in which case it sets negative precedent. Legal outcome prediction, the prediction of positive outcome, is an increasingly popular task in AI. In contrast, we turn our focus to negative outcomes here, and introduce a new task of negative outcome prediction. We discover an asymmetry in existing models’ ability to predict positive and negative outcomes. Where the state-of-the-art outcome prediction model we used predicts positive outcomes at 75.06 F1, it predicts negative outcomes at only 10.09 F1, worse than a random baseline. To address this performance gap, we develop two new models inspired by the dynamics of a court process. Our first model significantly improves positive outcome prediction score to 77.15 F1 and our second model more than doubles the negative outcome prediction performance to 24.01 F1. Despite this improvement, shifting focus to negative outcomes reveals that there is still much room for improvement for outcome prediction models. https://github.com/valvoda/Negative-Precedent-in-Legal-Outcome-Prediction

A Self-Attention-Guided 3D Deep Residual Network With Big Transfer to Predict Local Failure in Brain Metastasis After Radiotherapy Using Multi-Channel MRI

A noticeable proportion of larger brain metastases (BMs) are not locally controlled after stereotactic radiotherapy, and it may take months before local progression is apparent on standard follow-up imaging. This work proposes and investigates new explainable deep-learning models to predict the radiotherapy outcome for BM. A novel self-attention-guided 3D residual network is introduced for predicting the outcome of local failure (LF) after radiotherapy using the baseline treatment-planning MRI. The 3D self-attention modules facilitate capturing long-range intra/inter slice dependencies which are often overlooked by convolution layers. The proposed model was compared to a vanilla 3D residual network and 3D residual network with CBAM attention in terms of performance in outcome prediction. A training recipe was adapted for the outcome prediction models during pretraining and training the down-stream task based on the recently proposed big transfer principles. A novel 3D visualization module was coupled with the model to demonstrate the impact of various intra/peri-lesion regions on volumetric multi-channel MRI upon the network's prediction. The proposed self-attention-guided 3D residual network outperforms the vanilla residual network and the residual network with CBAM attention in accuracy, F1-score, and AUC. The visualization results show the importance of peri-lesional characteristics on treatment-planning MRI in predicting local outcome after radiotherapy. This study demonstrates the potential of self-attention-guided deep-learning features derived from volumetric MRI in radiotherapy outcome prediction for BM. The insights obtained via the developed visualization module for individual lesions can possibly be applied during radiotherapy planning to decrease the chance of LF.

A Systematic Review of Polygenic Models for Predicting Drug Outcomes.

Polygenic models have emerged as promising prediction tools for the prediction of complex traits. Currently, the majority of polygenic models are developed in the context of predicting disease risk, but polygenic models may also prove useful in predicting drug outcomes. This study sought to understand how polygenic models incorporating pharmacogenetic variants are being used in the prediction of drug outcomes. A systematic review was conducted with the aim of gaining insights into the methods used to construct polygenic models, as well as their performance in drug outcome prediction. The search uncovered 89 papers that incorporated pharmacogenetic variants in the development of polygenic models. It was found that the most common polygenic models were constructed for drug dosing predictions in anticoagulant therapies (n = 27). While nearly all studies found a significant association with their polygenic model and the investigated drug outcome (93.3%), less than half (47.2%) compared the performance of the polygenic model against clinical predictors, and even fewer (40.4%) sought to validate model predictions in an independent cohort. Additionally, the heterogeneity of reported performance measures makes the comparison of models across studies challenging. These findings highlight key considerations for future work in developing polygenic models in pharmacogenomic research.

Improving the Performance of Outcome Prediction for Inpatients With Acute Myocardial Infarction Based on Embedding Representation Learned From Electronic Medical Records: Development and Validation Study.

BackgroundThe widespread secondary use of electronic medical records (EMRs) promotes health care quality improvement. Representation learning that can automatically extract hidden information from EMR data has gained increasing attention.ObjectiveWe aimed to propose a patient representation with more feature associations and task-specific feature importance to improve the outcome prediction performance for inpatients with acute myocardial infarction (AMI).MethodsMedical concepts, including patients’ age, gender, disease diagnoses, laboratory tests, structured radiological features, procedures, and medications, were first embedded into real-value vectors using the improved skip-gram algorithm, where concepts in the context windows were selected by feature association strengths measured by association rule confidence. Then, each patient was represented as the sum of the feature embeddings weighted by the task-specific feature importance, which was applied to facilitate predictive model prediction from global and local perspectives. We finally applied the proposed patient representation into mortality risk prediction for 3010 and 1671 AMI inpatients from a public data set and a private data set, respectively, and compared it with several reference representation methods in terms of the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and F1-score.ResultsCompared with the reference methods, the proposed embedding-based representation showed consistently superior predictive performance on the 2 data sets, achieving mean AUROCs of 0.878 and 0.973, AUPRCs of 0.220 and 0.505, and F1-scores of 0.376 and 0.674 for the public and private data sets, respectively, while the greatest AUROCs, AUPRCs, and F1-scores among the reference methods were 0.847 and 0.939, 0.196 and 0.283, and 0.344 and 0.361 for the public and private data sets, respectively. Feature importance integrated in patient representation reflected features that were also critical in prediction tasks and clinical practice.ConclusionsThe introduction of feature associations and feature importance facilitated an effective patient representation and contributed to prediction performance improvement and model interpretation.

