AKIpredictor, an online prognostic calculator for acute kidney injury in adult critically ill patients: development, validation and comparison to serum neutrophil gelatinase-associated lipocalin

[Recent developments in acute kidney injury : Definition, biomarkers, subphenotypes, and management]

Acute kidney injury (AKI) is a common problem in critically ill patients and is associated with increased morbidity and mortality. Since 2012, AKI has been defined according to the KDIGO (Kidney Disease Improving Global Outcome) guidelines. As some biomarkers are now available that can provide useful clinical information, a new definition including a new stage 1S has been proposed by an expert group of the Acute Disease Quality Initiative (ADQI). At this stage, classic AKI criteria are not yet met, but biomarkers are already positive defining subclinical AKI. This stage 1S is associated with a worse patient outcome, regardless of the biomarker chosen. The PrevAKI and PrevAKI-Multicenter trial also showed that risk stratification with a biomarker and implementation of the KDIGO bundle (in the high-risk group) can reduce the rate of moderate and severe AKI. In the absence of a successful clinical trial, conservative management remains the primary focus of treatment. This mainly involves optimization of hemodynamics and an individualized (restrictive) fluid management. The STARRT-AKI trial has shown that there is no benefit from accelerated initiation of renal replacement therapy. However, delaying too long might be associated with potential harm, as shown in the AKIKI2 study. Prospective studies are needed to determine whether artificial intelligence will play a role in AKI in the future, helping to guide treatment decisions and improve outcomes.

Development and validation of a machine learning model to predict the use of renal replacement therapy in 14,374 patients with COVID-19

PURPOSE

To develop a model to predict the use of renal replacement therapy (RRT) in COVID-19 patients.

MATERIALS AND METHODS

Retrospective analysis of multicenter cohort of intensive care unit (ICU) admissions of Brazil involving COVID-19 critically adult patients, requiring ventilatory support, admitted to 126 Brazilian ICUs, from February 2020 to December 2021 (development) and January to May 2022 (validation). No interventions were performed.

RESULTS

Eight machine learning models' classifications were evaluated. Models were developed using an 80/20 testing/train split ratio and cross-validation. Thirteen candidate predictors were selected using the Recursive Feature Elimination (RFE) algorithm. Discrimination and calibration were assessed. Temporal validation was performed using data from 2022. Of 14,374 COVID-19 patients with initial respiratory support, 1924 (13%) required RRT. RRT patients were older (65 [53-75] vs. 55 [42-68]), had more comorbidities (Charlson's Comorbidity Index 1.0 [0.00-2.00] vs 0.0 [0.00-1.00]), had higher severity (SAPS-3 median: 61 [51-74] vs 48 [41-58]), and had higher in-hospital mortality (71% vs 22%) compared to non-RRT. Risk factors for RRT, such as Creatinine, Glasgow Coma Scale, Urea, Invasive Mechanical Ventilation, Age, Chronic Kidney Disease, Platelets count, Vasopressors, Noninvasive Ventilation, Hypertension, Diabetes, modified frailty index (mFI) and Gender, were identified. The best discrimination and calibration were found in the Random Forest (AUC [95%CI]: 0.78 [0.75-0.81] and Brier's Score: 0.09 [95%CI: 0.08-0.10]). The final model (Random Forest) showed comparable performance in the temporal validation (AUC [95%CI]: 0.79 [0.75-0.84] and Brier's Score, 0.08 [95%CI: 0.08-0.1]).

CONCLUSIONS

An early ML model using easily available clinical and laboratory data accurately predicted the use of RRT in critically ill patients with COVID-19. Our study demonstrates that using ML techniques is feasible to provide early prediction of use of RRT in COVID-19 patients.

[Artificial intelligence and acute kidney injury]

Digitalization is increasingly finding its way into intensive care units and with it artificial intelligence (AI) for critically ill patients. One promising area for the use of AI is in the field of acute kidney injury (AKI). The use of AI is primarily focused on the prediction of AKI, but further approaches are also being used to classify existing AKI into different phenotypes. Different AI models are used for prediction. The area under the receiver operating characteristic curve values (AUROC) achieved with these models vary and are influenced by several factors, such as the prediction time and the definition of AKI. Most models have an AUROC between 0.650 and 0.900, with lower values for predictions further into the future and when applying Acute Kidney Injury Network (AKIN) instead of KDIGO criteria. Classification into phenotypes already makes it possible to categorize patients into groups with different risks of mortality or requirement of renal replacement therapy (RRT), but the etiologies or therapeutic consequences derived from this are still lacking. However, all the models suffer from AI-specific shortcomings. The use of large databases does not make it possible to promptly include recent changes in therapy and the implementation of new biomarkers in a relevant proportion. For this reason, serum creatinine and urinary output, with their known limitations, dominate current AI models for prediction impairing the performance of the current models. On the other hand, the increasingly complex models no longer allow physicians to understand the basis on which the warning of a threatening AKI is calculated and subsequent initiation of therapy should take place. The successful use of AIs in routine clinical practice will be highly determined by the trust of the physicians in the systems and overcoming the aforementioned weaknesses. However, the clinician will remain irreplaceable as the decisive authority for critically ill patients by combining measurable and nonmeasurable parameters.

Endothelium-related biomarkers enhanced prediction of kidney support therapy in critically ill patients with non-oliguric acute kidney injury

Acute kidney injury (AKI) is a common condition in hospitalized patients who often requires kidney support therapy (KST). However, predicting the need for KST in critically ill patients remains challenging. This study aimed to analyze endothelium-related biomarkers as predictors of KST need in critically ill patients with stage 2 AKI. A prospective observational study was conducted on 127 adult ICU patients with stage 2 AKI by serum creatinine only. Endothelium-related biomarkers, including vascular cell adhesion protein-1 (VCAM-1), angiopoietin (AGPT) 1 and 2, and syndecan-1, were measured. Clinical parameters and outcomes were recorded. Logistic regression models, receiver operating characteristic (ROC) curves, continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used for analysis. Among the patients, 22 (17.2%) required KST within 72 h. AGPT2 and syndecan-1 levels were significantly greater in patients who progressed to the KST. Multivariate analysis revealed that AGPT2 and syndecan-1 were independently associated with the need for KST. The area under the ROC curve (AUC-ROC) for AGPT2 and syndecan-1 performed better than did the constructed clinical model in predicting KST. The combination of AGPT2 and syndecan-1 improved the discrimination capacity of predicting KST beyond that of the clinical model alone. Additionally, this combination improved the classification accuracy of the NRI and IDI. AGPT2 and syndecan-1 demonstrated predictive value for the need for KST in critically ill patients with stage 2 AKI. The combination of AGPT2 and syndecan-1 alone enhanced the predictive capacity of predicting KST beyond clinical variables alone. These findings may contribute to the early identification of patients who will benefit from KST and aid in the management of AKI in critically ill patients.

