Early Prediction of Acute Kidney Injury in the Emergency Department With Machine-Learning Methods Applied to Electronic Health Record Data

Predicting acute kidney injury in trauma using an extreme gradient boosting model

Background

Acute kidney injury (AKI) is a significant complication in patients with trauma. The early identification of AKI in these patients poses challenges. This study aimed to predict AKI in trauma patients 24 or 48 hours in advance using an extreme gradient boosting (XGBoost) model.

Methods

We analyzed 17 859 trauma patients admitted to a regional trauma center between January 2015 and July 2023. Demographic, clinical, and laboratory parameters were collected. The model was developed using data until July 2021 and validated using data from August 2021. We developed models to predict AKI stages 1-3 and AKI stages 2 and 3 occurring 48 and 24 hours later and measured predictive performance in the validation group. The models' performance was evaluated using the area under the receiver operating characteristic curve (AUROC), and feature importance was assessed through SHapley Additive exPlanations values.

Results

The study population exhibited an incidence of AKI of 6.6% in the development group and 5.4% in the validation group. The models demonstrated predictive performance with AUROCs of 0.864 and 0.886 for 48-hour predictions of AKI stages 1-3 and stages 2 and 3, and 0.904 and 0.903 for 24-hour predictions of AKI stages 1-3 and stages 2 and 3, respectively. Key features influencing model predictions included baseline and in-hospital serum creatinine values, injury severity score, age, lactate dehydrogenase, D-dimer, platelets, albumin, and C-reactive protein levels.

Conclusions

The XGBoost models effectively predicted AKI in trauma patients up to 48 hours in advance using clinical data.

Identifying and forecasting importation and asymptomatic spreaders of multi-drug resistant organisms in hospital settings

Healthcare-associated infections (HAIs) from multi-drug resistant organisms (MDROs) pose a significant challenge for healthcare systems. Patients can arrive at hospitals already infected ("importation") or acquire infections during their stay ("nosocomial infection"). Many cases, often asymptomatic, complicate rapid identification due to testing limitations and delays. Although recent advancements in mathematical modeling and machine learning have aimed to identify at-risk patients, these methods face challenges: transmission models often overlook valuable electronic health record (EHR) data, while machine learning approaches typically lack mechanistic insights into underlying processes. To address these issues, we propose NeurABM, a novel framework that integrates neural networks and agent-based models (ABM) to leverage the strengths of both methods. NeurABM simultaneously learns a neural network for patient-level importation predictions and an ABM for infection identification. Our findings show that NeurABM significantly outperforms existing methods, marking a breakthrough in accurately identifying importation cases and forecasting future nosocomial infections in clinical practice.

Identifying and Forecasting Importation and Asymptomatic Spreaders of Multi-drug Resistant Organisms in Hospital Settings

Healthcare-associated infections (HAIs) from multi-drug resistant organisms (MDROs) pose a significant challenge for healthcare systems. Patients can arrive at hospitals already infected ("importation") or acquire infections during their stay ("nosocomial infection"). Many cases, often asymptomatic, complicate rapid identification due to testing limitations and delays. Although recent advancements in mathematical modeling and machine learning have aimed to identify at-risk patients, these methods face challenges: transmission models often overlook valuable electronic health record (EHR) data, while machine learning approaches typically lack mechanistic insights into underlying processes. To address these issues, we propose NeurABM, a novel framework that integrates neural networks and agent-based models (ABM) to leverage the strengths of both methods. NeurABM simultaneously learns a neural network for patient-level importation predictions and an ABM for infection identification. Our findings show that NeurABM significantly outperforms existing methods, marking a breakthrough in accurately identifying importation cases and forecasting future nosocomial infections in clinical practice.

Methicillin-resistant Staphylococcus aureus and Vancomycin Prescribing in the Emergency Department: A Single-center Study Assessing Antibiotic Prescribing

Objectives

Given the support for methicillin-resistant Staphylococcus aureus (MRSA) antimicrobial stewardship in the 2021 Surviving Sepsis Campaign Guidelines, we sought to measure the use of vancomycin in the emergency department (ED) in the years preceding these recommendations.

Methods

A retrospective cohort study was conducted of all patients aged ≥ 18 years presenting to 5 emergency departments within a university-based health system who were given intravenous (IV) vancomycin during their ED index visit. The primary outcome assessed the proportion of patients with MRSA-positive blood cultures who received IV vancomycin in the ED. We also measured associations between clinical attributes associated with any MRSA infection.

Results

Of the 20,212 unique ED visits for patients who received IV vancomycin, 63% (n = 12,755) had at least 1 MRSA risk factor. Only 2.4% (n = 494) and 14.1% (n = 2850) of patients receiving IV vancomycin in the ED were found to have MRSA bacteremia or any MRSA-positive culture, respectively. A total of 3160 patients met Sepsis-3 criteria and received IV vancomycin, though 65% (n = 2064) had no MRSA risk factors. For any patient with culture-proven MRSA, 63.8% (n = 315) and 43.4% (n = 1236) received an MRSA antimicrobial in the ED. MRSA risk factors were not associated with MRSA bacteremia (≥1 MRSA risk factor: odds ratio, 1.3, 95% CI, 0.9-1.8) or an MRSA-positive culture of any type (odds ratio, 0.9, 95% CI, 0.7-1.1).

Conclusion

Within our hospital system, MRSA was an infrequent cause of bacteremia for patients presenting to the ED with sepsis or septic shock. Although vancomycin is frequently used in the ED, many patients with culture-proven MRSA did not receive MRSA antimicrobials. Notably, one-third of patients with culture-proven MRSA had no MRSA risk factors. MRSA risk factors were not predictive of culture-proven MRSA, thus highlighting the complexity of antimicrobial stewardship in the ED without validated clinical decision rules.

