Outcome Prediction Models Research Articles

A Novel AUC Maximization Imbalanced Learning Approach for Predicting Composite Outcomes in COVID-19 Hospitalized Patients.

The COVID-19 patient data for composite outcome pre- diction often come with class imbalance issues, i.e., only a small group of patients develop severe composite events after hospital admission while the rest do not. An ideal COVID-19 composite outcomes prediction model should seek strong imbalanced learning capability. Such a model also should have fewer tuning hyperparameters to preserve its good usability and exhibit its potential for fast incremen- tal learning. Towards this goal, by means of the classical proximal support vector machine (PSVM), this study pro- poses a novel imbalanced learning approach called Imbalanced maximizing-Area Under the Curve (AUC) Proximal Support Vector Machine (ImAUC-PSVM) to predict the hospitalized COVID-19 patients' composite outcomes within 30 days of hospitalization. ImAUC-PSVM has the following merits: (1) it incorporates the straightforward AUC maximization into the objective function and hence has fewer parameters to tune. This makes it naturally suitable for handling imbalanced COVID-19 data with a simplified training process. (2) Our theoretical derivations reveal that it has the same form of the analytical solution as PSVM. Therefore, it inherits the advantages of PSVM in dealing with incremental COVID- 19 cases using the fast incremental updating. Our proposed classifier was built and validated internally and externally on the real COVID-19 patient data retrieved from three separate sites of Mayo Clinic in the United States and also validated on the public datasets using various performance metrics. Experi- mental results demonstrate that ImAUC-PSVM outperforms in most cases, showing its potential to assist clinicians in triaging COVID-19 patients at early stage in the hospital settings.

Developing predictive models for surgical outcomes in patients with degenerative cervical myelopathy: a comparison of statistical and machine learning approaches

BACKGROUND CONTEXTMachine learning (ML) is widely used to predict the prognosis of numerous diseases. PURPOSEThis retrospective analysis aimed to develop a prognostic prediction model using ML algorithms and identify predictors associated with poor surgical outcomes in patients with degenerative cervical myelopathy (DCM). STUDY DESIGNA retrospective study. PATIENT SAMPLEA total of 406 symptomatic DCM patients who underwent surgical decompression were enrolled and analyzed from three independent medical centers. OUTCOME MEASURESWe calculated the area under the curve (AUC), classification accuracy, sensitivity, and specificity of each model. METHODSThe Japanese Orthopedic Association (JOA) score was obtained before and 1 year following decompression surgery, and patients were grouped into good and poor outcome groups based on a cut-off value of 60% based on a previous study. Two datasets were fused for training, 1 dataset was held out as an external validation set. Optimal feature-subset and hyperparameters for each model were adjusted based on a 2,000-resample bootstrap-based internal validation via exhaustive search and grid search. The performance of each model was then tested on the external validation set. RESULTSThe Support Vector Machine (SVM) model showed the highest predictive accuracy compared to other methods, with an AUC of 0.82 and an accuracy of 75.7%. Age, sex, disease duration, and preoperative JOA score were identified as the most commonly selected features by both the ML and statistical models. Grid search optimization for hyperparameters successfully enhanced the predictive performance of each ML model, and the SVM model still had the best performance with an AUC of 0.93 and an accuracy of 86.4%. CONCLUSIONSOverall, the study demonstrated that ML classifiers such as SVM can effectively predict surgical outcomes for patients with DCM while identifying associated predictors in a multivariate manner.

ASPECTS-based net water uptake outperforms target mismatch for outcome prediction in patients with acute ischemic stroke and late therapeutic window.

To compare the prognostic value of net water uptake (NWU) and target mismatch (TM) on CT perfusion (CTP) in acute ischemic stroke (AIS) patients with late time window. One hundred and nine consecutive AIS patients with anterior-circulation large vessel occlusion presenting within 6-24 h from onset/last seen well were enrolled. Automated Alberta Stroke Program Early CT Score-based NWU (ASPECTS-NWU) was calculated from admission CT. The correlation between ASPECTS-NWU and CTP parameters was assessed. Predictors for favorable outcome (modified Rankin Scale score ≤ 2) at 90 days were assessed using logistic regression analysis. The ability of outcome prediction between ASPECTS-NWU and TM (an ischemic core < 70 mL, a mismatch ratio ≥ 1.8, and an absolute difference ≥ 15 mL) was compared using receiver operating characteristic (ROC) curve. A higher level of ASPECTS-NWU was associated with a larger ischemic core (r = 0.66, p < 0.001) and a larger hypoperfusion volume (r = 0.38, p < 0.001). ASPECTS-NWU performed better than TM for outcome stratification (area under the curve [AUC], 0.738 vs 0.583, p = 0.004) and was the only independent neuroimaging marker associated with favorable outcomes compared with CTP parameters (odds ratio, 0.73; 95% confidence interval [CI] 0.62-0.87, p < 0.001). An outcome prediction model including ASPECTS-NWU and clinical variables (National Institutes of Health Stroke Scale scores and age) yielded an AUC of 0.828 (95% CI 0.744-0.893; sensitivity 65.4%; specificity 87.7%). ASPECTS-NWU performed better than TM for outcome prediction in AIS patients with late time window and might be an alternative imaging biomarker to CTP for patient selection. Automated Alberta Stroke Program Early CT Score-based net water uptake outperforms target mismatch on CT perfusion for the outcome prediction in patients with acute ischemic stroke and can be an alternative imaging biomarker for patient selection in late therapeutic window. • A higher ASPECTS-based net water uptake was associated with larger ischemic cores and hypoperfusion volumes on CT perfusion. • ASPECTS-based net water uptake outperformed target mismatch for outcome prediction in acute ischemic stroke with extended therapeutic window. • ASPECTS-based net water uptake can be an alternative biomarkerto target mismatch for selecting acute ischemic stroke patients with late therapeutic window.