Prediction of Mortality and Major Adverse Kidney Events in Critically Ill Patients With Acute Kidney Injury

Risk prediction tools for assisting acute kidney injury (AKI) management have focused on AKI onset but have infrequently addressed kidney recovery. We developed clinical models for risk stratification of mortality and major adverse kidney events (MAKE) in critically ill patients with incident AKI. Multicenter cohort study. 9,587 adult patients admitted to heterogeneous intensive care units (ICUs; March 2009 to February 2017) who experienced AKI within the first 3 days of their ICU stays. Multimodal clinical data consisting of 71 features collected in the first 3 days of ICU stay. (1) Hospital mortality and (2) MAKE, defined as the composite of death during hospitalization or within 120 days of discharge, receipt of kidney replacement therapy in the last 48 hours of hospital stay, initiation of maintenance kidney replacement therapy within 120 days, or a≥50% decrease in estimated glomerular filtration rate from baseline to 120 days from hospital discharge. Four machine-learning algorithms (logistic regression, random forest, support vector machine, and extreme gradient boosting) and the SHAP (Shapley Additive Explanations) framework were used for feature selection and interpretation. Model performance was evaluated by 10-fold cross-validation and external validation. One developed model including 15 features outperformed the SOFA (Sequential Organ Failure Assessment) score for the prediction of hospital mortality, with areas under the curve of 0.79 (95% CI, 0.79-0.80) and 0.71 (95% CI, 0.71-0.71) in the development cohort and 0.74 (95% CI, 0.73-0.74) and 0.71 (95% CI, 0.71-0.71) in the validation cohort (P< 0.001 for both). A second developed model including 14 features outperformed KDIGO (Kidney Disease: Improving Global Outcomes) AKI severity staging for the prediction of MAKE: 0.78 (95% CI, 0.78-0.78) versus 0.66 (95% CI, 0.66-0.66) in the development cohort and 0.73 (95% CI, 0.72-0.74) versus 0.67 (95% CI, 0.67-0.67) in the validation cohort (P< 0.001 for both). The models are applicable only to critically ill adult patients with incident AKI within the first 3 days of an ICU stay. The reported clinical models exhibited better performance for mortality and kidney recovery prediction than standard scoring tools commonly used in critically ill patients with AKI in the ICU. Additional validation is needed to support the utility and implementation of these models. Acute kidney injury (AKI) occurs commonly in critically ill patients admitted to the intensive care unit (ICU) and is associated with high morbidity and mortality rates. Prediction of mortality and recovery after an episode of AKI may assist bedside decision making. In this report, we describe the development and validation of a clinical model using data from the first 3 days of an ICU stay to predict hospital mortality and major adverse kidney events occurring as long as 120 days after hospital discharge among critically ill adult patients who experienced AKI within the first 3 days of an ICU stay. The proposed clinical models exhibited good performance for outcome prediction and, if further validated, could enable risk stratification for timely interventions that promote kidney recovery.

CT severity score in COVID-19 patients, assessment of performance in triage and outcome prediction: a comparative study of different methods

BackgroundLung involvement in COVID-19 can be quantified by chest CT scan with some triage and prognostication value. Optimizing initial triage of patients could help decrease adverse health impacts of the disease through better clinical management. At least 6 CT severity score (CTSS) systems have been proposed. We aimed to evaluate triage and prognostication performance of seven different CTSSs, including one proposed by ourselves, in hospitalized COVID-19 patients diagnosed by positive polymerase chain reaction (PCR).ResultsAfter exclusion of 14 heart failure and significant preexisting pulmonary disease patients, 96 COVID-19, PCR-positive patients were included into our retrospective study, admitted from February 20, 2020, to July 22. Their mean age was 63.6 ± 17.4 years (range 21–88, median 67). Fifty-seven (59.4%) were men, and 39 (40.6%) were women. All CTSSs showed good interrater reliability as calculated intraclass correlation coefficients (ICCs) between two radiologists were 0.764–0.837. Those CTSSs with more numerous segmentations showed the best ICCs. As judged by area under curve (AUC) for each receiver operator characteristic (ROC) curve, only three CTSSs showed acceptable AUCs (AUC = 0.7) for triage of severe/critical patients. All CTSSs showed acceptable AUCs for prognostication (AUCs = 0.76–0.79). Calculated AUCs for different CTSSs were not significantly different for triage and for prediction of severe/critical disease, but some difference was shown for prediction of critical disease.ConclusionsMen are probably affected more frequently than women by COVID-19. Quantification of lung disease in COVID-19 is a readily available and easy tool to be used in triage and prognostication, but we do not advocate its use in heart failure or chronic respiratory disease patients. The scoring systems with more numerous segmentations are recommended if any future imaging for comparison is contemplated. CTSS performance in triage was much lower than earlier reports, and only three CTSSs showed acceptable AUCs in this regard. CTSS performed better for prognostic purposes than for triage as all 7 CTSSs showed acceptable AUCs in both types of prognostic ROC curves. There is not much difference among performance of different CTSSs.

