Clinical Prediction Algorithms Research Articles

Prospective Validation and Usability Evaluation of a Mobile Diagnostic App for Obstructive Sleep Apnea.

Background/Objectives: Obstructive sleep apnea (OSA) classification relies on polysomnography (PSG) results. Current guidelines recommend the development of clinical prediction algorithms in screening prior to PSG. A recent intuitive and user-friendly tool (OSABayes), based on a Bayesian network model using six clinical variables, has been proposed to quantify the probability of OSA. Our aims are (1) to validate OSABayes prospectively, (2) to build a smartphone app based on the proposed model, and (3) to evaluate app usability. Methods: We prospectively included adult patients suspected of OSA, without suspicion of other sleep disorders, who underwent level I or III diagnostic PSG. Apnea-hypopnea index (AHI) and OSABayes probabilities were obtained and compared using the area under the ROC curve (AUC [95%CI]) for OSA diagnosis (AHI ≥ 5/h) and higher severity levels (AHI ≥ 15/h) prediction. We built the OSABayes app on 'App Inventor 2', and the usability was assessed with a cognitive walkthrough method and a general evaluation. Results: 216 subjects were included in the validation cohort, performing PSG levels I (34%) and III (66%). OSABayes presented an AUC of 83.6% [77.3-90.0%] for OSA diagnosis and 76.3% [69.9-82.7%] for moderate/severe OSA prediction, showing good response for both types of PSG. The OSABayes smartphone application allows one to calculate the probability of having OSA and consult information about OSA and the tool. In the usability evaluation, 96% of the proposed tasks were carried out. Conclusions: These results show the good discrimination power of OSABayes and validate its applicability in identifying patients with a high pre-test probability of OSA. The tool is available as an online form and as a smartphone app, allowing a quick and accessible calculation of OSA probability.

Letter to the Editor: Benchmarking clinical risk prediction algorithms with ensemble machine learning for the noninvasive diagnosis of liver fibrosis in NAFLD.

Letter to the Editor: Benchmarking clinical risk prediction algorithms with ensemble machine learning for the noninvasive diagnosis of liver fibrosis in NAFLD.

Benchmarking clinical risk prediction algorithms with ensemble machine learning for the noninvasive diagnosis of liver fibrosis in NAFLD.

Ensemble machine-learning methods, like the superlearner, combine multiple models into a single one to enhance predictive accuracy. Here we explore the potential of the superlearner as a benchmarking tool for clinical risk prediction, illustrating the approach to identifying significant liver fibrosis among patients with NAFLD. We used 23 demographic/clinical variables to train superlearner(s) on data from the NASH-clinical research network observational study (n = 648) and validated models with data from the FLINT trial (n = 270) and National Health and Nutrition Examination Survey (NHANES) participants with NAFLD (n = 1244). Comparing the superlearner's performance to existing models (Fibrosis-4 [FIB-4], NAFLD fibrosis score, Forns, AST to Platelet Ratio Index [APRI], BARD, and Steatosis-Associated Fibrosis Estimator [SAFE]), it exhibited strong discriminative ability in the FLINT and NHANES validation sets, with AUCs of 0.79 (95% CI: 0.73-0.84) and 0.74 (95% CI: 0.68-0.79) respectively. Notably, the SAFE score performed similarly to the superlearner, both of which outperformed FIB-4, APRI, Forns, and BARD scores in the validation data sets. Surprisingly, the superlearner derived from 12 base models matched the performance of one with 90 base models. Overall, the superlearner, being the "best-in-class" machine-learning predictor, excelled in detecting fibrotic NASH, and this approach can be used to benchmark the performance of conventional clinical risk prediction models.

754 Prediction of 30-day All-Cause Readmission of Neurosurgery Patients Using Large Language Models

INTRODUCTION: Existing clinical prediction algorithms mostly leverage small cohorts of structured data (e.g., medical imaging or laboratory data). However, large language models have demonstrated the ability to utilize unstructured data to outperform other machine learning approaches given sufficient data. Training large language models on unstructured clinical notes offers a possible alternative to structured data algorithm development in clinical tasks. METHODS: An unlabeled dataset of over seven million unstructured clinical notes (e.g., radiology reports and patient histories) was collected from four hospitals within the NYU Langone Health (NYULH) system and utilized to pre-train a bidirectional encoder representation with transformer (BERT) model. This model was further fine-tuned using a labeled dataset of discharge summaries to predict 30-day all-cause readmission. The resulting model, termed NYUTron, was assessed using a held-out retrospective cohort of patients from June to December 2021. RESULTS: Over the period of the retrospective study, there were a total 1,072 neurosurgery patients. The BERT model achieved an area under the receiver operating curve of 0.7883, a recall of 95.1% at a precision of 27.3%, and an accuracy of 82.1%. CONCLUSIONS: This study demonstrates how large language models and unstructured clinical notes can be used to provide information to physicians on clinical tasks with a flexible framework that is amenable to modification for other clinical tasks.