The prediction model of the short-term outcome in elderly heart failure patients.

This study was designed to investigate the effect of the comprehensive geriatric assessment on the short-term prognosis of the elderly heart failure patients (EHFP), analyze the relevant risk factors, and construct an effective risk prediction model. According to the selection and exclusion criteria, 617 patients were filtered from 800 patients from the cadre ward database of the first Hospital of Jilin University. The EHFP were randomly divided into the model group (432 cases) and the validation group (185 cases). A retrospective study on the general clinical data of patients in the model group was conducted to analyze the risk factors associated with the short-term outcomes of EHFP. Based on the risk factors, the risk prediction model was established and validated through the validation group. In the model group, the following independent risk factors were identified for the short-term outcomes in EHFP in the light of univariate logistic and cox regression analysis: female (β = 0.989, OR = 1.277, 95% CI: 1.090-1.847, P = 0.024), age (65-75years, β = 0.654, OR = 2.320, 95% CI: 1.135-3.136, P = 0.012; 75-85years, β = 1.123, OR = 3.159, 95% CI: 1.532-5.189, P = 0.001; age > 85years old, β = 1.513, OR = 4.895, 95% CI: 1.866-979, P = 0.001), frailty (β = 1.015, OR = 2.761, 95% CI: 1.097-6.945, P = 0.031), malnutrition (β = 1.271, OR = 3.560, 95% CI: 1.122-11.325, P = 0.002), and EF≦40% (β = 1.250, OR = 3.498, 95% CI: 1.898-6.447, P = 0.001). The simple risk prediction score was set up in line with the five risk factors, including range (1-7), the area under ROC curve (0.771, 95% CI: 0.723-0.819), and H-L test (P = 0.393), so patients were divided into the low-risk group (1-3) and the high-risk group (4-8). As a result, the number of EHFP in the high-risk group was significantly much more than that in the low-risk group (70.1% versus 29.9%, P < 0.001). Besides, the area under ROC curve (0.758, 95% CI: 0.682-0.835) and H-L test (P = 0.669) of the validation group indicated that this model could be a promising prediction model for the short-term outcomes of EHFP. Female, age, frailty, malnutrition, and EF ≦ 40% are independent risk factors for short-term outcomes of EHFP. The risk prediction model based on the five risk factors provided compelling clinic predictive value for the short-term prognosis of EHFP.

Skeletal muscle-focused guideline development: hierarchical model incorporating muscle form, function, and clinical outcomes.

Sarcopenia, sarcopenic obesity, malnutrition, and cachexia clinical guidelines were created by expert consensus over the past decade. These pathological states all share in common deficits in skeletal muscle mass, and in some cases muscle function, which adversely impact patient outcomes. Early identification is key as some detrimental outcomes are potentially preventable with available treatments. The four guidelines share common design features: patients suspected of having the condition are first screened with a focused clinical history; if positive, the next step is evaluation with either a measure of body "form" (e.g., mass, shape, and composition) or function (e.g., mechanical, endurance, and metabolic); combined form and functional criteria are also recognized. The form and functional "gateway" nodes establish whether or not to proceed with further evaluations and treatments. Intensive discussions among experts focus on selection of these gateway nodes and the final choice is made when consensus is reached. Form and functional measures are often treated as equivalent alternatives when framed in the context of "outcomes" for which they are intended to predict. Here we adapt a classic biological concept stating that "function follows form" to show that pathophysiological links are present between these two different muscle qualities and clinical outcomes. We argue that a hierarchy exists such that outcomes closely follow functions that, in turn, follow form…the OFF rule. The OFF rule explains why functional measures often show stronger associations with outcomes than those quantifying form, helps to frame debates on how to structure the gateway nodes used to identify patients for further evaluation and treatment, and sets out a pathophysiological structure for developing future outcome prediction models.