Urinary biomarkers for diagnosing acute kidney injury in sepsis in the emergency department

Background

Development of acute kidney injury (AKI) in patients with sepsis is associated with increased mortality, highlighting the importance of early detection and management. However, baseline creatinine or urine output measurements are required for AKI diagnosis, which can be challenging in emergency departments (EDs). We aimed to evaluate the association between urinary biomarkers and the AKI diagnosis or 30-day survival status in patients with sepsis in the ED.

Methods

This prospective observational study enrolled patients from a single ED. We enrolled adult patients presenting to the ED with symptoms suggestive of infection and an initial quick sequential organ failure assessment score ≥2. Initial urine samples were collected, and urinary biomarkers (dickkopf-3, soluble triggering receptor expressed on myeloid cells-1, kidney injury molecule-1, neutrophil gelatinase-associated lipocalin (NGAL), and tissue inhibitor of metalloproteinases-2 (TIMP-2), and insulin-like growth factor binding protein-7 (IGFBP-7), and TIMP-2 × IGFBP-7) were analyzed using an enzyme-linked immunosorbent assay kit. Multivariable logistic regression models were used to evaluate biomarker performance.

Results

Of 84 patients, 63 (75.0 %) were diagnosed with AKI and 16 (19.0 %) died within 30 days. None of the urinary biomarkers demonstrated significant differences between the survivors and non-survivors. NGAL (p = 0.014) and TIMP-2 × IGFBP-7 (p = 0.027) levels were different between the AKI and non-AKI groups. The multivariable logistic regression model suggested a higher area under the receiver operating characteristic curve for models, including TIMP-2 × IGFBP-7 (from 0.853 to 0.889, p = 0.018).

Conclusion

None of the urinary biomarkers in the initial urine sample demonstrated an independent association with AKI diagnosis or 30-day survival status in patients with sepsis presenting to the ED. Further studies with larger population are necessary to confirm its clinical utility and explore its role.

A two-tier feature selection method for predicting mortality risk in ICU patients with acute kidney injury

Acute kidney injury (AKI) is one of the most important lethal factors for patients admitted to intensive care units (ICUs), and timely high-risk prognostic assessment and intervention are essential to improving patient prognosis. In this study, a stacking model using the MIMIC-III dataset with a two-tier feature selection approach was developed to predict the risk of in-hospital mortality in ICU patients admitted for AKI. External validation was performed using separate MIMIC-IV and eICU-CRD. The area under the curve (AUC) was calculated using the stacking model, and features were selected using the Boruta and XGBoost feature selection methods. This study compares the performance of a stacking model using two-tier feature selection with a model using single-tier feature selection (XGBoost: 85; Boruta: 83; two-tier: 0.91). The predictive effectiveness of the stacking model was further validated by using different datasets (Validation 1: 0.83; Validation 2: 0.85) and comparing it with a simpler model and traditional clinical scores (SOFA: 0.65; APACH IV: 0.61). In addition, this study combined interpretable techniques and causal inference to analyze the causal relationship between features and predicted outcomes.

Association between Pan-Immune-Inflammation Value and Contrast-Induced Nephropathy with Coronary Angiography

Background: Contrast-induced nephropathy (CIN) is one of the most important complications after invasive cardiovascular procedures. Considering the pivotal role of inflammation in CIN development, the use of peripheral blood-based indexes may be an easily available biomarker to predict CIN risk. Therefore, in the present study, we evaluated the association between the pan-immune-inflammation value (PIV) and the risk of CIN. Patients and Methods: A total of 1343 patients undergoing coronary angiography (CAG) were included. The PIV was calculated with the following equation: (neutrophil count × platelet count × monocyte count)/lymphocyte count. Multivariable regression analyses were used to determine the association between clinical and laboratory parameters and CIN development. Results: The median age of the cohort was 58 (IQR 50-67), and 48.2% of the patients were female. CIN developed in 202 patients (15%) in follow-up. In multivariate analyses, older age (OR: 1.015, 95% CI: 1.002-1.028, p = 0.020) and higher PIV levels (OR: 1.016, 95% CI: 1.004-1.028, p = 0.008) were associated with a higher CIN risk, while the use of antiplatelet agents was associated with a lower risk of CIN (OR: 0.670, 95% CI: 0.475-0.945, p = 0.022). Conclusions: We demonstrated that the risk of CIN was significantly higher in patients with higher PIV and older patients in a large cohort of patients undergoing CAG for stable ischemic heart disease. If supported with prospective evidence, PIV levels could be used as a minimally invasive reflector of CIN.

Biomarkers vs Machines: The Race to Predict Acute Kidney Injury

BACKGROUND

Acute kidney injury (AKI) is a serious complication affecting up to 15% of hospitalized patients. Early diagnosis is critical to prevent irreversible kidney damage that could otherwise lead to significant morbidity and mortality. However, AKI is a clinically silent syndrome, and current detection primarily relies on measuring a rise in serum creatinine, an imperfect marker that can be slow to react to developing AKI. Over the past decade, new innovations have emerged in the form of biomarkers and artificial intelligence tools to aid in the early diagnosis and prediction of imminent AKI.

CONTENT

This review summarizes and critically evaluates the latest developments in AKI detection and prediction by emerging biomarkers and artificial intelligence. Main guidelines and studies discussed herein include those evaluating clinical utilitiy of alternate filtration markers such as cystatin C and structural injury markers such as neutrophil gelatinase-associated lipocalin and tissue inhibitor of metalloprotease 2 with insulin-like growth factor binding protein 7 and machine learning algorithms for the detection and prediction of AKI in adult and pediatric populations. Recommendations for clinical practices considering the adoption of these new tools are also provided.

SUMMARY

The race to detect AKI is heating up. Regulatory approval of select biomarkers for clinical use and the emergence of machine learning algorithms that can predict imminent AKI with high accuracy are all promising developments. But the race is far from being won. Future research focusing on clinical outcome studies that demonstrate the utility and validity of implementing these new tools into clinical practice is needed.