Prognostic Impact of Preoperative Thrombocytosis on Recurrence-Free Survival in Patients with Upper Tract Urothelial Carcinoma.

The aim of this work was to evaluate the prognostic potential of preoperative thrombocytosis for recurrence-free survival (RFS) and cancer-specific survival (CSS) among patients subjected to radical nephroureterectomy (RNU) due to UTUC. Analytical cohort was composed of a single-center series of 405 patients treated between January 1999 and December 2020. Thrombocytosis was defined as a platelet count exceeding the threshold value of 400 × 109 per L. Along with the Kaplan-Meier survival probability, Cox proportional hazard regression models were used. Preoperative thrombocytosis confirmed in 71 patients (17.5%) was significantly associated with the higher pathological tumor stage, lymph node metastasis, prior bladder cancer diagnosis, and preoperative anemia. With a median post-surgical follow-up period of 33.5 months, 125 patients (30.9%) experienced disease recurrence. The recurrence rate among patients with normal platelet levels was 13.6%, compared with 22.2% in those with preoperative thrombocytosis (p < 0.03). The 5-year RFS estimates reached 36.6% in the thrombocytosis-confirmed group. Multivariate analysis implied that preoperative thrombocytosis was a significant independent prognosticator of both poor RFS (HR 2.22, 95% CI 1.14-4.31, p = 0.02) and CSS (HR 2.48, 95% CI 1.14-3.09, p = 0.01). Patients with a clinically significant elevation of platelet count prior to RNU were more likely to have UTUC with advanced tumor stages and lymph node metastases. Preoperative thrombocytosis was an independent predictor of RFS and CSS in patients who underwent radical nephroureterectomy. Furthermore, preoperative thrombocytosis may complement and refine UTUC clinical prediction algorithms as an independent indicator of adverse survival outcomes.

Abstract 12129: Using a Polygenic Risk Score to Account for Genomic Risk Factors in a Model to Detect Individuals With Dilated Ascending Thoracic Aortas

Background: Ascending thoracic aortic dilation is a complex trait that involves modifiable and non-modifiable risk factors and can lead to thoracic aortic aneurysm and dissection. Clinical risk factors have been shown to predict ascending thoracic aortic diameter. Polygenic risk scores (PRS) are increasingly used to assess clinical risk for multifactorial diseases. The degree to which a PRS improves aortic diameter risk prediction is not known. In this study we tested the extent to which the addition of PRS to clinical prediction algorithms improves the prediction of aortic diameter. Methods: Training and validation cohorts comprised 6,790 Penn Medicine Biobank (PMBB) participants with clinical data linked to whole exome sequencing. Linear regression models integrated PRS weights derived from a large genome wide association study of thoracic aortic diameter in the UK biobank and were compared to the performance of the standard and a reweighted variation of the recently published AORTA Score. Results: The training cohort of 5,092 participants (56% male) had a median age of 61 years (IQR 52-70) with a mean ascending aortic diameter of 3.4 cm (SD 0.5). Compared to the AORTA Score which explained 29.1% (95% CI 27.9%-30.8%) of the variance in aortic diameter, AORTA Score + PRS (explained 29.7%, 95% CI 28.5%-30.9%), the reweighted AORTA score (explained 30.7%, 95% CI 29.4%-32.3%), and the reweighted AORTA Score + PRS (explained 31.5%, 95% CI 30.0%-32.8%) had superior performance. For the predictive ability to identify thoracic aortic diameter of ≥ 4 cm, the addition of a PRS to both the AORTA Score and the reweighted AORTA Score improved sensitivity at a range of model threshold values. The respective areas under the receiver operator characteristic curve for the AORTA Score + PRS (0.762, 95% CI 0.731-0.794) and reweighted AORTA Score + PRS (0.779, 95% CI 0.749-0.809) were greater than the standard AORTA Score (0.758, 95% CI 0.727-0.789) and reweighted AORTA Score (0.775, 95% CI 0.745-0.805). Conclusions: We demonstrated that inclusion of a PRS to the AORTA Score results in a small performance enhancement. Further investigation is necessary to determine if incorporation of genetics into clinical risk prediction improve outcomes for thoracic aortic disease.