Texas Youth Depression and Suicide Research Network (TX-YDSRN) research registry and learning healthcare network: Rationale, design, and baseline characteristics

BackgroundAmerican youth are seriously impacted by depression and suicide. The Texas Youth Depression and Suicide Research Network (TX-YDSRN) Participant Registry Study was initiated in 2020 to develop predictive models for treatment outcomes in youth with depression and/or suicidality. This report presents the study rationale, design and baseline characteristics of the first 1000 participants. MethodsTX-YDSRN consists of the Network Hub (coordinating center), 12 medical school “Nodes” (manage/implement study), each with 1–5 primary care, inpatient, and/or outpatient Sub-Sites (recruitment, data collection). Participants are 8–20-year-olds who receive treatment or screen positive for depression and/or suicidality. Baseline data include mood and suicidality symptoms, associated comorbidities, treatment history, services used, and social determinants of health. Subsequent assessments occur every two months for 24 months. ResultsAmong 1000 participants, 68.7 % were 12–17 years, 24.6 % were ≥ 18 years, and 6.7 % were < 12. Overall, 36.8 % were non-Hispanic Caucasian, 73.4 % were female, and 79.9 % had a primary depressive disorder. Nearly half of the sample reported ≥1 suicide attempt, with rates similar in youth 12–17 years old (49.9 %) and those 18 years and older (45.5 %); 29.9 % of children <12 reported at least one suicide attempt. Depression and anxiety scores were in the moderate-severe range for all age groups (Patient Health Questionnaire for Adolescents [PHQ-A]: 12.9 ± 6.4; Generalized Anxiety Disorder [GAD-7]: 11.3 ± 5.9). LimitationsThe sample includes youth who are receiving depression care at enrollment and may not be representative of non-diagnosed, non-treatment seeking youth. ConclusionsThe TX-YDSRN is one of the largest prospective longitudinal cohort registries designed to develop predictive models for outcome trajectories based on disorder heterogeneity, social determinants of health, and treatment availability.

Interpretable machine learning to predict adverse perinatal outcomes: examining marginal predictive value of risk factors during pregnancy

The timely identification of nulliparas at high risk of adverse fetal and neonatal outcomes during pregnancy is crucial for initiating clinical interventions to prevent perinatal complications. Although machine learning methods have been applied to predict preterm birth and other pregnancy complications, many models do not provide explanations of their predictions, limiting the clinical use of the model. This study aimed to develop interpretable prediction models for a composite adverse perinatal outcome (stillbirth, neonatal death, estimated Combined Apgar score of <10, or preterm birth) at different points in time during the pregnancy and to evaluate the marginal predictive value of individual predictors in the context of a machine learning model. This was a secondary analysis of the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be data, a prospective cohort study in which 10,038 nulliparous pregnant individuals with singleton pregnancies were enrolled. Here, interpretable prediction models were developed using L1-regularized logistic regression for adverse perinatal outcomes using data available at 3 study visits during the pregnancy (visit 1: 6 0/7 to 13 6/7 weeks of gestation; visit 2: 16 0/7 to 21 6/7 weeks of gestation; visit 3: 22 0/7 to 29 6/7 weeks of gestation). We identified the important predictors for each model using SHapley Additive exPlanations, a model-agnostic method of computing explanations of model predictions, and evaluated the marginal predictive value of each predictor using the DeLong test. Our interpretable machine learning model had an area under the receiver operating characteristic curves of 0.617 (95% confidence interval, 0.595-0.639; all predictor variables at visit 1), 0.652 (95% confidence interval, 0.631-0.673; all predictor variables at visit 2), and 0.673 (95% confidence interval, 0.651-0.694; all predictor variables at visit 3). For all visits, the placental biomarker inhibin A was a valuable predictor, as including inhibin A resulted in better performance in predicting adverse perinatal outcomes (P<.001, all visits). At visit 1, endoglin was also a valuable predictor (P<.001). At visit 2, free beta human chorionic gonadotropin (P=.001) and uterine artery pulsatility index (P=.023) were also valuable predictors. At visit 3, cervical length was also a valuable predictor (P<.001). Despite various advances in predictive modeling in obstetrics, the accurate prediction of adverse perinatal outcomes remains difficult. Interpretable machine learning can help clinicians understand how predictions are made, but barriers exist to the widespread clinical adoption of machine learning models for adverse perinatal outcomes. A better understanding of the evolution of risk factors for adverse perinatal outcomes throughout pregnancy is necessary for the development of effective interventions.

Predicting Prostate Surgery Outcomes from Standard Clinical Assessments of Lower Urinary Tract Symptoms To Derive Prognostic Symptom and Flowmetry Criteria