Identification and validation of an explainable prediction model of acute kidney injury with prognostic implications in critically ill children: a prospective multicenter cohort study

Background

Acute kidney injury (AKI) is a common and serious organ dysfunction in critically ill children. Early identification and prediction of AKI are of great significance. However, current AKI criteria are insufficiently sensitive and specific, and AKI heterogeneity limits the clinical value of AKI biomarkers. This study aimed to establish and validate an explainable prediction model based on the machine learning (ML) approach for AKI, and assess its prognostic implications in children admitted to the pediatric intensive care unit (PICU).

Methods

This multicenter prospective study in China was conducted on critically ill children for the derivation and validation of the prediction model. The derivation cohort, consisting of 957 children admitted to four independent PICUs from September 2020 to January 2021, was separated for training and internal validation, and an external data set of 866 children admitted from February 2021 to February 2022 was employed for external validation. AKI was defined based on serum creatinine and urine output using the Kidney Disease: Improving Global Outcome (KDIGO) criteria. With 33 medical characteristics easily obtained or evaluated during the first 24 h after PICU admission, 11 ML algorithms were used to construct prediction models. Several evaluation indexes, including the area under the receiver-operating-characteristic curve (AUC), were used to compare the predictive performance. The SHapley Additive exPlanation method was used to rank the feature importance and explain the final model. A probability threshold for the final model was identified for AKI prediction and subgrouping. Clinical outcomes were evaluated in various subgroups determined by a combination of the final model and KDIGO criteria.

Findings

The random forest (RF) model performed best in discriminative ability among the 11 ML models. After reducing features according to feature importance rank, an explainable final RF model was established with 8 features. The final model could accurately predict AKI in both internal (AUC = 0.929) and external (AUC = 0.910) validations, and has been translated into a convenient tool to facilitate its utility in clinical settings. Critically ill children with a probability exceeding or equal to the threshold in the final model had a higher risk of death and multiple organ dysfunctions, regardless of whether they met the KDIGO criteria for AKI.

Interpretation

Our explainable ML model was not only successfully developed to accurately predict AKI but was also highly relevant to adverse outcomes in individual children at an early stage of PICU admission, and it mitigated the concern of the "black-box" issue with an undirect interpretation of the ML technique.

Funding

The National Natural Science Foundation of China, Jiangsu Province Science and Technology Support Program, Key talent of women's and children's health of Jiangsu Province, and Postgraduate Research & Practice Innovation Program of Jiangsu Province.

Development and external validation of multimodal postoperative acute kidney injury risk machine learning models

Objectives

To develop and externally validate machine learning models using structured and unstructured electronic health record data to predict postoperative acute kidney injury (AKI) across inpatient settings.

Materials and Methods

Data for adult postoperative admissions to the Loyola University Medical Center (2009-2017) were used for model development and admissions to the University of Wisconsin-Madison (2009-2020) were used for validation. Structured features included demographics, vital signs, laboratory results, and nurse-documented scores. Unstructured text from clinical notes were converted into concept unique identifiers (CUIs) using the clinical Text Analysis and Knowledge Extraction System. The primary outcome was the development of Kidney Disease Improvement Global Outcomes stage 2 AKI within 7 days after leaving the operating room. We derived unimodal extreme gradient boosting machines (XGBoost) and elastic net logistic regression (GLMNET) models using structured-only data and multimodal models combining structured data with CUI features. Model comparison was performed using the receiver operating characteristic curve (AUROC), with Delong's test for statistical differences.

Results

The study cohort included 138 389 adult patient admissions (mean [SD] age 58 [16] years; 11 506 [8%] African-American; and 70 826 [51%] female) across the 2 sites. Of those, 2959 (2.1%) developed stage 2 AKI or higher. Across all data types, XGBoost outperformed GLMNET (mean AUROC 0.81 [95% confidence interval (CI), 0.80-0.82] vs 0.78 [95% CI, 0.77-0.79]). The multimodal XGBoost model incorporating CUIs parameterized as term frequency-inverse document frequency (TF-IDF) showed the highest discrimination performance (AUROC 0.82 [95% CI, 0.81-0.83]) over unimodal models (AUROC 0.79 [95% CI, 0.78-0.80]).

Discussion

A multimodality approach with structured data and TF-IDF weighting of CUIs increased model performance over structured data-only models.

Conclusion

These findings highlight the predictive power of CUIs when merged with structured data for clinical prediction models, which may improve the detection of postoperative AKI.

Expectation of clinical decision support systems: a survey study among nephrologist end-users

BACKGROUND

Chronic kidney disease (CKD), a major public health problem with differing disease etiologies, leads to complications, comorbidities, polypharmacy, and mortality. Monitoring disease progression and personalized treatment efforts are crucial for long-term patient outcomes. Physicians need to integrate different data levels, e.g., clinical parameters, biomarkers, and drug information, with medical knowledge. Clinical decision support systems (CDSS) can tackle these issues and improve patient management. Knowledge about the awareness and implementation of CDSS in Germany within the field of nephrology is scarce.

PURPOSE

Nephrologists' attitude towards any CDSS and potential CDSS features of interest, like adverse event prediction algorithms, is important for a successful implementation. This survey investigates nephrologists' experiences with and expectations towards a useful CDSS for daily medical routine in the outpatient setting.

METHODS

The 38-item questionnaire survey was conducted either by telephone or as a do-it-yourself online interview amongst nephrologists across all of Germany. Answers were collected and analysed using the Electronic Data Capture System REDCap, as well as Stata SE 15.1, and Excel. The survey consisted of four modules: experiences with CDSS (M1), expectations towards a helpful CDSS (M2), evaluation of adverse event prediction algorithms (M3), and ethical aspects of CDSS (M4). Descriptive statistical analyses of all questions were conducted.

RESULTS

The study population comprised 54 physicians, with a response rate of about 80-100% per question. Most participants were aged between 51-60 years (45.1%), 64% were male, and most participants had been working in nephrology out-patient clinics for a median of 10.5 years. Overall, CDSS use was poor (81.2%), often due to lack of knowledge about existing CDSS. Most participants (79%) believed CDSS to be helpful in the management of CKD patients with a high willingness to try out a CDSS. Of all adverse event prediction algorithms, prediction of CKD progression (97.8%) and in-silico simulations of disease progression when changing, e. g., lifestyle or medication (97.7%) were rated most important. The spectrum of answers on ethical aspects of CDSS was diverse.

CONCLUSION

This survey provides insights into experience with and expectations of out-patient nephrologists on CDSS. Despite the current lack of knowledge on CDSS, the willingness to integrate CDSS into daily patient care, and the need for adverse event prediction algorithms was high.

Prediction differences and implications of acute kidney injury with and without urine output criteria in adult critically ill patients

BACKGROUND

Due to the convenience of serum creatinine (SCr) monitoring and the relative complexity of urine output (UO) monitoring, most studies have predicted acute kidney injury (AKI) only based on SCr criteria. This study aimed to compare the differences between SCr alone and combined UO criteria in predicting AKI.