Defining and evaluating potentially preventable ED visits in oncology patients including using early machine learning models.

463 Background: Cancer patients are frequent users of the emergency department (ED); preventing such visits is important for improving the quality of cancer care and decreasing health care system burden. This study describes oncology ED visits at a single institution, and seeks to define a novel definition of potentially preventable ED visits (PPEDs), particularly considering novel cancer therapies. It also uses machine learning (ML) to create early models to predict patients at highest risk of PPEDs. Methods: A retrospective study of ED visits by oncology patients between April 1st, 2019 and April 1st, 2021 at a single academic institution was completed. PPEDs were defined as ED visits resulting in discharge from the ED or an admission of less than 48 hours. PPEDs versus non-PPEDs were evaluated using descriptive statistics, logistic regression, and machine learning (ML) modelling. Results: 17% of cancer patients (n=6,689) had 13,415 ED visits during the study period. PPEDs differed in characteristics from non-PPEDs. Stage 1-3 patients with breast cancer, and those on IV systemic therapy were more likely to have PPEDs.Logistic regression models of pre-ED visit variables showed that stage, distance, and neoadjuvant systemic therapy were associated with PPEDs. Logistic regression of ED visit-related variables revealed that genitourinary, eye, or skin-related diagnoses were associated with PPEDs. The highest-performing ML model scored an AUC of 0.819, and combined pre-ED visit and ED visit related variables; stage 1-3 breast cancer patients, all IV chemotherapy patients, and stage 4 patients living within 15km of the ED, or with GI, oral or female gynecologic cancers were associated with PPEDs. ED diagnoses of “cancer” and MSK issues were associated with PPEDs. For those patients on IV systemic therapy specifically, having breast cancer, having a palliative consult within 60 days of the ED visit, or presenting with ED diagnoses of toxicities from therapy or blood disorders, pain, or rash were associated with PPEDs. Conclusions: Our novel definition of PPEDs versus non-PPEDs appears reasonable, and using machine learning has helped to describe groups of patients at highest likelihood of such visits at our institution. Future work to validate and advance these models into clinical prediction tools and algorithms could help proactively target high risk patients, and ideally decrease preventable ED visits in oncology.

Machine learning-based prediction of in-hospital mortality in patients with pneumonic chronic obstructive pulmonary disease exacerbations

Objective While linear regression and LASSO models have been established for predicting in-hospital mortality, there is currently no validated clinical prediction algorithm to predict in-hospital mortality for patients with chronic obstructive pulmonary disease (COPD) exacerbations using machine learning. Thus, we will evaluate the BAP-65 and CURB-65, and construct a novel prediction model using the random forest (RF) technique. Methods A dataset of 1,418 patients with COPD exacerbations was collected. Age, gender, mental status, vital signs, and laboratory results were all taken into account for predictors. The categorical outcome variable was hospital-based mortality of people over 65 years. The dataset was divided randomly into a training dataset (70%) and a testing dataset (30%). We trained three prediction models, BAP-65, CURB-65, and the RF model, estimated the area under the receiver operating characteristic curve (AUROC) for the entire dataset. We also conducted a comparison of the AUROC values using the Delong test. Results A total of 658 individuals with COPD acute exacerbations were enrolled. Our analysis using the receiver operating characteristic curve demonstrated that the RF model exhibited excellent performance, with an AUROC of 0.80 (95% confidence interval: 0.75-0.84). In comparison, the BAP-65 prediction model yielded an AUROC of 0.72 (0.68-0.75), while the CURB-65 prediction model achieved an AUROC of 0.69 (0.67-0.73). Conclusions The RF model demonstrated superior predictive capabilities than the BAP-65 and CURB-65 models in predicting in-hospital mortality. The results further highlighted significant factors for predicting in-hospital mortality, including blood eosinophil count, systolic blood pressure, and prior history of asthma.

Use of race in clinical algorithms.