BackgroundAssessment of male lower urinary tract symptoms (LUTS) needs to identify predictors of symptom outcomes when interventional treatment is planned. ObjectiveTo develop a novel prediction model for prostate surgery outcomes and validate it using a separate patient cohort and derive thresholds for key clinical parameters. Design, setting, and participantsFrom the UPSTREAM trial of 820 men seeking treatment for LUTS, analysis of bladder diary (BD), International Prostate Symptom Score (IPSS), IPSS-quality of life, and uroflowmetry data was performed for 176 participants who underwent prostate surgery and provided complete data. For external validation, data from a retrospective database of surgery outcomes in a Japanese urology department (n = 227) were used. Outcome measurements and statistical analysisSymptom improvement was defined as a reduction in total IPSS of ≥3 points. Multiple logistic regression, classification tree analysis, and random forest models were generated, including versions with and without BD data. Results and limitationsMultiple logistic regression without BD data identified age (p = 0.029), total IPSS (p = 0.0016), and maximum flow rate (Qmax; p = 0.066) as predictors of outcomes, with area under the receiver operating characteristic curve (AUC) of 77.1%. Classification tree analysis without BD data gave thresholds of IPSS <16 and Qmax ≥13 ml/s (AUC 75.0%). The random forest model, which included all clinical parameters except BD data, had an AUC of 94.7%. Internal validation using the bootstrap method showed reasonable AUCs (69.6–85.8%). Analyses using BD data marginally improved the model fits. External validation gave comparable AUCs for logistic regression, classification tree analysis, and random forest models (all without BD; 70.9%, 67.3%, and 68.5%, respectively). Limitations include the significant number of men with incomplete baseline data and limited assessments in the external validation cohort. ConclusionsOutcomes of prostate surgery can be predicted preoperatively using age, total IPSS, and uroflowmetry data, with prognostic thresholds of 16 for IPSS and 13 ml/s for Qmax. Patient summaryThis study identified key preoperative factors that can predict outcomes of prostate surgery for bothersome urinary symptoms, including which patients are at risk of a poor outcome.

Prediction model for long-term seizure and developmental outcomes among children with infantile epileptic spasms syndrome.

Children with infantile epileptic spasms syndrome (IESS) are likely to experience poor outcomes. Researchers have investigated the factors related to its long-term prognosis; however, none of them developed a predictive model. This study aimed to clarify the factors that influence the long-term prognosis of seizures and their development and to create a prediction model for IESS. We conducted a retrospective cohort study enrolling participants diagnosed with IESS at the Tottori University Hospital. We examined the seizure and developmental status at 3 and 7 years after the IESS onset and divided the participants into favorable and poor outcome groups. Subsequently, we analyzed the factors associated with the poor outcome group and developed a prediction model at 3 years by setting cutoff values using the receiver operating characteristic curve. Data were obtained from 44 patients with IESS (19 female patients and 25 male patients). Three years after epileptic spasms (ES) onset, seizure and development were the poor outcomes in 15 (34.9%) and 27 (61.4%) patients, respectively. The persistence of ES or tonic seizures (TS) after 90 days of onset, moderate or severe magnetic resonance imaging abnormalities, and developmental delay before IESS onset were significantly associated with poor outcomes. Seven years after the onset of ES, seizures and development were the poor outcomes in 9 (45.0%) and 13 (72.2%) patients, respectively. We found that no factor was significantly associated with poor seizure outcomes, and only developmental delay before IESS onset was significantly associated with poor developmental outcomes. Our prediction model demonstrated 86.7% sensitivity and 64.3% specificity for predicting poor seizure outcomes and 88.9% sensitivity and 100% specificity for predicting poor developmental outcomes. Our prediction model may be useful for predicting the long-term prognosis of seizures and their development after 3 years. Understanding the long-term prognosis during the initial treatment may facilitate the selection of appropriate treatment.

Factors influencing clinical pregnancy outcome of in vitro fertilization/intracytoplasmic sperm injection in older women.

To analyse the factors influencing clinical pregnancy outcome of in vitro fertilization/intracytoplasmic sperm injection (IVF/ICSI) in older women, and to establish a risk prediction model. A total of 425 patients receiving IVF/ICSI from March 2018 to March 2020 were divided into pregnancy group (n=194) and non-pregnancy group (n=231). The factors affecting the outcomes of IVF/ICSI were explored by univariate and multivariate logistic regression analyses. A nomogram prediction model was constructed. The two groups had significantly different age, body mass index, dysmenorrhea, parity, times of full-term births, history of cesarean section, basal follicle stimulating hormone, basal antral follicle count (AFC), number of high-quality embryos, and basal estradiol, luteinizing hormone and endometrial thickness on the day of human chorionic gonadotropin (HCG) administration (P<0.05). Age ≥40 years old, dysmenorrhea, history of cesarean section, basal AFC<9, number of high-quality embryos <4, and endometrial thickness on the day of HCG administration <11 mm led to IVF/ICSI failure. The established model exhibited high calibration and discrimination degrees in predicting the outcome of IVF/ICSI. The risk prediction model for the pregnancy outcome of IVF/ICSI in older women helps evaluate the fertility probability and risk, providing references for formulating reasonable assisted reproduction plans.

The Role of Levodopa Challenge in Predicting the Outcome of Subthalamic Deep Brain Stimulation.