METHODS

We applied machine learning methods to evaluate the performance of 13 prediction models composed of different feature categories on 16 risk assessment tasks (half used only SCr criteria, half used both SCr and UO criteria). The area under receiver operator characteristic curve (AUROC), the area under precision recall curve (AUPRC) and calibration were used to assess the prediction performance.

RESULTS

In the first week after ICU admission, the prevalence of any AKI was 29% under SCr criteria alone and increased to 60% when the UO criteria was combined. Adding UO to SCr criteria can significantly identify more AKI patients. The predictive importance of feature types with and without UO was different. Using only laboratory data maintained similar predictive performance to the full feature model under only SCr criteria [e.g. for AKI within the 48-h time window after 1 day of ICU admission, AUROC (95% confidence interval) 0.83 (0.82, 0.84) vs 0.84 (0.83, 0.85)], but it was not sufficient when the UO was added [corresponding AUROC (95% confidence interval) 0.75 (0.74, 0.76) vs 0.84 (0.83, 0.85)].

CONCLUSIONS

This study found that SCr and UO measures should not be regarded as equivalent criteria for AKI staging, and emphasizes the importance and necessity of UO criteria in AKI risk assessment.

Continuous and early prediction of future moderate and severe Acute Kidney Injury in critically ill patients: Development and multi-centric, multi-national external validation of a machine-learning model

BACKGROUND

Acute Kidney Injury (AKI) is a major complication in patients admitted to Intensive Care Units (ICU), causing both clinical and economic burden on the healthcare system. This study develops a novel machine-learning (ML) model to predict, with several hours in advance, the AKI episodes of stage 2 and 3 (according to KDIGO definition) acquired in ICU.

METHODS

A total of 16'760 ICU adult patients from 145 different ICU centers and 3 different countries (US, Netherland, Italy) are retrospectively enrolled for the study. Every hour the model continuously analyzes the routinely-collected clinical data to generate a new probability of developing AKI stage 2 and 3, according to KDIGO definition, during the ICU stay.

RESULTS

The predictive model obtains an auROC of 0.884 for AKI (stage 2/3 KDIGO) prediction, when evaluated on the internal test set composed by 1'749 ICU stays from US and EU centers. When externally tested on a multi-centric US dataset of 6'985 ICU stays and multi-centric Italian dataset of 1'025 ICU stays, the model achieves an auROC of 0.877 and of 0.911, respectively. In all datasets, the time between model prediction and AKI (stage 2/3 KDIGO) onset is at least of 14 hours after the first day of ICU hospitalization.

CONCLUSIONS

In this study, a novel ML model for continuous and early AKI (stage 2/3 KDIGO) prediction is successfully developed, leveraging only routinely-available data. It continuously predicts AKI episodes during ICU stay, at least 14 hours in advance when the AKI episode happens after the first 24 hours of ICU admission. Its performances are validated in an extensive, multi-national and multi-centric cohort of ICU adult patients. This ML model overcomes the main limitations of currently available predictive models. The benefits of its real-world implementation enable an early proactive clinical management and the prevention of AKI episodes in ICU patients. Furthermore, the software could be directly integrated with IT system of the ICU.

Development and validation of the creatinine clearance predictor machine learning models in critically ill adults

BACKGROUND

In critically ill patients, measured creatinine clearance (CrCl) is the most reliable method to evaluate glomerular filtration rate in routine clinical practice and may vary subsequently on a day-to-day basis. We developed and externally validated models to predict CrCl one day ahead and compared them with a reference reflecting current clinical practice.

METHODS

A gradient boosting method (GBM) machine-learning algorithm was used to develop the models on data from 2825 patients from the EPaNIC multicenter randomized controlled trial database. We externally validated the models on 9576 patients from the University Hospitals Leuven, included in the M@tric database. Three models were developed: a "Core" model based on demographic, admission diagnosis, and daily laboratory results; a "Core + BGA" model adding blood gas analysis results; and a "Core + BGA + Monitoring" model also including high-resolution monitoring data. Model performance was evaluated against the actual CrCl by mean absolute error (MAE) and root-mean-square error (RMSE).

RESULTS

All three developed models showed smaller prediction errors than the reference. Assuming the same CrCl of the day of prediction showed 20.6 (95% CI 20.3-20.9) ml/min MAE and 40.1 (95% CI 37.9-42.3) ml/min RMSE in the external validation cohort, while the developed model having the smallest RMSE (the Core + BGA + Monitoring model) had 18.1 (95% CI 17.9-18.3) ml/min MAE and 28.9 (95% CI 28-29.7) ml/min RMSE.

CONCLUSIONS

Prediction models based on routinely collected clinical data in the ICU were able to accurately predict next-day CrCl. These models could be useful for hydrophilic drug dosage adjustment or stratification of patients at risk.

TRIAL REGISTRATION

Not applicable.

Predicting outcomes of acute kidney injury in critically ill patients using machine learning

Acute Kidney Injury (AKI) is a sudden episode of kidney failure that is frequently seen in critically ill patients. AKI has been linked to chronic kidney disease (CKD) and mortality. We developed machine learning-based prediction models to predict outcomes following AKI stage 3 events in the intensive care unit. We conducted a prospective observational study that used the medical records of ICU patients diagnosed with AKI stage 3. A random forest algorithm was used to develop two models that can predict patients who will progress to CKD after three and six months of experiencing AKI stage 3. To predict mortality, two survival prediction models have been presented using random survival forests and survival XGBoost. We evaluated established CKD prediction models using AUCROC, and AUPR curves and compared them with the baseline logistic regression models. The mortality prediction models were evaluated with an external test set, and the C-indices were compared to baseline COXPH. We included 101 critically ill patients who experienced AKI stage 3. To increase the training set for the mortality prediction task, an unlabeled dataset has been added. The RF (AUPR: 0.895 and 0.848) and XGBoost (c-index: 0.8248) models have a better performance than the baseline models in predicting CKD and mortality, respectively Machine learning-based models can assist clinicians in making clinical decisions regarding critically ill patients with severe AKI who are likely to develop CKD following discharge. Additionally, we have shown better performance when unlabeled data are incorporated into the survival analysis task.

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.

Application of urinary biomarkers for diagnosing acute kidney injury in critically ill patients without baseline renal function data

PURPOSE

Estimating the baseline renal function of patients without prior creatinine measurement is crucial for diagnosing acute kidney injury (AKI). This study aimed to incorporate AKI biomarkers into a new AKI diagnosis rule when no premorbid baseline is available.

METHODS

This prospective observational study was conducted in an adult intensive care unit (ICU). Urinary neutrophil gelatinase-associated lipocalin (NGAL) and L-type fatty acid-binding protein (L-FABP) were measured at ICU admission. An AKI diagnostic rule was composed by classification and regression tree (CART) analysis.