Forecasting acute kidney injury and resource utilization in ICU patients using longitudinal, multimodal models

BACKGROUND

Advances in artificial intelligence (AI) have realized the potential of revolutionizing healthcare, such as predicting disease progression via longitudinal inspection of Electronic Health Records (EHRs) and lab tests from patients admitted to Intensive Care Units (ICU). Although substantial literature exists addressing broad subjects, including the prediction of mortality, length-of-stay, and readmission, studies focusing on forecasting Acute Kidney Injury (AKI), specifically dialysis anticipation like Continuous Renal Replacement Therapy (CRRT) are scarce. The technicality of how to implement AI remains elusive.

OBJECTIVE

This study aims to elucidate the important factors and methods that are required to develop effective predictive models of AKI and CRRT for patients admitted to ICU, using EHRs in the Medical Information Mart for Intensive Care (MIMIC) database.

METHODS

We conducted a comprehensive comparative analysis of established predictive models, considering both time-series measurements and clinical notes from MIMIC-IV databases. Subsequently, we proposed a novel multi-modal model which integrates embeddings of top-performing unimodal models, including Long Short-Term Memory (LSTM) and BioMedBERT, and leverages both unstructured clinical notes and structured time series measurements derived from EHRs to enable the early prediction of AKI and CRRT.

RESULTS

Our multimodal model achieved a lead time of at least 12 h ahead of clinical manifestation, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.888 for AKI and 0.997 for CRRT, as well as an Area Under the Precision Recall Curve (AUPRC) of 0.727 for AKI and 0.840 for CRRT, respectively, which significantly outperformed the baseline models. Additionally, we performed a SHapley Additive exPlanation (SHAP) analysis using the expected gradients algorithm, which highlighted important, previously underappreciated predictive features for AKI and CRRT.

CONCLUSION

Our study revealed the importance and the technicality of applying longitudinal, multimodal modeling to improve early prediction of AKI and CRRT, offering insights for timely interventions. The performance and interpretability of our model indicate its potential for further assessment towards clinical applications, to ultimately optimize AKI management and enhance patient outcomes.

Multisite development and validation of machine learning models to predict severe outcomes and guide decision-making for emergency department patients with influenza

Objective

Millions of Americans are infected by influenza annually. A minority seek care in the emergency department (ED) and, of those, only a limited number experience severe disease or death. ED clinicians must distinguish those at risk for deterioration from those who can be safely discharged.

Methods

We developed random forest machine learning (ML) models to estimate needs for critical care within 24 h and inpatient care within 72 h in ED patients with influenza. Predictor data were limited to those recorded prior to ED disposition decision: demographics, ED complaint, medical problems, vital signs, supplemental oxygen use, and laboratory results. Our study population was comprised of adults diagnosed with influenza at one of five EDs in our university health system between January 1, 2017 and May 18, 2022; visits were divided into two cohorts to facilitate model development and validation. Prediction performance was assessed by the area under the receiver operating characteristic curve (AUC) and the Brier score.

Results

Among 8032 patients with laboratory-confirmed influenza, incidence of critical care needs was 6.3% and incidence of inpatient care needs was 19.6%. The most common reasons for ED visit were symptoms of respiratory tract infection, fever, and shortness of breath. Model AUCs were 0.89 (95% CI 0.86-0.93) for prediction of critical care and 0.90 (95% CI 0.88-0.93) for inpatient care needs; Brier scores were 0.026 and 0.042, respectively. Importantpredictors included shortness of breath, increasing respiratory rate, and a high number of comorbid diseases.

Conclusions

ML methods can be used to accurately predict clinical deterioration in ED patients with influenza and have potential to support ED disposition decision-making.

Forecasting Acute Kidney Injury and Resource Utilization in ICU patients using longitudinal, multimodal models

Background

Advances in artificial intelligence (AI) have realized the potential of revolutionizing healthcare, such as predicting disease progression via longitudinal inspection of Electronic Health Records (EHRs) and lab tests from patients admitted to Intensive Care Units (ICU). Although substantial literature exists addressing broad subjects, including the prediction of mortality, length-of-stay, and readmission, studies focusing on forecasting Acute Kidney Injury (AKI), specifically dialysis anticipation like Continuous Renal Replacement Therapy (CRRT) are scarce. The technicality of how to implement AI remains elusive.

Objective

This study aims to elucidate the important factors and methods that are required to develop effective predictive models of AKI and CRRT for patients admitted to ICU, using EHRs in the Medical Information Mart for Intensive Care (MIMIC) database.

Methods

We conducted a comprehensive comparative analysis of established predictive models, considering both time-series measurements and clinical notes from MIMIC-IV databases. Subsequently, we proposed a novel multi-modal model which integrates embeddings of top-performing unimodal models, including Long Short-Term Memory (LSTM) and BioMedBERT, and leverages both unstructured clinical notes and structured time series measurements derived from EHRs to enable the early prediction of AKI and CRRT.

Results

Our multimodal model achieved a lead time of at least 12 hours ahead of clinical manifestation, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.888 for AKI and 0.997 for CRRT, as well as an Area Under the Precision Recall Curve (AUPRC) of 0.727 for AKI and 0.840 for CRRT, respectively, which significantly outperformed the baseline models. Additionally, we performed a SHapley Additive exPlanation (SHAP) analysis using the expected gradients algorithm, which highlighted important, previously underappreciated predictive features for AKI and CRRT.

Conclusion

Our study revealed the importance and the technicality of applying longitudinal, multimodal modeling to improve early prediction of AKI and CRRT, offering insights for timely interventions. The performance and interpretability of our model indicate its potential for further assessment towards clinical applications, to ultimately optimize AKI management and enhance patient outcomes.

Automated machine learning for early prediction of acute kidney injury in acute pancreatitis

BACKGROUND

Acute kidney injury (AKI) represents a frequent and grave complication associated with acute pancreatitis (AP), substantially elevating both mortality rates and the financial burden of hospitalization. The aim of our study is to construct a predictive model utilizing automated machine learning (AutoML) algorithms for the early prediction of AKI in patients with AP.