Deep brain stimulation of the subthalamic nucleus (STN-DBS) is an effective and evidence-based treatment for idiopathic Parkinson's disease (iPD). A minority of patients does not sufficiently benefit from STN-DBS. The predictive validity of the levodopa challenge for individual patients is analyzed. Data from patients assessed with a preoperative Levodopa-test and a follow-up examination (mean ± standard deviation: 9.15 months ±3.39) from Kiel (n = 253), Berlin (n = 78) and Toronto (n = 98) were studied. Insufficient DBS outcome was defined as an overall UPDRS-III reduction <33% compared to UPDRS-III in med-off at baseline or alternatively if the minimal clinically important improvement of 5 points was not reached. Single UPDRS-items and sub-scores were dichotomized. Following exploratory analysis, we trained supervised regression- and classification models for outcome prediction. Data analysis confirmed significant correlation between the absolute UPDRS-III reduction during Levodopa challenge and after stimulation. But individual improvement was inaccurately predicted with a large range of up to 30 UPDRS III points. Further analysis identified preoperative UPDRS-III/med-off-scores and preoperative Levodopa-improvement as most influential factors. The models for UPDRS-III and sub-scores improvement achieved comparably low accuracy. With large prediction intervals, the Levodopa challenge use for patient counseling is limited, though remains important for excluding non-responders to Levodopa. Despite these deficiencies, the current practice of patient selection is highly successful and builds not only on the Levodopa challenge. However, more specific motor tasks and further paraclinical tools for prediction need to be developed.

Anterior ischemic stroke: Analysis of the multivariable CT-based models for prediction of clinical outcome

ObjectiveTo determine the predictive value of multiple CT-based measurements, individually and collectively, including arterial collateral filling (AC), tissue perfusion parameters, as well as cortical venous (CV) and medullary venous (MV) outflow, in patients with acute ischemic stroke (AIS). MethodsWe retrospectively reviewed a database of patients with AIS in the middle cerebral artery distribution, who underwent evaluation by multiphase CT-angiography and perfusion. AC pial filling was evaluated using a multiphase CTA imaging. The CV status was scored using the adopted PRECISE system based on contrast opacification of the main cortical veins. The MV status was defined by the degree of contrast opacification of medullary veins in one cerebral hemisphere as compared to the contralateral hemisphere. The perfusion parameters were calculated using FDA-approved automated software. A good clinical outcome was defined as a Modified Rankin Scale of 0-2 at 90 days. ResultsA total of 64 patients were included. Each of the CT-based measurements could predict clinical outcomes independently (P<0.05). AC pial filling and perfusion core based models did slightly better compared to each of the other models (AUC = 0.66). Among models with two variables, the perfusion core combined with MV status had the highest AUC=0.73 followed by a combination of MV status and AC (AUC=0.72). Multivariable modeling with all four variables resulted in the highest predictive value (AUC=0.77). ConclusionThe combination of arterial collateral flow, tissue perfusion, and venous outflow provides a more accurate prediction of clinical outcome in AIS than each variable alone. This additive effect of these techniques suggests that the information collected by each of these methods only partially overlaps.

Interpretable machine learning model for imaging-based outcome prediction after cardiac arrest

IntroductionEarly identification of brain injury patterns in computerized tomography (CT) imaging is crucial for post-cardiac arrest prognostication. Lack of interpretability of machine learning prediction reduces trustworthiness by clinicians and prevents translation to clinical practice. We aimed to identify CT imaging patterns associated with prognosis with interpretable machine learning. MethodsIn this IRB-approved retrospective study, we included consecutive comatose adult patients hospitalized at a single academic medical center after resuscitation from in- and out-of-hospital cardiac arrest between August 2011 and August 2019 who underwent unenhanced CT imaging of the brain within 24 hours of their arrest. We decomposed the CT images into subspaces to identify interpretable and informative patterns of injury, and developed machine learning models to predict patient outcomes (i.e., survival and awakening status) using the identified imaging patterns. Practicing physicians visually examined the imaging patterns to assess clinical relevance. We evaluated machine learning models using 80%-20% random data split and reported AUC values to measure the model performance. ResultsWe included 1284 subjects of whom 35% awakened from coma and 34% survived hospital discharge. Our expert physicians were able to visualize decomposed image patterns and identify those believed to be clinically relevant on multiple brain locations. For machine learning models, the AUC was 0.710 ± 0.012 for predicting survival and 0.702 ± 0.053 for predicting awakening, respectively. DiscussionWe developed an interpretable method to identify patterns of early post-cardiac arrest brain injury on CT imaging and showed these imaging patterns are predictive of patient outcomes (i.e., survival and awakening status).

Risk Prediction Models and Novel Prognostic Factors for Heart Failure with Preserved Ejection Fraction: A Systematic and Comprehensive Review.