RESULTS

A total of 243 patients were enrolled. In the development cohort, CART analysis composed a decision tree for AKI diagnosis selecting serum creatinine and urinary NGAL at ICU admission as predictors. In the validation cohort, the novel decision rule was superior to the imputation strategy based on Modification of Diet in Renal Disease (MDRD) equation regarding misclassification rate (13.0% vs. 29.6%, p = 0.002). Decision curve analysis demonstrated that the net benefit of the decision rule exceeded the MDRD approach in a threshold probability range of 25% and higher.

CONCLUSIONS

The novel diagnostic rule incorporating serum creatinine and urinary NGAL at ICU admission showed superiority to the MDRD approach in AKI diagnosis without baseline renal function data.

Kidney Dysfunction After Traumatic Brain Injury: Pathophysiology and General Management

Traumatic brain injury (TBI) remains a major cause of mortality and morbidity, and almost half of these patients are admitted to the intensive care unit. Of those, 10% develop acute kidney injury (AKI) and 2% even need kidney replacement therapy (KRT). Although clinical trials in patients with TBI who have AKI are lacking, some general principles in this population may apply. The present review is an overview on the epidemiology and pathophysiology of AKI in patients with TBI admitted to the intensive care unit who are at risk for or who have developed AKI. A cornerstone in severe TBI management is preventing secondary brain damage, in which reducing the intracranial pressure (ICP) and optimizing the cerebral perfusion pressure (CPP) remain important therapeutic targets. To treat episodes of elevated ICP, osmolar agents such as mannitol and hypertonic saline are frequently administered. Although we are currently awaiting the results of a prospective randomized controlled trial that compares both agents, it is important to realize that both agents have been associated with an increased risk of developing AKI which is probably higher for mannitol compared with hypertonic saline. For the brain, as well as for the kidney, targeting an adequate perfusion pressure is important. Hemodynamic management based on the combined use of intravascular fluids and vasopressors is ideally guided by hemodynamic monitoring. Hypotonic albumin or crystalloid resuscitation solutions may increase the risk of brain edema, and saline-based solutions are frequently used but have a risk of hyperchloremia, which might jeopardize kidney function. In patients at risk, frequent assessment of serum chloride might be advised. Maintenance of an adequate CPP involves the optimization of circulating blood volume, often combined with vasopressor agents. Whether individualized CPP targets based on cerebrovascular autoregulation monitoring are beneficial need to be further investigated. Interestingly, such individualized perfusion targets are also under investigation in patients as a strategy to mitigate the risk for AKI in patients with chronic hypertension. In the small proportion of patients with TBI who need KRT, continuous techniques are advised based on pathophysiology and expert opinion. The need for KRT is associated with a higher risk of intracranial hypertension, especially if osmolar clearance occurs fast, which can even occur in continuous techniques. Precise ICP and CPP monitoring is mandatory, especially at the initiation of KRT.

Bibliometric and visual analysis of machine learning-based research in acute kidney injury worldwide

Background

Acute kidney injury (AKI) is a serious clinical complication associated with adverse short-term and long-term outcomes. In recent years, with the rapid popularization of electronic health records and artificial intelligence machine learning technology, the detection rate and treatment of AKI have been greatly improved. At present, there are many studies in this field, and a large number of articles have been published, but we do not know much about the quality of research production in this field, as well as the focus and trend of current research.

Methods

Based on the Web of Science Core Collection, studies reporting machine learning-based AKI research that were published from 2013 to 2022 were retrieved and collected after manual review. VOSviewer and other software were used for bibliometric visualization analysis, including publication trends, geographical distribution characteristics, journal distribution characteristics, author contributions, citations, funding source characteristics, and keyword clustering.

Results

A total of 336 documents were analyzed. Since 2018, publications and citations have increased dramatically, with the United States (143) and China (101) as the main contributors. Regarding authors, Bihorac, A and Ozrazgat-Baslanti, T from the Kansas City Medical Center have published 10 articles. Regarding institutions, the University of California (18) had the most publications. Approximately 1/3 of the publications were published in Q1 and Q2 journals, of which Scientific Reports (19) was the most prolific journal. Tomašev et al.'s study that was published in 2019 has been widely cited by researchers. The results of cluster analysis of co-occurrence keywords suggest that the construction of AKI prediction model related to critical patients and sepsis patients is the research frontier, and XGBoost algorithm is also popular.

Conclusion

This study first provides an updated perspective on machine learning-based AKI research, which may be beneficial for subsequent researchers to choose suitable journals and collaborators and may provide a more convenient and in-depth understanding of the research basis, hotspots and frontiers.

Development and validation of clinical prediction models for acute kidney injury recovery at hospital discharge in critically ill adults

PURPOSE

Acute kidney injury (AKI) recovery prediction remains challenging. The purpose of the present study is to develop and validate prediction models for AKI recovery at hospital discharge in critically ill patients with ICU-acquired AKI stage 3 (AKI-3).

METHODS

Models were developed and validated in a development cohort (n = 229) and a matched validation cohort (n = 244) from the multicenter EPaNIC database to create prediction models with the least absolute shrinkage and selection operator (Lasso) machine-learning algorithm. We evaluated the discrimination and calibration of the models and compared their performance with plasma neutrophil gelatinase-associated lipocalin (NGAL) measured on first AKI-3 day (NGAL_AKI3) and reference model that only based on age.

RESULTS

Complete recovery and complete or partial recovery occurred in 33.20% and 51.23% of the validation cohort patients respectively. The prediction model for complete recovery based on age, need for renal replacement therapy (RRT), diagnostic group (cardiac/surgical/trauma/others), and sepsis on admission had an area under the receiver operating characteristics curve (AUROC) of 0.53. The prediction model for complete or partial recovery based on age, need for RRT, platelet count, urea, and white blood cell count had an AUROC of 0.61. NGAL_AKI3 showed AUROCs of 0.55 and 0.53 respectively. In cardiac patients, the models had higher AUROCs of 0.60 and 0.71 than NGAL_AKI3's AUROCs of 0.52 and 0.54. The developed models demonstrated a better performance over the reference models (only based on age) for cardiac surgery patients, but not for patients with sepsis and for a general ICU population.

CONCLUSION

Models to predict AKI recovery upon hospital discharge in critically ill patients with AKI-3 showed poor performance in the general ICU population, similar to the biomarker NGAL. In cardiac surgery patients, discrimination was acceptable, and better than NGAL. These findings demonstrate the difficulty of predicting non-reversible AKI early.

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

RATIONALE & OBJECTIVE

Risk prediction tools for assisting acute kidney injury (AKI) management have focused on AKI onset but have infrequently addressed kidney recovery. We developed clinical models for risk stratification of mortality and major adverse kidney events (MAKE) in critically ill patients with incident AKI.

STUDY DESIGN

Multicenter cohort study.

SETTING & PARTICIPANTS

9,587 adult patients admitted to heterogeneous intensive care units (ICUs; March 2009 to February 2017) who experienced AKI within the first 3 days of their ICU stays.