To answer whether patients' race belongs in clinical prediction algorithms, two types of prediction models are considered: (i) diagnostic, which describes a patient's clinical characteristics, and (ii) prognostic, which forecasts a clinical risk or treatment effect that a patient is likely to experience in the future. The ex ante equality of opportunity framework is used, where specific health outcomes, which are prediction targets, evolve dynamically due to the effects of legacy levels of outcomes, circumstances, and current individual efforts. In practical settings, this study shows that failure to include race corrections will propagate systemic inequities and discrimination in any diagnostic model and specific prognostic models that inform decisions by invoking an ex ante compensation principle. In contrast, including race in prognostic models that inform resource allocations following an ex ante reward principle can compromise the equality of opportunities for patients from different races. Simulation results demonstrate these arguments.

Joint polygenic and environmental risks for childhood attention-deficit/hyperactivity disorder (ADHD) and ADHD symptom dimensions.

attention-deficit/hyperactivity disorder (ADHD) is associated with both polygenic liability and environmental exposures, both intrinsic to the family, such as family conflict, and extrinsic, such as air pollution. However, much less is known about the interplay between environmental and genetic risks relevant to ADHD-a better understanding of which could inform both mechanistic models and clinical prediction algorithms. Two independent data sets, the population-based Adolescent Brain Cognitive Development Study (ABCD) (N=11,876) and the case-control Oregon-ADHD-1000 (N=1449), were used to examine additive (G+E) and interactive (GxE) effects of selected polygenic risk scores (PRS) and environmental factors in a cross-sectional design. Genetic risk was measured using PRS for nine mental health disorders/traits. Exposures included family income, family conflict/negative sentiment, and geocoded measures of area deprivation, lead exposure risk, and air pollution exposure (nitrogen dioxide and fine particulate matter). ADHD PRS and family conflict jointly predicted concurrent ADHD symptoms in both cohorts. Additive-effects models, including both genetic and environmental factors, explained significantly more variation in symptoms than any individual factor alone (joint R 2=.091 for total symptoms in ABCD; joint R 2=.173 in Oregon-ADHD-1000; all delta-R 2 p-values <2e-7). Significant effect size heterogeneity across ancestry groups was observed for genetic and environmental factors (e.g., Q=9.01, p=.011 for major depressive disorder PRS; Q=13.34, p=.001 for area deprivation). GxE interactions observed in the full ABCD cohort suggested stronger environmental effects when genetic risk is low, though they did not replicate. Reproducible additive effects of PRS and family environment on ADHD symptoms were found, but GxE interaction effects were not replicated and appeared confounded by ancestry. Results highlight the potential value of combining exposures and PRS in clinical prediction algorithms. The observed differences in risks across ancestry groups warrant further study to avoid health care disparities.

Association of Continuously Measured Vital Signs With Respiratory Insufficiency in Hospitalized COVID-19 Patients: Retrospective Cohort Study.

Continuous monitoring of vital signs has the potential to assist in the recognition of deterioration of patients admitted to the general ward. However, methods to efficiently process and use continuously measured vital sign data remain unclear. The aim of this study was to explore methods to summarize continuously measured vital sign data and evaluate their association with respiratory insufficiency in COVID-19 patients at the general ward. In this retrospective cohort study, we included patients admitted to a designated COVID-19 cohort ward equipped with continuous vital sign monitoring. We collected continuously measured data of respiratory rate, heart rate, and oxygen saturation. For each patient, 7 metrics to summarize vital sign data were calculated: mean, slope, variance, occurrence of a threshold breach, number of episodes, total duration, and area above/under a threshold. These summary measures were calculated over timeframes of either 4 or 8 hours, with a pause between the last data point and the endpoint (the "lead") of 4, 2, 1, or 0 hours, and with 3 predefined thresholds per vital sign. The association between each of the summary measures and the occurrence of respiratory insufficiency was calculated using logistic regression analysis. Of the 429 patients that were monitored, 334 were included for analysis. Of these, 66 (19.8%) patients developed respiratory insufficiency. Summarized continuously measured vital sign data in timeframes close to the endpoint showed stronger associations than data measured further in the past (ie, lead 0 vs 1, 2, or 4 hours), and summarized estimates over 4 hours of data had stronger associations than estimates taken over 8 hours of data. The mean was consistently strongly associated with respiratory insufficiency for the three vital signs: in a 4-hour timeframe without a lead, the standardized odds ratio for heart rate, respiratory rate, and oxygen saturation was 2.59 (99% CI 1.74-4.04), 5.05 (99% CI 2.87-10.03), and 3.16 (99% CI 1.78-6.26), respectively. The strength of associations of summary measures varied per vital sign, timeframe, and lead. The mean of a vital sign showed a relatively strong association with respiratory insufficiency for the majority of vital signs and timeframes. The type of vital sign, length of the timeframe, and length of the lead influenced the strength of associations. Highly associated summary measures and their combinations could be used in a clinical prediction score or algorithm for an automatic alarm system.