Patients with heart failure with preserved ejection fraction (HFpEF) have large individual differences, unclear risk stratification, and imperfect treatment plans. Risk prediction models are helpful for the dynamic assessment of patients' prognostic risk and early intensive therapy of high-risk patients. The purpose of this study is to systematically summarize the existing risk prediction models and novel prognostic factors for HFpEF, to provide a reference for the construction of convenient and efficient HFpEF risk prediction models. Studies on risk prediction models and prognostic factors for HFpEF were systematically searched in relevant databases including PubMed and Embase. The retrieval time was from inception to February 1, 2023. The Quality in Prognosis Studies (QUIPS) tool was used to assess the risk of bias in included studies. The predictive value of risk prediction models for end outcomes was evaluated by sensitivity, specificity, the area under the curve, C-statistic, C-index, etc. In the literature screening process, potential novel prognostic factors with high value were explored. A total of 21 eligible HFpEF risk prediction models and 22 relevant studies were included. Except for 2 studies with a high risk of bias and 2 studies with a moderate risk of bias, other studies that proposed risk prediction models had a low risk of bias overall. Potential novel prognostic factors for HFpEF were classified and described in terms of demographic characteristics (age, sex, and race), lifestyle (physical activity, body mass index, weight change, and smoking history), laboratory tests (biomarkers), physical inspection (blood pressure, electrocardiogram, imaging examination), and comorbidities. It is of great significance to explore the potential novel prognostic factors of HFpEF and build a more convenient and efficient risk prediction model for improving the overall prognosis of patients. This review can provide a substantial reference for further research.

Uneven Between-Hospital Distribution of Patient-Related Risk Factors for Adverse Outcomes of Colorectal Cancer Treatment: A Population-Based Register Study.

The purpose of this study was to elucidate between-hospital variation in the prevalence at the time of diagnosis of patient-related risk factors for adverse outcomes of colorectal cancer (CRC) treatment. A register-based national cohort of 44,471 patients diagnosed with CRC and registered in the Danish Colorectal Cancer Group database in 2009-2018 was included in the study. Patient-related risk factors present at diagnosis were collected from national Danish registers within the areas of demography, lifestyle factors, comorbidity, participation in screening, disease-related factors and socioeconomic factors. Prediction models of short-term postoperative outcomes and mortality were modelled to examine the potential aggregated impact of patient-related risk factors on outcomes, and variations between hospitals were examined. The most conspicuous variations found were for old age (75+ years), ranging from 31% (95% confidence interval (95% CI): 29-33%) to 46% (95% CI: 43-48%), Union for International Cancer Control Stage I ranging from 12% (95% CI: 10-14%) to 21% (95% CI: 19-22%), Stage IV ranging from 23% (95% CI: 21-25%) to 35% (95% CI: 34-37%) and American Society of Anesthesiologists score ≥III ranging from 18% (95% CI: 16-19%) to 40% (95% CI: 37-43%). Clinically significant variations were found in predicted probability of 30-day surgical complications which varied from 17% (95% CI: 16-17%) to 23% (95% CI: 22-23%) and 90-day postoperative mortality which varied between 3.2% (95% CI: 3-3.4%) and 5.5% (95% CI: 4.9-6%). Marked variation in the prevalence of patient-related risk factors for adverse outcomes of colorectal cancer treatment exists between hospitals in Denmark. It seems reasonable to take these differences into account when comparing outcomes between hospitals.

Major trauma in pregnancy: prediction of maternal and perinatal adverse outcomes

Trauma, largely the consequence of motor vehicle crashes, is the leading cause of pregnancy-associated maternal mortality. Prediction of adverse outcomes has been difficult in pregnancy given the infrequent occurrence of traumatic events and anatomic considerations unique to pregnancy. The injury severity score, an anatomic scoring system with weighting dependent on severity and anatomic region of injury, is used in the prediction of adverse outcomes in the nonpregnant population but has yet to be validated in pregnancy. This study aimed to estimate the associations between risk factors and adverse pregnancy outcomes after major trauma in pregnancy and to develop a clinical prediction model for adverse maternal and perinatal outcomes. This was a retrospective analysis of a cohort of pregnant patients who sustained major trauma and who were admitted to 1 of 2 level 1 trauma centers. Three composite adverse pregnancy outcomes were evaluated, namely adverse maternal outcomes and short- and long-term adverse perinatal outcomes, defined as outcomes occurring within the first 72 hours of the traumatic event or encompassing the entire pregnancy. Bivariate analyses were performed to estimate the associations between clinical or trauma-related variables and adverse pregnancy outcomes. Multivariable logistic regression analyses were performed to predict each adverse pregnancy outcome. The predictive performance of each model was estimated using receiver operating characteristic curve analyses. A total of 119 pregnant trauma patients were included, 26.1% of whom met the severe adverse maternal pregnancy outcome criteria, 29.4% of whom met the severe short-term adverse perinatal pregnancy outcome definition, and 51.3% of whom met the severe long-term adverse perinatal pregnancy outcome definition. Injury severity score and gestational age were associated with the composite short-term adverse perinatal pregnancy outcome with an adjusted odds ratio of 1.20 (95% confidence interval, 1.11-1.30). The injury severity score was solely predictive of the adverse maternal and long-term adverse perinatal pregnancy outcomes with odds ratios of 1.65 (95% confidence interval, 1.31-2.09) and 1.14 (95% confidence interval, 1.07-1.23), respectively. An injury severity score ≥8 was the best cutoff for predicting adverse maternal outcomes with 96.8% sensitivity and 92.0% specificity (area under the receiver operating characteristic curve, 0.990±0.006). An injury severity score ≥3 was the best cutoff for the short-term adverse perinatal outcomes, which correlates with a 68.6% sensitivity and 65.1% specificity (area under the receiver operating characteristic curve, 0.755±0.055). An injury severity score ≥2 was the best cutoff for the long-term adverse perinatal outcomes, yielding a 68.3% sensitivity and 72.4% specificity (area under the receiver operating characteristic curve, 0.763±0.042). For pregnant trauma patients, an injury severity score of ≥8 was predictive of severe adverse maternal outcomes. Minor trauma in pregnancy, defined in this study as an injury severity score <2, was not associated with maternal or perinatal morbidity or mortality. These data can guide management decisions for pregnant patients who present after trauma.