PREDICTORS

Multimodal clinical data consisting of 71 features collected in the first 3 days of ICU stay.

OUTCOMES

(1) Hospital mortality and (2) MAKE, defined as the composite of death during hospitalization or within 120 days of discharge, receipt of kidney replacement therapy in the last 48 hours of hospital stay, initiation of maintenance kidney replacement therapy within 120 days, or a ≥50% decrease in estimated glomerular filtration rate from baseline to 120 days from hospital discharge.

ANALYTICAL APPROACH

Four machine-learning algorithms (logistic regression, random forest, support vector machine, and extreme gradient boosting) and the SHAP (Shapley Additive Explanations) framework were used for feature selection and interpretation. Model performance was evaluated by 10-fold cross-validation and external validation.

RESULTS

One developed model including 15 features outperformed the SOFA (Sequential Organ Failure Assessment) score for the prediction of hospital mortality, with areas under the curve of 0.79 (95% CI, 0.79-0.80) and 0.71 (95% CI, 0.71-0.71) in the development cohort and 0.74 (95% CI, 0.73-0.74) and 0.71 (95% CI, 0.71-0.71) in the validation cohort (P < 0.001 for both). A second developed model including 14 features outperformed KDIGO (Kidney Disease: Improving Global Outcomes) AKI severity staging for the prediction of MAKE: 0.78 (95% CI, 0.78-0.78) versus 0.66 (95% CI, 0.66-0.66) in the development cohort and 0.73 (95% CI, 0.72-0.74) versus 0.67 (95% CI, 0.67-0.67) in the validation cohort (P < 0.001 for both).

LIMITATIONS

The models are applicable only to critically ill adult patients with incident AKI within the first 3 days of an ICU stay.

CONCLUSIONS

The reported clinical models exhibited better performance for mortality and kidney recovery prediction than standard scoring tools commonly used in critically ill patients with AKI in the ICU. Additional validation is needed to support the utility and implementation of these models.

PLAIN-LANGUAGE SUMMARY

Acute kidney injury (AKI) occurs commonly in critically ill patients admitted to the intensive care unit (ICU) and is associated with high morbidity and mortality rates. Prediction of mortality and recovery after an episode of AKI may assist bedside decision making. In this report, we describe the development and validation of a clinical model using data from the first 3 days of an ICU stay to predict hospital mortality and major adverse kidney events occurring as long as 120 days after hospital discharge among critically ill adult patients who experienced AKI within the first 3 days of an ICU stay. The proposed clinical models exhibited good performance for outcome prediction and, if further validated, could enable risk stratification for timely interventions that promote kidney recovery.

Artificial intelligence for the prediction of acute kidney injury during the perioperative period: systematic review and Meta-analysis of diagnostic test accuracy

BACKGROUND

Acute kidney injury (AKI) is independently associated with morbidity and mortality in a wide range of surgical settings. Nowadays, with the increasing use of electronic health records (EHR), advances in patient information retrieval, and cost reduction in clinical informatics, artificial intelligence is increasingly being used to improve early recognition and management for perioperative AKI. However, there is no quantitative synthesis of the performance of these methods. We conducted this systematic review and meta-analysis to estimate the sensitivity and specificity of artificial intelligence for the prediction of acute kidney injury during the perioperative period.

METHODS

Pubmed, Embase, and Cochrane Library were searched to 2nd October 2021. Studies presenting diagnostic performance of artificial intelligence in the early detection of perioperative acute kidney injury were included. True positives, false positives, true negatives and false negatives were pooled to collate specificity and sensitivity with 95% CIs and results were portrayed in forest plots. The risk of bias of eligible studies was assessed using the PROBAST tool.

RESULTS

Nineteen studies involving 304,076 patients were included. Quantitative random-effects meta-analysis using the Rutter and Gatsonis hierarchical summary receiver operating characteristics (HSROC) model revealed pooled sensitivity, specificity, and diagnostic odds ratio of 0.77 (95% CI: 0.73 to 0.81),0.75 (95% CI: 0.71 to 0.80), and 10.7 (95% CI 8.5 to 13.5), respectively. Threshold effect was found to be the only source of heterogeneity, and there was no evidence of publication bias.

CONCLUSIONS

Our review demonstrates the promising performance of artificial intelligence for early prediction of perioperative AKI. The limitations of lacking external validation performance and being conducted only at a single center should be overcome.

TRIAL REGISTRATION

This study was not registered with PROSPERO.

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.

External validation of a deep-learning model to predict severe acute kidney injury based on urine output changes in critically ill patients

OBJECTIVES

The purpose of this study was to externally validate algorithms (previously developed and trained in two United States populations) aimed at early detection of severe oliguric AKI (stage 2/3 KDIGO) in intensive care units patients.

METHODS

The independent cohort was composed of 10'596 patients from the university hospital ICU of Amsterdam (the "AmsterdamUMC database") admitted to their intensive care units. In this cohort, we analysed the accuracy of algorithms based on logistic regression and deep learning methods. The accuracy of investigated algorithms had previously been tested with electronic intensive care unit (eICU) and MIMIC-III patients.

RESULTS

The deep learning model had an area under the ROC curve (AUC) of 0,907 (± 0,007SE) with a sensitivity and specificity of 80% and 89%, respectively, for identifying oliguric AKI episodes. Logistic regression models had an AUC of 0,877 (± 0,005SE) with a sensitivity and specificity of 80% and 81%, respectively. These results were comparable to those obtained in the two US populations upon which the algorithms were previously developed and trained.

CONCLUSION

External validation on the European sample confirmed the accuracy of the algorithms, previously investigated in the US population. The models show high accuracy in both the European and the American databases even though the two cohorts differ in a range of demographic and clinical characteristics, further underlining the validity and the generalizability of the two analytical approaches.

Nomogram Prediction Model of Serum Chloride and Sodium Ions on the Risk of Acute Kidney Injury in Critically Ill Patients

Purpose

This study aims to investigate the effect of serum chloride and sodium ions on AKI occurrence in ICU patients, and further constructs a prediction model containing these factors to explore the predictive value of these ions in AKI.

Methods

The clinical information of patients admitted to ICU of Beijing Friendship Hospital Affiliated to Capital Medical University was collected for retrospective analysis. Logistic regression analysis was used to analyzing the influencing factors. A nomogram for predicting AKI risk was constructed with R software and validated by repeated sampling. Afterwards, the effectiveness and accuracy of the model were tested and evaluated.

Results

A total of 446 cases met the requirements of this study, of which 178 developed AKI during their stay in ICU, with an incidence rate of 39.9%. Hypernatremia, heart failure, sepsis, APACHE II score, and initial creatinine value and BE value at ICU admission before the diagnosis of AKI were identified as independent risk factors for developing AKI during ICU stay. These predictors were incorporated into the nomogram of AKI risk in critically ill patients, which was constructed by using R software. Receiver operating characteristic curve analysis was further used and showed that the area under the curve of the model was 0.7934 (95% CI 0.742–0.8447), indicating that the model had an ideal value. Finally, further evaluated its clinical effectiveness. The clinical effect curve and decision curve showed that most areas of the decision curve of this model were greater than 0, indicating that this model owned a certain clinical effectiveness.