METHODS

We retrospectively analyzed patients who were diagnosed with AP in our hospital from January 2017 to December 2021. These patients were randomly allocated into a training set and a validation set at a ratio of 7:3. To develop predictive models for each set, we employed the least absolute shrinkage and selection operator (LASSO) algorithm along with AutoML. A nomogram was developed based on multivariate logistic regression analysis outcomes. The model's efficacy was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Additionally, the performance of the model constructed via AutoML was evaluated using decision curve analysis (DCA), feature importance, SHapley Additive exPlanations (SHAP) plots, and locally interpretable model-agnostic explanations (LIME).

RESULTS

This study incorporated a total of 437 patients who met the inclusion criteria. Out of these, 313 were assigned to the training cohort and 124 to the validation cohort. In the training and validation cohorts, AKI occurred in 68 (21.7%) and 29(23.4%) patients, respectively. Comparative analysis revealed that the AutoML models exhibited enhanced performance over traditional logistic regression (LR). Furthermore, the deep learning (DL) model demonstrated superior predictive accuracy, evidenced by an area under the ROC curve of 0.963 in the training set and 0.830 in the validation set, surpassing other comparative models. The key variables identified as significant in the DL model within the training dataset included creatinine (Cr), urea (Urea), international normalized ratio (INR), etiology, smoking, alanine aminotransferase (ALT), hypertension, prothrombin time (PT), lactate dehydrogenase (LDH), and diabetes.

CONCLUSION

The AutoML model, utilizing DL algorithm, offers considerable clinical significance in the early detection of AKI among patients with AP.

Digital health and acute kidney injury: consensus report of the 27th Acute Disease Quality Initiative workgroup

Acute kidney injury (AKI), which is a common complication of acute illnesses, affects the health of individuals in community, acute care and post-acute care settings. Although the recognition, prevention and management of AKI has advanced over the past decades, its incidence and related morbidity, mortality and health care burden remain overwhelming. The rapid growth of digital technologies has provided a new platform to improve patient care, and reports show demonstrable benefits in care processes and, in some instances, in patient outcomes. However, despite great progress, the potential benefits of using digital technology to manage AKI has not yet been fully explored or implemented in clinical practice. Digital health studies in AKI have shown variable evidence of benefits, and the digital divide means that access to digital technologies is not equitable. Upstream research and development costs, limited stakeholder participation and acceptance, and poor scalability of digital health solutions have hindered their widespread implementation and use. Here, we provide recommendations from the Acute Disease Quality Initiative consensus meeting, which involved experts in adult and paediatric nephrology, critical care, pharmacy and data science, at which the use of digital health for risk prediction, prevention, identification and management of AKI and its consequences was discussed.

Predicting in-hospital outcomes of patients with acute kidney injury

Acute kidney injury (AKI) is prevalent and a leading cause of in-hospital death worldwide. Early prediction of AKI-related clinical events and timely intervention for high-risk patients could improve outcomes. We develop a deep learning model based on a nationwide multicenter cooperative network across China that includes 7,084,339 hospitalized patients, to dynamically predict the risk of in-hospital death (primary outcome) and dialysis (secondary outcome) for patients who developed AKI during hospitalization. A total of 137,084 eligible patients with AKI constitute the analysis set. In the derivation cohort, the area under the receiver operator curve (AUROC) for 24-h, 48-h, 72-h, and 7-day death are 95·05%, 94·23%, 93·53%, and 93·09%, respectively. For dialysis outcome, the AUROC of each time span are 88·32%, 83·31%, 83·20%, and 77·99%, respectively. The predictive performance is consistent in both internal and external validation cohorts. The model can predict important outcomes of patients with AKI, which could be helpful for the early management of AKI.

Characterization of Risk Prediction Models for Acute Kidney Injury: A Systematic Review and Meta-analysis

Importance

Despite the expansion of published prediction models for acute kidney injury (AKI), there is little evidence of uptake of these models beyond their local derivation nor data on their association with patient outcomes.

Objective

To systematically review published AKI prediction models across all clinical subsettings.

Data Sources

MEDLINE via PubMed (January 1946 to April 2021) and Embase (January 1947 to April 2021) were searched using medical subject headings and text words related to AKI and prediction models.

Study Selection

All studies that developed a prediction model for AKI, defined as a statistical model with at least 2 predictive variables to estimate future occurrence of AKI, were eligible for inclusion. There was no limitation on study populations or methodological designs.

Data Extraction and Synthesis

Two authors independently searched the literature, screened the studies, and extracted and analyzed the data following the Preferred Reporting Items for Systematic Review and Meta-analyses guideline. The data were pooled using a random-effects model, with subgroups defined by 4 clinical settings. Between-study heterogeneity was explored using multiple methods, and funnel plot analysis was used to identify publication bias.

Main Outcomes and Measures

C statistic was used to measure the discrimination of prediction models.

Results

Of the 6955 studies initially identified through literature searching, 150 studies, with 14.4 million participants, met the inclusion criteria. The study characteristics differed widely in design, population, AKI definition, and model performance assessments. The overall pooled C statistic was 0.80 (95% CI, 0.79-0.81), with pooled C statistics in different clinical subsettings ranging from 0.78 (95% CI, 0.75-0.80) to 0.82 (95% CI, 0.78-0.86). Between-study heterogeneity was high overall and in the different clinical settings (eg, contrast medium-associated AKI: I2 = 99.9%; P < .001), and multiple methods did not identify any clear sources. A high proportion of models had a high risk of bias (126 [84.4%]) according to the Prediction Model Risk Of Bias Assessment Tool.

Conclusions and Relevance

In this study, the discrimination of the published AKI prediction models was good, reflected by high C statistics; however, the wide variation in the clinical settings, populations, and predictive variables likely drives the highly heterogenous findings that limit clinical utility. Standardized procedures for development and validation of prediction models are urgently needed.