Hypoperfusion intensity ratio correlates with clinical outcome of endovascular thrombectomy in acute ischaemic stroke patients with late therapeutic window

To evaluate the prognostic value of the hypoperfusion intensity ratio (HIR) on 90-day clinical outcome in acute ischaemic stroke (AIS) patients with late therapeutic window. One hundred and sixty-eight consecutive AIS patients with anterior-circulation large-vessel occlusion who underwent endovascular thrombectomy during the late window were enrolled retrospectively. Clinical data, Alberta Stroke Program Early Computed Tomography Score (ASPECTS) based on unenhanced computed tomography (CT), and perfusion parameters included ischaemic core, hypoperfusion volume, mismatch volume between core and penumbra, and the HIR were assessed and compared between patients with or without favourable outcomes (defined as modified Rankin Scale score of 0-2). Statistical analysis included binary logistic regression and receiver operating characteristic (ROC) analyses. A favourable outcome was achieved in 76 (45.2%) patients. In univariable analysis, age, National Institutes of Health Stroke Scale (NIHSS) score at admission, ASPECTS score, HIR, ischaemic core, and hypoperfusion volume were significantly associated with functional outcome (p<0.05). In multivariate analyses, age (OR 0.95; 95% CI 0.92-0.99), NIHSS score at admission (OR 0.89, 95% CI 0.84-0.96) and HIR (OR 0.018, 95% CI 0.003-0.113) remained as independent outcome predictors (p<0.01). The optimal threshold of HIR was 0.36 (sensitivity 70.7%, specificity 61.8%). The combination of age, NIHSS score at admission, and HIR yield good performance for outcome prediction with an area under the ROC curve of 0.815 (sensitivity 88.2%, specificity 64.1%), significantly higher than individual variable (p<0.05). Low HIR was a predictor for favourable outcome in AIS patients with late therapeutic window. Integrating HIR with clinical variables improved the ability for outcome classification.

Neurological Pupil Index for the Early Prediction of Outcome in Severe Acute Brain Injury Patients.

In this study, we examined the early value of automated quantitative pupillary examination, using the Neurological Pupil index (NPi), to predict the long-term outcome of acute brain injured (ABI) patients. We performed a single-centre retrospective study (October 2016–March 2019) in ABI patients who underwent NPi measurement during the first 3 days following brain insult. We examined the performance of NPi—alone or in combination with other baseline demographic (age) and radiologic (CT midline shift) predictors—to prognosticate unfavourable 6-month outcome (Glasgow Outcome Scale 1–3). A total of 145 severely brain-injured subjects (65 traumatic brain injury, TBI; 80 non-TBI) were studied. At each time point tested, NPi <3 was highly predictive of unfavourable outcome, with highest specificity (100% (90–100)) at day 3 (sensitivity 24% (15–35), negative predictive value 36% (34–39)). The addition of NPi, from day 1 following ABI to age and cerebral CT scan, provided the best prognostic performance (AUROC curve 0.85 vs. 0.78 without NPi, p = 0.008; DeLong test) for 6-month neurological outcome prediction. NPi, assessed at the early post-injury phase, has a superior ability to predict unfavourable long-term neurological outcomes in severely brain-injured patients. The added prognostic value of NPi was most significant when complemented with baseline demographic and radiologic information.

Unfinished Business in Classifying HPV-Positive Oropharyngeal Carcinoma: Identifying the Bad Apples in a Good Staging Barrel.

This commentary highlights three important findings in the study by Vijayvargiya et al, published in this journal, involving 9554 oropharyngeal cancer patients from the SEER database. Firstly, there is improved performance in outcome prediction with TNM-8 in HPV+ OPC. However, heterogeneity exists, especially in TNM-8 stage I disease, and there is need for ongoing improvement in risk stratification. Several anatomical and non-anatomical prognostic factors have been proposed. Among them, radiologic extranodal extension has emerged as one of the promising parameters to be considered for future staging. These baseline prognostic factors should address sensitivity, specificity, and diagnostic accuracy to serve different clinical needs. Secondly, cure is possible for some patients presenting with M1 disease. Optimal management of such patients remains to be explored, and clinical trials targeting de novo M1 disease should be encouraged to optimize outcomes for this subset. Finally, methodologies to address missing tumor HPV status in historical cohorts have been discussed, including using baseline demographics and clinical characteristics, as well as statistical procedures such as multiple imputation.