Conclusion

The nomogram based on hypernatremia, heart failure, sepsis, APACHE II score, and initial creatinine and BE value in ICU can predict the individualized risk of AKI with satisfactory distinguishability and accuracy.

Development and validation of the MMCD score to predict kidney replacement therapy in COVID-19 patients

BACKGROUND

Acute kidney injury (AKI) is frequently associated with COVID-19, and the need for kidney replacement therapy (KRT) is considered an indicator of disease severity. This study aimed to develop a prognostic score for predicting the need for KRT in hospitalised COVID-19 patients, and to assess the incidence of AKI and KRT requirement.

METHODS

This study is part of a multicentre cohort, the Brazilian COVID-19 Registry. A total of 5212 adult COVID-19 patients were included between March/2020 and September/2020. Variable selection was performed using generalised additive models (GAM), and least absolute shrinkage and selection operator (LASSO) regression was used for score derivation. Accuracy was assessed using the area under the receiver operating characteristic curve (AUC-ROC).

RESULTS

The median age of the model-derivation cohort was 59 (IQR 47-70) years, 54.5% were men, 34.3% required ICU admission, 20.9% evolved with AKI, 9.3% required KRT, and 15.1% died during hospitalisation. The temporal validation cohort had similar age, sex, ICU admission, AKI, required KRT distribution and in-hospital mortality. The geographic validation cohort had similar age and sex; however, this cohort had higher rates of ICU admission, AKI, need for KRT and in-hospital mortality. Four predictors of the need for KRT were identified using GAM: need for mechanical ventilation, male sex, higher creatinine at hospital presentation and diabetes. The MMCD score had excellent discrimination in derivation (AUROC 0.929, 95% CI 0.918-0.939) and validation (temporal AUROC 0.927, 95% CI 0.911-0.941; geographic AUROC 0.819, 95% CI 0.792-0.845) cohorts and good overall performance (Brier score: 0.057, 0.056 and 0.122, respectively). The score is implemented in a freely available online risk calculator ( https://www.mmcdscore.com/ ).

CONCLUSIONS

The use of the MMCD score to predict the need for KRT may assist healthcare workers in identifying hospitalised COVID-19 patients who may require more intensive monitoring, and can be useful for resource allocation.

Artificial Intelligence in Acute Kidney Injury Prediction

The use of artificial intelligence (AI) in nephrology and its associated clinical research is growing. Recent years have seen increased interest in utilizing AI to predict the development of hospital-based acute kidney injury (AKI). Several AI techniques have been employed to improve the ability to detect AKI across a variety of hospitalized settings. This review discusses the evolutions of AKI risk prediction discussing the static risk assessment models of yesteryear as well as the more recent trend toward AI and advanced learning techniques. We discuss the relative improvement in AKI detection as well as the relative dearth of data around the clinical implementation and patient outcomes using these models. The use of AI for AKI detection and clinical care is in its infancy, and this review describes how we arrived at our current position and hints at the promise of the future.

Quantitative EEG parameters can improve the predictive value of the non-traumatic neurological ICU patient prognosis through the machine learning method

Background

Better outcome prediction could assist in reliable classification of the illnesses in neurological intensive care unit (ICU) severity to support clinical decision-making. We developed a multifactorial model including quantitative electroencephalography (QEEG) parameters for outcome prediction of patients in neurological ICU.

Methods

We retrospectively analyzed neurological ICU patients from November 2018 to November 2021. We used 3-month mortality as the outcome. Prediction models were created using a linear discriminant analysis (LDA) based on QEEG parameters, APACHEII score, and clinically relevant features. Additionally, we compared our best models with APACHEII score and Glasgow Coma Scale (GCS). The DeLong test was carried out to compare the ROC curves in different models.

Results

A total of 110 patients were included and divided into a training set (n=80) and a validation set (n = 30). The best performing model had an AUC of 0.85 in the training set and an AUC of 0.82 in the validation set, which were better than that of GCS (training set 0.64, validation set 0.61). Models in which we selected only the 4 best QEEG parameters had an AUC of 0.77 in the training set and an AUC of 0.71 in the validation set, which were similar to that of APACHEII (training set 0.75, validation set 0.73). The models also identified the relative importance of each feature.

Conclusion

Multifactorial machine learning models using QEEG parameters, clinical data, and APACHEII score have a better potential to predict 3-month mortality in non-traumatic patients in neurological ICU.

Prediction of acute kidney injury in patients with femoral neck fracture utilizing machine learning

Background

Acute kidney injury (AKI) is a common complication associated with significant morbidity and mortality in high-energy trauma patients. Given the poor efficacy of interventions after AKI development, it is important to predict AKI before its diagnosis. Therefore, this study aimed to develop models using machine learning algorithms to predict the risk of AKI in patients with femoral neck fractures.

Methods

We developed machine-learning models using the Medical Information Mart from Intensive Care (MIMIC)-IV database. AKI was predicted using 10 predictive models in three-time windows, 24, 48, and 72 h. Three optimal models were selected according to the accuracy and area under the receiver operating characteristic curve (AUROC), and the hyperparameters were adjusted using a random search algorithm. The Shapley additive explanation (SHAP) analysis was used to determine the impact and importance of each feature on the prediction. Compact models were developed using important features chosen based on their SHAP values and clinical availability. Finally, we evaluated the models using metrics such as accuracy, precision, AUROC, recall, F1 scores, and kappa values on the test set after hyperparameter tuning.

Results

A total of 1,596 patients in MIMIC-IV were included in the final cohort, and 402 (25%) patients developed AKI after surgery. The light gradient boosting machine (LightGBM) model showed the best overall performance for predicting AKI before 24, 48, and 72 h. AUROCs were 0.929, 0.862, and 0.904. The SHAP value was used to interpret the prediction models. Renal function markers and perioperative blood transfusions are the most critical features for predicting AKI. In compact models, LightGBM still performs the best. AUROCs were 0.930, 0.859, and 0.901.

Conclusions

In our analysis, we discovered that LightGBM had the best metrics among all algorithms used. Our study identified the LightGBM as a solid first-choice algorithm for early AKI prediction in patients after femoral neck fracture surgery.

Internal and external validation of machine learning-assisted prediction models for mechanical ventilation-associated severe acute kidney injury

BACKGROUND

Currently, very few preventive or therapeutic strategies are used for mechanical ventilation (MV)-associated severe acute kidney injury (AKI).

OBJECTIVES

We developed clinical prediction models to detect the onset of severe AKI in the first week of intensive care unit (ICU) stay during the initiation of MV.