Patient Embeddings From Diagnosis Codes for Health Care Prediction Tasks: Pat2Vec Machine Learning Framework

BACKGROUND

In health care, diagnosis codes in claims data and electronic health records (EHRs) play an important role in data-driven decision making. Any analysis that uses a patient's diagnosis codes to predict future outcomes or describe morbidity requires a numerical representation of this diagnosis profile made up of string-based diagnosis codes. These numerical representations are especially important for machine learning models. Most commonly, binary-encoded representations have been used, usually for a subset of diagnoses. In real-world health care applications, several issues arise: patient profiles show high variability even when the underlying diseases are the same, they may have gaps and not contain all available information, and a large number of appropriate diagnoses must be considered.

OBJECTIVE

We herein present Pat2Vec, a self-supervised machine learning framework inspired by neural network-based natural language processing that embeds complete diagnosis profiles into a small real-valued numerical vector.

METHODS

Based on German outpatient claims data with diagnosis codes according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10), we discovered an optimal vectorization embedding model for patient diagnosis profiles with Bayesian optimization for the hyperparameters. The calibration process ensured a robust embedding model for health care-relevant tasks by aggregating the metrics of different regression and classification tasks using different machine learning algorithms (linear and logistic regression as well as gradient-boosted trees). The models were tested against a baseline model that binary encodes the most common diagnoses. The study used diagnosis profiles and supplementary data from more than 10 million patients from 2016 to 2019 and was based on the largest German ambulatory claims data set. To describe subpopulations in health care, we identified clusters (via density-based clustering) and visualized patient vectors in 2D (via dimensionality reduction with uniform manifold approximation). Furthermore, we applied our vectorization model to predict prospective drug prescription costs based on patients' diagnoses.

RESULTS

Our final models outperform the baseline model (binary encoding) with equal dimensions. They are more robust to missing data and show large performance gains, particularly in lower dimensions, demonstrating the embedding model's compression of nonlinear information. In the future, other sources of health care data can be integrated into the current diagnosis-based framework. Other researchers can apply our publicly shared embedding model to their own diagnosis data.

CONCLUSIONS

We envision a wide range of applications for Pat2Vec that will improve health care quality, including personalized prevention and signal detection in patient surveillance as well as health care resource planning based on subcohorts identified by our data-driven machine learning framework.

Predictive models in emergency medicine and their missing data strategies: a systematic review

In the field of emergency medicine (EM), the use of decision support tools based on artificial intelligence has increased markedly in recent years. In some cases, data are omitted deliberately and thus constitute "data not purposely collected" (DNPC). This accepted information bias can be managed in various ways: dropping patients with missing data, imputing with the mean, or using automatic techniques (e.g., machine learning) to handle or impute the data. Here, we systematically reviewed the methods used to handle missing data in EM research. A systematic review was performed after searching PubMed with the query "(emergency medicine OR emergency service) AND (artificial intelligence OR machine learning)". Seventy-two studies were included in the review. The trained models variously predicted diagnosis in 25 (35%) publications, mortality in 21 (29%) publications, and probability of admission in 21 (29%) publications. Eight publications (11%) predicted two outcomes. Only 15 (21%) publications described their missing data. DNPC constitute the "missing data" in EM machine learning studies. Although DNPC have been described more rigorously since 2020, the descriptions in the literature are not exhaustive, systematic or homogeneous. Imputation appears to be the best strategy but requires more time and computational resources. To increase the quality and the comparability of studies, we recommend inclusion of the TRIPOD checklist in each new publication, summarizing the machine learning process in an explicit methodological diagram, and always publishing the area under the receiver operating characteristics curve-even when it is not the primary outcome.

Renal outcomes following intravenous contrast administration in patients with acute kidney injury: a multi-site retrospective propensity-adjusted analysis

PURPOSE

Evidence of an association between intravenous contrast media (CM) and persistent renal dysfunction is lacking for patients with pre-existing acute kidney injury (AKI). This study was designed to determine the association between intravenous CM administration and persistent AKI in patients with pre-existing AKI.

METHODS

A retrospective propensity-weighted and entropy-balanced observational cohort analysis of consecutive hospitalized patients ≥ 18 years old meeting Kidney Disease Improving Global Outcomes (KDIGO) creatinine-based criteria for AKI at time of arrival to one of three emergency departments between 7/1/2017 and 6/30/2021 who did or did not receive intravenous CM. Outcomes included persistent AKI at hospital discharge and initiation of dialysis within 180 days of index encounter.

RESULTS

Our analysis included 14,449 patient encounters, with 12.8% admitted to the intensive care unit (ICU). CM was administered in 18.4% of all encounters. AKI resolved prior to hospital discharge for 69.1%. No association between intravenous CM administration and persistent AKI was observed after unadjusted multivariable logistic regression modeling (OR 1; 95% CI 0.89-1.11), propensity weighting (OR 0.93; 95% CI 0.83-1.05), and entropy balancing (OR 0.94; 95% CI 0.83-1.05). Sub-group analysis in those admitted to the ICU yielded similar results. Initiation of dialysis within 180 days was observed in 5.4% of the cohort. An association between CM administration and increased risk of dialysis within 180 days was not observed.

CONCLUSION

Among patients with pre-existing AKI, contrast administration was not associated with either persistent AKI at hospital discharge or initiation of dialysis within 180 days. Current consensus recommendations for use of intravenous CM in patients with stable renal disease may also be applied to patients with pre-existing AKI.

Predicting hospital admission from emergency department triage data for patients presenting with fall-related fractures

PURPOSE

Predict in advance the need for hospitalization of adult patients for fall-related fractures based on information available at the time of triage to help decision-making at the emergency department (ED).