Developing Prediction Models for COVID-19 Outcomes: A Valuable Tool for Resource-Limited Hospitals.

Coronavirus disease is a global pandemic with millions of confirmed cases and hundreds of thousands of deaths worldwide that continues to create a significant burden on the healthcare systems. The aim of this study was to determine the patient clinical and paraclinical profiles that associate with COVID-19 unfavourable outcome and generate a prediction model that could separate between high-risk and low-risk groups. The present study is a multivariate observational retrospective study. A total of 483 patients, residents of the municipality of Timișoara, the biggest city in the Western Region of Romania, were included in the study group that was further divided into 3 sub-groups in accordance with the disease severity form. Increased age (cOR=1.09, 95% CI: 1.06-1.11, p<0.001), cardiovascular diseases (cOR=3.37, 95% CI: 1.96-6.08, p<0.001), renal disease (cOR=4.26, 95% CI: 2.13-8.52, p<0.001), and neurological disorder (cOR=5.46, 95% CI: 2.71-11.01, p<0.001) were all independently significantly correlated with an unfavourable outcome in the study group. The severe form increases the risk of an unfavourable outcome 19.59 times (95% CI: 11.57-34.10, p<0.001), while older age remains an independent risk factor even when disease severity is included in the statistical model. An unfavourable outcome was positively associated with increased values for the following paraclinical parameters: white blood count (WBC; cOR=1.10, 95% CI: 1.05-1.15, p<0.001), absolute neutrophil count (ANC; cOR=1.15, 95% CI: 1.09-1.21, p<0.001) and C-reactive protein (CRP; cOR=1.007, 95% CI: 1.004-1.009, p<0.001). The best prediction model including age, ANC and CRP achieved a receiver operating characteristic (ROC) curve with the area under the curve (AUC) = 0.845 (95% CI: 0.813-0.877, p<0.001); cut-off value = 0.12; sensitivity = 72.3%; specificity = 83.9%. This model and risk profiling may contribute to a more precise allocation of limited healthcare resources in a clinical setup and can guide the development of strategies for disease management.

Paper 42: A Novel Machine Learning (ML) Algorithm to Predict Outcomes after Revision ACLR (rACLR) in the Multicenter Anterior Cruciate Ligament Reconstruction Study (MARS) Cohort

Objectives: The multicenter anterior cruciate ligament reconstruction study (MARS) Group has implicated several factors that influence revision anterior cruciate ligament (rACLR) outcomes. As machine learning (ML) becomes increasingly utilized in the orthopaedic literature, the application of ML methodology to MARS cohort data presents a valuable opportunity to translate data into patient-specific insights. This study sought to apply novel ML methodology to the MARS cohort data in order to determine 1) an optimal predictive model of rACLR graft failure and 2) features that are most predictive of rACLR graft failure in context of the predictive model. Methods: The MARS Group of 83 surgeons and 52 sites prospectively enrolled a cohort of patients undergoing rACLR. Patients were followed up at 2 and 6 years postoperatively regarding patient-reported outcomes (PROs), additional surgeries, and incidence of graft failure. Surgeon reported intraoperative findings and preoperative radiographic measures were obtained. Data was preprocessed and 381 demographic, clinical, and surgical features were used to build five machine learning models predictive of graft failure at 6 years postoperatively. Models used included traditional logistic regression (LR), XGBoost, Gradient Boosting, Random Forest, and a validated ensemble algorithm, AutoPrognosis (AP). Validated performance metrics for binary outcome prediction models were used to determine discriminative power and calibration. Individual feature importance was calculated for the highest performing models using partial dependence and perturbation-based feature importance. Results: The cohort included 831 patients who completed six year follow up, and 5.8% (n=48) of whom experienced graft failure. While all models had moderate to good concordance, AP demonstrated the highest discriminative power compared to other models (Model: AUROC | AP: 0.722 | Random Forest: 0.621 | Gradient Boosting: 0.654 | XGBoost: 0.690 | Logistic Regression: 0.630). The AP model was well-calibrated, with calibration scores similar to the other studied models (Model: Brier Score | AP: 0.054 | Random Forest: 0.054 | Gradient Boosting: 0.059 | XGBoost: 0.059 | Logistic Regression: 0.102). Features deemed important for AP differ from those for LR model performance. For AP, partial dependence feature importance calculations demonstrated the following top five contributors to model predictive ability: surgeon years of experience, prior femoral tunnel position measured via preoperative radiograph, compromised prior ACLR femoral tunnel position and size, and baseline patient age. Similarly, AP perturbation-based calculations showed the following top five features that contribute most to the AUROC performance: baseline patient age, current ACLR graft type, years since previous ACLR, prior ACLR femoral tunnel position, and prior tibial tunnel position on sagittal view preoperative radiographs. Perturbation-based calculations showed the following top five features that contribute most to the AUPRC performance: current ACLR graft type, baseline patient age, prior tibial tunnel position on sagittal view preoperative radiographs, previous ipsilateral medial meniscus repair, and baseline SF-36 subscale scores for patient vitality. Other leading contributors to model predictive ability included baseline PROs including SF-36 subscale scores for physical function, mental health and composite mental health, KOOS quality of life, WOMAC pain, and MARX activity rating scores. Conclusions: Of the studied models in this preliminary analysis, AP appears to most accurately predict rACLR graft failure. These findings build on prior studies, identifying key surgical, clinical, radiographic, and patient-reported factors contributing to the model’s ability to predict rACLR graft failure at 6 years postoperatively. Further work includes creation of a clinical risk calculator using important feature inputs, and the use of survival modeling to determine risk scores at various postoperative time points.