METHODS

A large ICU database Medical Information Mart for Intensive Care IV (MIMIC-IV) was analysed retrospectively. Data were collected from the clinical information recorded at the time of ICU admission and during the initial 12 h of MV. Using univariate and multivariate analyses, the predictors were selected successively. For model development, two machine learning algorithms were compared. The primary goal was to predict the development of AKI stage 2 or 3 (AKI-23) and AKI stage 3 (AKI-3) in the first week of patients' ICU stay after initial 12 h of MV. The developed models were externally validated using another multicentre ICU database (eICU Collaborative Research Database, eICU) and evaluated in various patient subpopulations.

RESULTS

Models were developed using data from the development cohort (MIMIC-IV: 2008-2016; n = 3986); the random forest algorithm outperformed the logistic regression algorithm. In the internal (MIMIC-IV: 2017-2019; n = 1210) and external (eICU; n = 1494) validation cohorts, the incidences of AKI-23 were 154 (12.7%) and 119 (8.0%), respectively, with areas under the receiver operator characteristic curve of 0.78 (95% confidence interval [CI]: 0.74-0.82) and 0.80 (95% CI: 0.76-0.84); the incidences of AKI-3 were 81 (6.7%) and 67 (4.5%), with areas under the receiver operator characteristic curve of 0.81 (95% CI: 0.76-0.87) and 0.80 (95% CI: 0.73-0.86), respectively.

CONCLUSIONS

Models driven by machine learning and based on routine clinical data may facilitate the early prediction of MV-associated severe AKI. The validated models can be found at: https://apoet.shinyapps.io/mv_aki_2021_v2/.

Small, short-term, point-of-care creatinine changes as predictors of acute kidney injury in critically ill patients

PURPOSE

To assess short-term creatinine changes as predictors of acute kidney injury (AKI) when used alone and in combination with AKI risk factors.

METHODS

In this prospective cohort study, we identified all creatinine measurements from frequent point-of-care arterial blood gas measurements from ICU admission until AKI. We evaluated the predictive value of small changes between these creatinine measurements for AKI development, alone and with AKI risk factors.

RESULTS

Of 377 patients with 3235 creatinine measurements, generating 15,075 creatinine change episodes, 215 (57%) patients developed AKI, and 68 (18%) developed stage 2 or 3 AKI. In isolation, a creatinine increase over 4.1-7.3 h had a 0.65 area under the curve for predicting stage 2 or 3 AKI within 3-37.7 h. Combining creatinine increases of ≥1 μmol/L/h (≥0.0113 mg/dL/h) over 4-5.8 h with three AKI risk factors (cardiac surgery, use of vasopressors, chronic liver disease) had 83% sensitivity, 79% specificity and 0.87 area under the curve for stage 2 or 3 AKI occurring 8.7-25.6 h later.

CONCLUSION

In combination with key risk factors, frequent point-of-care creatinine assessment on arterial blood gases to detect small, short-term creatinine changes provides a robust, novel, low-cost, and rapid method for predicting AKI in critically ill patients.

Predicting acute kidney injury following open partial nephrectomy treatment using SAT-pruned explainable machine learning model

Background

One of the most prevalent complications of Partial Nephrectomy (PN) is Acute Kidney Injury (AKI), which could have a negative impact on subsequent renal function and occurs in up to 24.3% of patients undergoing PN. The aim of this study was to predict the occurrence of AKI following PN using preoperative parameters by applying machine learning algorithms.

Methods

We included all adult patients (n = 723) who underwent open PN in our department since 1995 and on whom we have data on the pre-operative renal function. We developed a random forest (RF) model with Boolean satisfaction-based pruned decision trees for binary classification (AKI or non-AKI). Hyper-parameter grid search was performed to optimize the model's performance. Fivefold cross-validation was applied to evaluate the model. We implemented a RF model with greedy feature selection to binary classify AKI and non-AKI cases based on pre-operative data.

Results

The best model obtained a 0.69 precision and 0.69 recall in classifying the AKI and non-AKI groups on average (k = 5). In addition, the model's probability to correctly classify a new prediction is 0.75. The proposed model is available as an online calculator.

Conclusions

Our model predicts the occurrence of AKI following open PN with (75%) accuracy. We plan to externally validate this model and modify it to minimally-invasive PN.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12911-022-01877-8.

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.

Risk Classification and Subphenotyping of Acute Kidney Injury: Concepts and Methodologies

Acute kidney injury (AKI) is a complex syndrome with a paucity of therapeutic development. One aspect that could explain the lack of implementation science in the AKI field is the vast heterogeneity of the AKI syndrome, which hinders precise therapeutic applications for specific AKI subpopulations. In this context, there is a consensual focus of the scientific community toward the development and validation of tools to better subphenotype AKI and therefore facilitate precision medicine approaches. The subphenotyping of AKI requires the use of specific methodologies suitable for interrogation of multimodal data inputs from different sources such as electronic health records, organ support devices, and/or biospecimens and tissues. Over the past years, the surge of artificial intelligence applied to health care has yielded novel machine learning methodologies for data acquisition, harmonization, and interrogation that can assist with subphenotyping of AKI. However, one should recognize that although risk classification and subphenotyping of AKI is critically important, testing their potential applications is even more important to promote implementation science. For example, risk-classification should support actionable interventions that could ameliorate or prevent the occurrence of the outcome being predicted. Furthermore, subphenotyping could be applied to predict therapeutic responses to support enrichment and adaptive platforms for pragmatic clinical trials.

A Novel Risk Prediction Model for Severe Acute Kidney Injury in Intensive Care Unit Patients Receiving Fluid Resuscitation

Background

To develop a risk prediction model for the occurrence of severe acute kidney injury (AKI) in intensive care unit (ICU) patients receiving fluid resuscitation.

Methods

We conducted a secondary analysis of the Crystalloid vs. Hydroxyethyl Starch Trial (CHEST) trial, a blinded randomized controlled trial that enrolled ICU patients who received intravenous fluid resuscitation. The primary outcome was the first event in a composite outcome of doubling of serum creatinine and/or treatment with renal replacement treatment (RRT) within 28 days of randomization. The final model developed using multivariable logistic regression with backwards elimination was validated internally and then translated into a predictive equation.

Results

Six thousand seven hundred twenty-seven ICU participants were studied, among whom 745 developed the study outcome. The final model having six variables, including admission diagnosis of sepsis, illness severity score, mechanical ventilation, tachycardia, baseline estimated glomerular filtration rate and emergency admission. The model had good discrimination (c-statistic = 0.72, 95% confidence interval 0.697–0.736) and calibration (Hosmer-Lemeshow test, χ² = 14.4, p = 0.07) for the composite outcome, with a c-statistic after internal bootstrapping validation of 0.72, which revealed a low degree of over-fitting. The positive predictive value and negative predictive value were 58.8 and 89.1%, respectively. The decision curve analysis indicates a net benefit in prediction of severe AKI using the model across a range of threshold probabilities between 5 and 35%.