METHODS

We developed machine learning models using routinely collected triage data at a regional hospital chain in Pennsylvania to predict admission to an inpatient unit. We considered all patients presenting to the ED for fall-related fractures. Patients who were 18 years or younger, who left the ED against medical advice, left the ED waiting room without being seen by a provider, and left the ED after initial diagnostics were excluded from the analysis. We compared models obtained using triage data (pre-model) with models developed using additional data obtained after physicians' diagnoses (post-model).

RESULTS

Our results show good discriminatory power on predicting hospital admissions. Neural network models performed the best (AUC: pre-model = 0.938 [CI 0.920-0.956], post-model = 0.983 [0.974-0.992]). The logistic regression analysis provides additional insights into the data and the relationships between the variables.

CONCLUSIONS

Using limited data available at the time of triage, we developed four machine learning models aimed at predicting hospitalization for patients presenting to the ED for fall-related fractures. All the four models were robust and performed well. Neural network method, however, performed the best for both pre- and post-models. Simple, parsimonious machine learning models can provide high accuracy for predicting hospital admission.

Machine learning models for predicting acute kidney injury: a systematic review and critical appraisal

BACKGROUND

The number of studies applying machine learning (ML) to predict acute kidney injury (AKI) has grown steadily over the past decade. We assess and critically appraise the state of the art in ML models for AKI prediction, considering performance, methodological soundness, and applicability.

METHODS

We searched PubMed and ArXiv, extracted data, and critically appraised studies based on the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD), Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS), and Prediction Model Risk of Bias Assessment Tool (PROBAST) guidelines.

RESULTS

Forty-six studies from 3166 titles were included. Thirty-eight studies developed a model, five developed and externally validated one, and three studies externally validated one. Flexible ML methods were used more often than deep learning, although the latter was common with temporal variables and text as predictors. Predictive performance showed an area under receiver operating curves ranging from 0.49 to 0.99. Our critical appraisal identified a high risk of bias in 39 studies. Some studies lacked internal validation, whereas external validation and interpretability of results were rarely considered. Fifteen studies focused on AKI prediction in the intensive care setting, and the US-derived Medical Information Mart for Intensive Care (MIMIC) data set was commonly used. Reproducibility was limited as data and code were usually unavailable.

CONCLUSIONS

Flexible ML methods are popular for the prediction of AKI, although more complex models based on deep learning are emerging. Our critical appraisal identified a high risk of bias in most models: Studies should use calibration measures and external validation more often, improve model interpretability, and share data and code to improve reproducibility.

Opportunities in digital health and electronic health records for acute kidney injury care

PURPOSE OF REVIEW

The field of digital health is evolving rapidly with applications relevant to the prediction, detection and management of acute kidney injury (AKI). This review will summarize recent publications in these areas.

RECENT FINDINGS

Machine learning (ML) approaches have been applied predominantly for AKI prediction, but also to identify patients with AKI at higher risk of adverse outcomes, and to discriminate different subgroups (subphenotypes) of AKI. There have been multiple publications in this area, but a smaller number of ML models have robust external validation or the ability to run in real-time in clinical systems. Recent studies of AKI alerting systems and clinical decision support systems continue to demonstrate variable results, which is likely to result from differences in local context and implementation strategies. In the design of AKI alerting systems, choice of baseline creatinine has a strong effect on performance of AKI detection algorithms.

SUMMARY

Further research is required to overcome barriers to the validation and implementation of ML models for AKI care. Simpler electronic systems within the electronic medical record can lead to improved care in some but not all settings, and careful consideration of local context and implementation strategy is recommended.

Machine learning and artificial intelligence: applications in healthcare epidemiology

Artificial intelligence (AI) refers to the performance of tasks by machines ordinarily associated with human intelligence. Machine learning (ML) is a subtype of AI; it refers to the ability of computers to draw conclusions (ie, learn) from data without being directly programmed. ML builds from traditional statistical methods and has drawn significant interest in healthcare epidemiology due to its potential for improving disease prediction and patient care. This review provides an overview of ML in healthcare epidemiology and practical examples of ML tools used to support healthcare decision making at 4 stages of hospital-based care: triage, diagnosis, treatment, and discharge. Examples include model-building efforts to assist emergency department triage, predicting time before septic shock onset, detecting community-acquired pneumonia, and classifying COVID-19 disposition risk level. Increasing availability and quality of electronic health record (EHR) data as well as computing power provides opportunities for ML to increase patient safety, improve the efficiency of clinical management, and reduce healthcare costs.

Machine Learning for Acute Kidney Injury Prediction in the Intensive Care Unit

Machine learning is the field of artificial intelligence in which computers are trained to make predictions or to identify patterns in data through complex mathematical algorithms. It has great potential in critical care to predict outcomes, such as acute kidney injury, and can be used for prognosis and to suggest management strategies. Machine learning can also be used as a research tool to advance our clinical and biochemical understanding of acute kidney injury. In this review, we introduce basic concepts in machine learning and review recent research in each of these domains.

Development and validation of a nomogram for predicting overall survival in cirrhotic patients with acute kidney injury

BACKGROUND

Acute kidney injury (AKI) is a common and severe complication in patients with cirrhosis, and is associated with poor prognosis. Therefore, identifying cirrhotic patients with AKI who are at high risk of mortality is very important and may be helpful for providing timely medical interventions to improve the prognosis of these patients. However, studies focused on investigating the risk factors for the mortality of cirrhotic patients with AKI were scarce.

AIM

To identify risk factors for mortality and establish a nomogram for predicting the prognosis of these patients.

METHODS

Two hundred fifty consecutive patients with cirrhosis and AKI were recruited and randomly divided into training cohort (n = 173) and validation cohort (n = 77). In the training cohort, potential risk factors for death were identified by performing a Cox regression analysis, and a nomogram was established. The predictive performance of the nomogram was internally and externally validated by calculating the area under the receiver operating characteristic curve (AUROC), constructing a calibration curve and performing decision curve analysis.