Enabling Early Obstructive Sleep Apnea Diagnosis With Machine Learning: Systematic Review.

BackgroundAmerican Academy of Sleep Medicine guidelines suggest that clinical prediction algorithms can be used to screen patients with obstructive sleep apnea (OSA) without replacing polysomnography, the gold standard.ObjectiveWe aimed to identify, gather, and analyze existing machine learning approaches that are being used for disease screening in adult patients with suspected OSA.MethodsWe searched the MEDLINE, Scopus, and ISI Web of Knowledge databases to evaluate the validity of different machine learning techniques, with polysomnography as the gold standard outcome measure and used the Prediction Model Risk of Bias Assessment Tool (Kleijnen Systematic Reviews Ltd) to assess risk of bias and applicability of each included study.ResultsOur search retrieved 5479 articles, of which 63 (1.15%) articles were included. We found 23 studies performing diagnostic model development alone, 26 with added internal validation, and 14 applying the clinical prediction algorithm to an independent sample (although not all reporting the most common discrimination metrics, sensitivity or specificity). Logistic regression was applied in 35 studies, linear regression in 16, support vector machine in 9, neural networks in 8, decision trees in 6, and Bayesian networks in 4. Random forest, discriminant analysis, classification and regression tree, and nomogram were each performed in 2 studies, whereas Pearson correlation, adaptive neuro-fuzzy inference system, artificial immune recognition system, genetic algorithm, supersparse linear integer models, and k-nearest neighbors algorithm were each performed in 1 study. The best area under the receiver operating curve was 0.98 (0.96-0.99) for age, waist circumference, Epworth Somnolence Scale score, and oxygen saturation as predictors in a logistic regression.ConclusionsAlthough high values were obtained, they still lacked external validation results in large cohorts and a standard OSA criteria definition.Trial RegistrationPROSPERO CRD42021221339; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=221339

DeepBackRib: Deep learning to understand factors associated with readmissions after rib fractures.

Deep neural networks yield high predictive performance, yet obscure interpretability limits clinical applicability. We aimed to build an explainable deep neural network that elucidates factors associated with readmissions after rib fractures among nonelderly adults, termed DeepBackRib . We hypothesized that DeepBackRib could accurately predict readmissions and a game theoretic approach to elucidate how predictions are made would facilitate model explainability. We queried the 2017 National Readmissions Database for index hospitalization encounters of adults aged 18 to 64 years hospitalized with multiple rib fractures. The primary outcome was 3-month readmission(s). Study cohort was split 60-20-20 into training-validation-test sets. Model input features included demographic/injury/index hospitalization characteristics and index hospitalization International Classification of Diseases, Tenth Revision , diagnosis codes. The seven-layer DeepBackRib comprised multipronged strategies to mitigate overfitting and was trained to optimize recall. Shapley additive explanation analysis identified the marginal contribution of each input feature for predicting readmissions. A total of 20,260 patients met the inclusion criteria, among whom 11% (n = 2,185) experienced 3-month readmissions. Feature selection narrowed 3,164 candidate input features to 61, and DeepBackRib yielded 91%, 85%, and 82% recall on the training, validation, and test sets, respectively. Shapley additive explanation analysis quantified the marginal contribution of each input feature in determining DeepBackRib's predictions: underlying chronic obstructive pulmonary disease and long index hospitalization length of stay had positive associations with 3-month readmissions, while private primary payer and diagnosis of pneumothorax during index admission had negative associations. We developed and internally validated a high-performing deep learning algorithm that elucidates factors associated with readmissions after rib fractures. Despite promising predictive performance, standalone deep learning algorithms are insufficient for clinical prediction tasks: a concerted effort is needed to ensure that clinical prediction algorithms remain explainable. Prognostic and Epidemiological; Level III.