Development and Validation of an Electronic Health Record-Based Risk Assessment Tool for Hypoglycemia in Patients With Type 2 Diabetes Mellitus.

The recent availability of high-quality data from clinical trials, together with machine learning (ML) techniques, presents exciting opportunities for developing prediction models for clinical outcomes. As a proof-of-concept, we translated a hypoglycemia risk model derived from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) study into the HypoHazardScore, a risk assessment tool applicable to electronic health record (EHR) data. To assess its performance, we conducted a 16-week clinical study at the University of Minnesota where participants (N = 40) with type 2 diabetes mellitus (T2DM) had hypoglycemia assessed prospectively by continuous glucose monitoring (CGM). The HypoHazardScore combines 16 risk factors commonly found within the EHR. The HypoHazardScore successfully predicted (area under the curve [AUC] = 0.723) whether participants experienced least one CGM-assessed hypoglycemic event (glucose <54 mg/dL for ≥15 minutes from two CGMs) while significantly correlating with frequency of CGM-assessed hypoglycemic events (r = 0.38) and percent time experiencing CGM-assessed hypoglycemia (r = 0.39). Compared to participants with a low HypoHazardScore (N = 19, score <4, median score of 4), participants with a high HypoHazardScore (N = 21, score ≥4) had more frequent CGM-assessed hypoglycemic events (high: 1.6 ± 2.2 events/week; low: 0.3 ± 0.5 events/week) and experienced more CGM-assessed hypoglycemia (high: 1.4% ± 2.0%; low: 0.2% ± 0.4% time) during the 16-week follow-up. We demonstrated that a hypoglycemia risk model can be successfully adapted from the ACCORD data to the EHR, with validation by a prospective study using CGM-assessed hypoglycemia. The HypoHazardScore represents a significant advancement toward implementing an EHR-based decision support system that can help reduce hypoglycemia in patients with T2DM.

A systematic review and meta-analysis of predictive and prognostic models for outcome prediction using positron emission tomography radiomics in head and neck squamous cell carcinoma patients.

Positron emission tomography (PET) images of head and neck squamous cell carcinoma (HNSCC) patients can assess the functional and biochemical processes at cellular levels. Therefore, PET radiomics-based prediction and prognostic models have the potentials to understand tumour heterogeneity and assist clinicians with diagnosis, prognosis and management of the disease. We conducted a systematic review of published modelling information to evaluate the usefulness of PET radiomics in the prediction and prognosis of HNSCC patients. We searched bibliographic databases (MEDLINE, Embase, Web of Science) from 2010 to 2021 and considered 31 studies with pre-defined inclusion criteria. We followed the CHARMS checklist for data extraction and performed quality assessment using the PROBAST tool. We conducted a meta-analysis to estimate the accuracy of the prediction and prognostic models using the diagnostic odds ratio (DOR) and average C-statistic, respectively. Manual segmentation method followed by 40% of the maximum standardised uptake value (SUVmax ) thresholding is a commonly used approach. The area under the receiver operating curves of externally validated prediction models ranged between 0.60-0.87, 0.65-0.86 and 0.62-0.75 for overall survival, distant metastasis and recurrence, respectively. Most studies highlighted an overall high risk of bias (outcome definition, statistical methodologies and external validation of models) and high unclear concern in terms of applicability. The meta-analysis showed the estimated pooled DOR of 6.75 (95% CI: 4.45, 10.23) for prediction models and the C-statistic of 0.71 (95% CI: 0.67, 0.74) for prognostic models. Both prediction and prognostic models using clinical variables and PET radiomics demonstrated reliable accuracy for detecting adverse outcomes in HNSCC, suggesting the prospect of PET radiomics in clinical settings for diagnosis, prognosis and management of HNSCC patients. Future studies of prediction and prognostic models should emphasise the quality of reporting, external model validation, generalisability to real clinical scenarios and enhanced reproducibility of results.