RESULTS

The serum sodium level, international normalized ratio, peak serum creatinine level > 1.5 mg/dL, the presence of hepatic encephalopathy and diabetes were potential risk factors for mortality of cirrhotic patients with AKI in the training dataset. A prognostic nomogram incorporating these variables was established for predicting the overall survival of these patients. Compared with Child-Turcotte-Pugh, the model for end-stage liver disease (MELD) and the MELD-Na scores, the nomogram in predicting 90- and 180-d mortality exhibited better discriminatory power with AUROCs of 0.792 and 0.801 for the training dataset and 0.817 and 0.862 for the validation dataset, respectively. With a nomogram score of 98, patients were divided into low- and high-risk groups, and high-risk patients had a higher mortality rate.

CONCLUSION

A prognostic nomogram displayed good performance for predicting the overall survival of cirrhotic patients with AKI, and will assist clinicians in evaluating the prognosis of these patients.

Multisite implementation of a workflow-integrated machine learning system to optimize COVID-19 hospital admission decisions

Demand has outstripped healthcare supply during the coronavirus disease 2019 (COVID-19) pandemic. Emergency departments (EDs) are tasked with distinguishing patients who require hospital resources from those who may be safely discharged to the community. The novelty and high variability of COVID-19 have made these determinations challenging. In this study, we developed, implemented and evaluated an electronic health record (EHR) embedded clinical decision support (CDS) system that leverages machine learning (ML) to estimate short-term risk for clinical deterioration in patients with or under investigation for COVID-19. The system translates model-generated risk for critical care needs within 24 h and inpatient care needs within 72 h into rapidly interpretable COVID-19 Deterioration Risk Levels made viewable within ED clinician workflow. ML models were derived in a retrospective cohort of 21,452 ED patients who visited one of five ED study sites and were prospectively validated in 15,670 ED visits that occurred before (n = 4322) or after (n = 11,348) CDS implementation; model performance and numerous patient-oriented outcomes including in-hospital mortality were measured across study periods. Incidence of critical care needs within 24 h and inpatient care needs within 72 h were 10.7% and 22.5%, respectively and were similar across study periods. ML model performance was excellent under all conditions, with AUC ranging from 0.85 to 0.91 for prediction of critical care needs and 0.80-0.90 for inpatient care needs. Total mortality was unchanged across study periods but was reduced among high-risk patients after CDS implementation.

Influence of artificial intelligence on the work design of emergency department clinicians a systematic literature review

Objective

This systematic literature review aims to demonstrate how Artificial Intelligence (AI) is currently used in emergency departments (ED) and how it alters the work design of ED clinicians. AI is still new and unknown to many healthcare professionals in emergency care, leading to unfamiliarity with its capabilities.

Method

Various criteria were used to establish the suitability of the articles to answer the research question. This study was based on 34 selected peer-reviewed papers on the use of Artificial Intelligence (AI) in the Emergency Department (ED), published in the last five years. Drawing on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, all articles were scanned, read full-text, and analyzed afterward.

Results

The majority of the AI applications consisted of AI-based tools to aid with clinical decisions and to relieve overcrowded EDs of their burden. AI support was mostly offered during triage, the moment that sets the patient trajectory. There is ample evidence that AI-based applications could improve the clinical decision-making process.

Conclusion

The use of AI in EDs is still in its nascent stages. Many studies focus on the question of whether AI has clinical utility, such as decision support, improving resource allocation, reducing diagnostic errors, and promoting proactivity. Some studies suggest that AI-based tools essentially have the ability to outperform human skills. However, it is evident from the literature that current technology does not have the aims or power to do so. Nevertheless, AI-based tools can impact clinician work design in the ED by providing support with clinical decisions, which could ultimately help alleviate a portion of the increasing clinical burden.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-022-08070-7.

Development and validation of an interpretable clinical score for early identification of acute kidney injury at the emergency department

Acute kidney injury (AKI) in hospitalised patients is a common syndrome associated with poorer patient outcomes. Clinical risk scores can be used for the early identification of patients at risk of AKI. We conducted a retrospective study using electronic health records of Singapore General Hospital emergency department patients who were admitted from 2008 to 2016. The primary outcome was inpatient AKI of any stage within 7 days of admission based on the Kidney Disease Improving Global Outcome (KDIGO) 2012 guidelines. A machine learning-based framework AutoScore was used to generate clinical scores from the study sample which was randomly divided into training, validation and testing cohorts. Model performance was evaluated using area under the curve (AUC). Among the 119,468 admissions, 10,693 (9.0%) developed AKI. 8491 were stage 1 (79.4%), 906 stage 2 (8.5%) and 1296 stage 3 (12.1%). The AKI Risk Score (AKI-RiSc) was a summation of the integer scores of 6 variables: serum creatinine, serum bicarbonate, pulse, systolic blood pressure, diastolic blood pressure, and age. AUC of AKI-RiSc was 0.730 (95% CI 0.714–0.747), outperforming an existing AKI Prediction Score model which achieved AUC of 0.665 (95% CI 0.646–0.679) on the testing cohort. At a cut-off of 4 points, AKI-RiSc had a sensitivity of 82.6% and specificity of 46.7%. AKI-RiSc is a simple clinical score that can be easily implemented on the ground for early identification of AKI and potentially be applied in international settings.

Nouvelles approches diagnostiques de l’insuffisance rénale aiguë

L’insuffisance rénale, véritable problème de santé publique, concernerait plus de 82 000 personnes en France. On estime que 5 à 10 % de la population française souffriraient d’une maladie rénale pouvant conduire à une insuffisance rénale avec un taux de mortalité de plus de 10 % par an. À l’inverse de la maladie rénale chronique irréversible, l’insuffisance rénale aiguë est considérée comme un dysfonctionnement transitoire et réversible. Au-delà de l’intérêt de la prévention, le diagnostic précoce de l’insuffisance rénale aiguë permettrait de mettre en place des thérapeutiques adaptées et ciblées afin d’éviter l’évolution vers des lésions rénales irréversibles. Cependant, il demeure un véritable challenge pour le praticien puisque l’on présume que près de 10 % de la population française présenteraient des lésions rénales asymptomatiques. Bien que la définition de l’insuffisance rénale aiguë ait été simplifiée durant ces dernières années, il existe de nombreuses limites. En parallèle, des progrès majeurs ont été réalisés notamment en termes de diagnostic. L’objectif de cette mise au point est de faire un rappel sur l’évolution de l’insuffisance rénale aiguë, les définitions actuelles et de présenter les nouvelles approches diagnostiques en cours de développement.