Artificial Intelligence, Machine Learning, and Deep Learning in Health Care

Artificial Intelligence (AI) and related technologies are widely used in business and community, and are integrated into health care. This technology has the potential to transform many aspects of patient care, as well as management functions within providers, payers, and pharmaceutical companies. A growing number of research shows that artificial intelligence (AI) can work better or better than humans in important health care tasks such as diagnostics. As machine learning (ML) models become more common in our daily lives, concerns about their potential for injury are growing. In medicine, optimism in the human-health function of ML is influenced by ethical concerns about the potential of tools to magnify existing health disparities. Recent research, for example, has found that when applied to women, minority races, and ethnic groups, as well as those with social insurance, modern clinical prediction algorithms do well. When popular language language models are trained in science textbooks, they complete the templates of clinical notes to compliment hospitals for violent white fathers, according to other studies. Conclusion: While in-depth learning as a means of analysis and modeling has some advantages at the moment, it also highlights a major trend: the integration of health sciences with data. Obstacles to access to health care will be higher than many other industries, where billions of people use in-depth learning and other forms of electronic learning software every day.

Prediction of Sudden Cardiac Death Manifesting With Documented Ventricular Fibrillation or Pulseless Ventricular Tachycardia.

This study aimed to develop a novel clinical prediction algorithm for avertable sudden cardiac death. Sudden cardiac death manifests as ventricular fibrillation (VF)/ ventricular tachycardia (VT) potentially treatable with defibrillation, or nonshockable rhythms (pulseless electrical activity/asystole) with low likelihood of survival. There are no available clinical risk scores for targeted prediction of VF/VT. Subjects with out-of-hospital sudden cardiac arrest presenting with documented VF or pulseless VT (33% of total cases) were ascertained prospectively from the Portland, Oregon, metro area with population ≈1 million residents (n=1,374, 2002-2019). Comparisons of lifetime clinical records were conducted with a control group (n=1,600) with ≈70% coronary disease prevalence. Prediction models were constructed from a training dataset using backwards stepwise logistic regression and applied to an internal validation dataset. Receiver operating characteristic curves (Cstatistic) were used to evaluate model discrimination. External validation was performed in a separate, geographically distinct population (Ventura County, California, population ≈850,000, 2015-2020). A clinical algorithm (VFRisk) constructed with 13 clinical, electrocardiogram, and echocardiographic variableshad very good discrimination in the training dataset (C statistic=0.808; [95% CI: 0.774-0.842]) and was successfully validated in internal (C statistic=0.776 [95% CI: 0.725-0.827]) and external (C statistic=0.782 [95% CI: 0.718-0.846]) datasets. The algorithm substantially outperformed the left ventricular ejection fraction (LVEF)≤35% (Cstatistic=0.638) and performed well across the LVEF spectrum. An algorithm for prediction of sudden cardiac arrest manifesting with VF/VT was successfully constructed using widely available clinical and noninvasive markers. These findings have potential to enhance primaryprevention, especially in patients with mid-range or preserved LVEF.

Predictive Algorithms in the Diagnosis and Management of Pediatric Hip and Periarticular Infection.

➤ Although the criteria of Kocher et al. were an important advancement in our ability to diagnose septic arthritis of the hip early, the changing microbial landscape and availability of advanced imaging have rendered it insufficient for contemporary clinical use.➤ Routine use of magnetic resonance imaging and recognition of disseminated disease have prompted the development of algorithms to predict concurrent osteoarticular infection in cases of septic arthritis and osteomyelitis that were previously assumed to be "isolated."➤ Recent research has attempted to stratify childhood bone and joint infection (BJI) by severity to guide treatment planning. This is valuable, as patients with multifocal disease, more virulent pathogens, and immunocompromise can have longer hospital stays and require multiple surgeries.➤ The increasing prevalence of clinical prediction algorithms in childhood BJI is not completely matched by quality in methodology. Clinicians need to be wary of adopting predictive algorithms prior to robust external validation.

Machine Learning Reimagined: The Promise of Interpretability to Combat Bias.

Machine Learning Reimagined: The Promise of Interpretability to Combat Bias.

Diagnostic Criteria for the Painful Swollen Pediatric Knee: Distinguishing Septic Arthritis From Aseptic Effusion in a Non-Lyme Endemic Area.