The incidence, mortality and renal outcomes of acute kidney injury in patients with suspected infection at the emergency department

Background

Acute kidney injury (AKI) is a major health problem associated with considerable mortality and morbidity. Studies on clinical outcomes and mortality of AKI in the emergency department are scarce. The aim of this study is to assess incidence, mortality and renal outcomes after AKI in patients with suspected infection at the emergency department.

Methods

We used data from the SPACE-cohort (SePsis in the ACutely ill patients in the Emergency department), which included consecutive patients that presented to the emergency department of the internal medicine with suspected infection. Hazard ratios (HR) were assessed using Cox regression to investigate the association between AKI, 30-days mortality and renal function decline up to 1 year after AKI. Survival in patients with and without AKI was assessed using Kaplan-Meier analyses.

Results

Of the 3105 patients in the SPACE-cohort, we included 1716 patients who fulfilled the inclusion criteria. Of these patients, 10.8% had an AKI episode. Mortality was 12.4% for the AKI group and 4.2% for the non-AKI patients. The adjusted HR for all-cause mortality at 30-days in AKI patients was 2.8 (95% CI 1.7–4.8). Moreover, the cumulative incidence of renal function decline was 69.8% for AKI patients and 39.3% for non-AKI patients. Patients with an episode of AKI had higher risk of developing renal function decline (adjusted HR 3.3, 95% CI 2.4–4.5) at one year after initial AKI-episode at the emergency department.

Conclusion

Acute kidney injury is common in patients with suspected infection in the emergency department and is significantly associated with 30-days mortality and renal function decline one year after AKI.

Does Artificial Intelligence Make Clinical Decision Better? A Review of Artificial Intelligence and Machine Learning in Acute Kidney Injury Prediction

Acute kidney injury (AKI) is a common complication of hospitalization that greatly and negatively affects the short-term and long-term outcomes of patients. Current guidelines use serum creatinine level and urine output rate for defining AKI and as the staging criteria of AKI. However, because they are not sensitive or specific markers of AKI, clinicians find it difficult to predict the occurrence of AKI and prescribe timely treatment. Advances in computing technology have led to the recent use of machine learning and artificial intelligence in AKI prediction, recent research reported that by using electronic health records (EHR) the AKI prediction via machine-learning models can reach AUROC over 0.80, in some studies even reach 0.93. Our review begins with the background and history of the definition of AKI, and the evolution of AKI risk factors and prediction models is also appraised. Then, we summarize the current evidence regarding the application of e-alert systems and machine-learning models in AKI prediction.

Machine Learning Models for Predicting In-Hospital Mortality in Acute Aortic Dissection Patients

Background: Acute aortic dissection is a potentially fatal cardiovascular disorder associated with high mortality. However, current predictive models show a limited ability to efficiently and flexibly detect this mortality risk, and have been unable to discover a relationship between the mortality rate and certain variables. Thus, this study takes an artificial intelligence approach, whereby clinical data-driven machine learning was utilized to predict the in-hospital mortality of acute aortic dissection. Methods: Patients diagnosed with acute aortic dissection between January 2015 to December 2018 were voluntarily enrolled from the Second Xiangya Hospital of Central South University in the study. The diagnosis was defined by magnetic resonance angiography or computed tomography angiography, with an onset time of the symptoms being within 14 days. The analytical variables included demographic characteristics, physical examination, symptoms, clinical condition, laboratory results, and treatment strategies. The machine learning algorithms included logistic regression, decision tree, K nearest neighbor, Gaussian naive bayes, and extreme gradient boost (XGBoost). Evaluation of the predictive performance of the models was mainly achieved using the area under the receiver operating characteristic curve. SHapley Additive exPlanation was also implemented to interpret the final prediction model. Results: A total of 1,344 acute aortic dissection patients were recruited, including 1,071 (79.7%) patients in the survivor group and 273 (20.3%) patients in non-survivor group. The extreme gradient boost model was found to be the most effective model with the greatest area under the receiver operating characteristic curve (0.927, 95% CI: 0.860-0.968). The three most significant aspects of the extreme gradient boost importance matrix plot were treatment, type of acute aortic dissection, and ischemia-modified albumin levels. In the SHapley Additive exPlanation summary plot, medical treatment, type A acute aortic dissection, and higher ischemia-modified albumin level were shown to increase the risk of hospital-based mortality.

Machine learning in patient flow: a review

This work is a review of the ways in which machine learning has been used in order to plan, improve or aid the problem of moving patients through healthcare services. We decompose the patient flow problem into four subcategories: prediction of demand on a healthcare institution, prediction of the demand and resource required to transfer patients from the emergency department to the hospital, prediction of potential resource required for the treatment and movement of inpatients and prediction of length-of-stay and discharge timing. We argue that there are benefits to both approaches of considering the healthcare institution as a whole as well as the patient by patient case and that ideally a combination of these would be best for improving patient flow through hospitals. We also argue that it is essential for there to be a shared dataset that will allow researchers to benchmark their algorithms on and thereby allow future researchers to build on that which has already been done. We conclude that machine learning for the improvement of patient flow is still a young field with very few papers tailor-making machine learning methods for the problem being considered. Future works should consider the need to transfer algorithms trained on a dataset to multiple hospitals and allowing for dynamic algorithms which will allow real-time decision-making to help clinical staff on the shop floor.