Purpose: The child with a painful swollen knee must be worked-up for possible septic arthritis; the classic clinical prediction algorithms for septic arthritis of the hip may not be the best models to apply to the knee.Materials and methods: This was a retrospective case-control study of 17 years of children presenting to one hospital with a chief complaint of a painful swollen knee, to evaluate the appropriateness of applying a previously described clinical practice algorithm for the hip in differentiating between the septic and aseptic causes of the painful knee effusions. The diagnoses of true septic arthritis, presumed septic arthritis, and aseptic effusion were established, based upon the cultures of synovial fluid, blood cultures, synovial cell counts, and clinical course. Using a logistic regression model, the disease status was regressed on both the demographic and clinical variables.Results: In the study, 122 patients were included: 51 with true septic arthritis, 37 with presumed septic arthritis, and 34 with aseptic knee effusion. After applying a backward elimination, age <5 years and C-reactive protein (CRP) >2.0 mg/dl remained in the model, and predicted probabilities of having septic knee arthritis ranged from 15% for the lowest risk to 95% for the highest risk. Adding a knee aspiration including percent polymorphonucleocytes (%PMN) substantially improved the overall model performance, lowering the lowest risk to 11% while raising the highest risk to 96%.Conclusions: This predictive model suggests that the likelihood of pediatric septic arthritis of the knee is >90% when both “age <5 years” and “CRP > 2.0 mg/dl” are present in a child with a painful swollen knee, though, in the absence of these factors, the risk of septic arthritis remains over 15%. Aspiration of the knee for those patients would be the best next step.

Using Combinations of Both Clinical and Radiographic Parameters to Develop a Diagnostic Prediction Model Demonstrated an Excellent Performance in Early Detection of Patients with Blount's Disease.

Early identification of pathological causes for pediatric genu varum (bowlegs) is crucial for preventing a progressive, irreversible knee deformity of the child. This study aims to develop and validate a diagnostic clinical prediction algorithm for assisting physicians in distinguishing an early stage of Blount’s disease from the physiologic bowlegs to provide an early treatment that could prevent the progressive, irreversible deformity. The diagnostic prediction model for differentiating an early stage of Blount’s disease from the physiologic bowlegs was developed under a retrospective case-control study from 2000 to 2017. Stepwise backward elimination of multivariable logistic regression modeling was used to derive a diagnostic model. A total of 158 limbs from 79 patients were included. Of those, 84 limbs (53.2%) were diagnosed as Blount’s disease. The final model that included age, BMI, MDA, and MMB showed excellent performance (area under the receiver operating characteristic (AuROC) curve: 0.85, 95% confidence interval 0.79 to 0.91) with good calibration. The proposed diagnostic prediction model for discriminating an early stage of Blount’s disease from physiologic bowlegs showed high discriminative ability with minimal optimism.

Simultaneous Pulmonary Artery Pressure and Left Ventricle Stroke Volume Assessment Predicts Adverse Events in Patients With Pulmonary Embolism.

BackgroundCertain echocardiographic parameters may serve as early predictors of adverse events in patients with hemodynamically compromising pulmonary embolism (PE).Methods and ResultsAn observational analysis was conducted for patients with acute pulmonary embolism evaluated by a Pulmonary Embolism Response Team (PERT) between 2014 and 2020. The performance of clinical prediction algorithms including the Pulmonary Embolism Severity Index and Carl Bova score were compared using a ratio of right ventricle and left ventricle hemodynamics by dividing the pulmonary artery systolic pressure by the left ventricle stroke volume. The primary outcome of in‐hospital mortality, cardiac arrest, and the need for advanced therapies was evaluated by univariate and multivariable analyses. Of the 343 patients meeting the inclusion criteria, 215 had complete data. Pulmonary artery systolic pressure/left ventricle stroke volume was a clear predictor of the primary end point (odds ratio [OR], 2.31; P=0.005), performing as well or better than the Pulmonary Embolism Severity Index (OR, 1.43; P=0.06) or the Bova score (OR, 1.28; P=0.01).ConclusionsThis study is the first study to demonstrate the utility of early pulmonary artery systolic pressure/left ventricle stroke volume in predicting adverse clinical events in patients with acute pulmonary embolism. Pulmonary artery systolic pressure/left ventricle stroke volume may be a surrogate marker of ventricular asynchrony in high‐risk pulmonary embolism and should be prognostically evaluated.