Conclusions

Our model, using readily available clinical variables to identify ICU patients at high risk of severe AKI achieved good predictive performance in a clinically relevant population.

Serum IL-17 levels are higher in critically ill patients with AKI and associated with worse outcomes

Background

Interleukin-17 (IL-17) antagonism in rats reduces the severity and progression of AKI. IL-17-producing circulating T helper-17 (TH17) cells is increased in critically ill patients with AKI indicating that this pathway is also activated in humans. We aim to compare serum IL-17A levels in critically ill patients with versus without AKI and to examine their relationship with mortality and major adverse kidney events (MAKE).

Methods

Multicenter, prospective study of ICU patients with AKI stage 2 or 3 and without AKI. Samples were collected at 24–48 h after AKI diagnosis or ICU admission (in those without AKI) [timepoint 1, T1] and 5–7 days later [timepoint 2, T2]. MAKE was defined as the composite of death, dependence on kidney replacement therapy or a reduction in eGFR of ≥ 30% from baseline up to 90 days following hospital discharge.

Results

A total of 299 patients were evaluated. Patients in the highest IL-17A tertile (versus lower tertiles) at T1 had higher acuity of illness and comorbidity scores. Patients with AKI had higher levels of IL-17A than those without AKI: T1 1918.6 fg/ml (692.0–5860.9) versus 623.1 fg/ml (331.7–1503.4), p < 0.001; T2 2167.7 fg/ml (839.9–4618.9) versus 1193.5 fg/ml (523.8–2198.7), p = 0.006. Every onefold higher serum IL-17A at T1 was independently associated with increased risk of hospital mortality (aOR 1.35, 95% CI: 1.06–1.73) and MAKE (aOR 1.26, 95% CI: 1.02–1.55). The highest tertile of IL-17A (vs. the lowest tertile) was also independently associated with higher risk of MAKE (aOR 3.03, 95% CI: 1.34–6.87). There was no effect modification of these associations by AKI status. IL-17A levels remained significantly elevated at T2 in patients that died or developed MAKE.

Conclusions

Serum IL-17A levels measured by the time of AKI diagnosis or ICU admission were differentially elevated in critically ill patients with AKI when compared to those without AKI and were independently associated with hospital mortality and MAKE.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-022-03976-4.

Development of a knowledge mining approach to uncover heterogeneous risk predictors of acute kidney injury across age groups

OBJECTIVES

Acute kidney injury (AKI) risk increases with age and the underlying clinical predictors may be heterogeneous across age strata. This study aims to uncover the AKI risk factor heterogeneity among general inpatients across age groups using electronic medical records (EMR).

METHODS

Patient data (n = 179,370 encounters) were collected from an academic hospital between 2007 and 2016, and were stratified into four age groups: 18-35, 36-55, 56-65, and > 65. Potential risk factors extracted for the cohort included demographics, vital signs, laboratory values, past medical diagnoses, medications and admission diagnoses. We developed a data driven knowledge mining approach consisting of a machine learning algorithm to identify AKI predictors across age strata and a statistical method to quantify the impact of those factors on AKI risk. Identified predictors were evaluated for their predictability of AKI in terms of area-under-the-receiver-operating-characteristic-curve (AUC) and validated against expert knowledge.

RESULTS

Among the final analysis cohort of 76,957 hospital admissions, AKI prediction across age groups 18-35 (16.73%), 36-55 (32.74%), 56-65 (23.52%), and > 65 years (27.01%) achieved AUC of 0.85 (95% CI, 0.80-0.88), 0.86 (95% CI, 0.83-0.89), 0.87 (95% CI, 0.86-0.90), and 0.87 (95% CI, 0.86-0.90), respectively. Compared to expert knowledge, absolute consistency rates of the top-150 identified risk factors for each group were 78.4%, 77.2%, 81.3%, and 79.5%, respectively. Impact of many predictors on AKI varied across age groups; for example, high body mass index (BMI) was found to be associated with higher AKI risk in elderly patients, but low BMI was found to be associated with higher AKI risk in younger patients.

CONCLUSIONS

We verified the effectiveness of the knowledge mining method from the perspectives of accuracy, stability and credibility, and used this approach to clarify the heterogeneity of AKI risk factors between age groups. Future decision support systems need to consider such heterogeneity to enhance personalized patient care.

A deep-learning model to continuously predict severe acute kidney injury based on urine output changes in critically ill patients

BACKGROUND

Acute Kidney Injury (AKI), a frequent complication of pateints in the Intensive Care Unit (ICU), is associated with a high mortality rate. Early prediction of AKI is essential in order to trigger the use of preventive care actions.

METHODS

The aim of this study was to ascertain the accuracy of two mathematical analysis models in obtaining a predictive score for AKI development. A deep learning model based on a urine output trends was compared with a logistic regression analysis for AKI prediction in stages 2 and 3 (defined as the simultaneous increase of serum creatinine and decrease of urine output, according to  the Acute Kidney Injury Network (AKIN) guidelines). Two retrospective datasets including 35,573 ICU patients were analyzed. Urine output data were used to train and test the logistic regression and the deep learning model.

RESULTS

The deep learning model defined an area under the curve (AUC) of 0.89 (± 0.01), sensitivity = 0.8 and specificity = 0.84, which was higher than the logistic regression analysis. The deep learning model was able to predict 88% of AKI cases more than 12 h before their onset: for every 6 patients identified as being at risk of AKI by the deep learning model, 5 experienced the event. On the contrary, for every 12 patients not considered to be at risk by the model, 2 developed AKI.

CONCLUSION

In conclusion, by using urine output trends, deep learning analysis was able to predict AKI episodes more than 12 h in advance, and with a higher accuracy than the classical urine output thresholds. We suggest that this algorithm could be integrated in the ICU setting to better manage, and potentially prevent, AKI episodes.

Prediction of cardiac surgery associated - acute kidney injury (CSA-AKI) by healthcare professionals and urine cell cycle arrest AKI biomarkers [TIMP-2]*[IGFBP7]: A single center prospective study (the PREDICTAKI trial)

PURPOSE

Cardiac surgery associated acute kidney injury (CSA-AKI) is a contributor to adverse outcomes. Preventive measures reduce AKI incidence in high risk patients, identified by biomarkers [TIMP-2]*[IGFBP7] (Nephrocheck®). This study investigate clinical AKI risk assessment by healthcare professionals and the added value of the biomarker result.

MATERIALS AND METHODS

Adult patients were prospectively included. Healthcare professionals predicted CSA-AKI, with and without biomarker result knowledge. Predicted outcomes were AKI based on creatinine, AKI stage 3 on urine output, anuria and use of kidney replacement therapy (KRT).

RESULTS

One-hundred patients were included. Consultant and ICU residents were best in AKI prediction, respectively AUROC 0.769 (95% CI, 0.672-0.850) and 0.702 (95% CI, 0.599-0.791). AUROC of NephroCheck® was 0.541 (95% CI, 0.438-0.642). AKI 3 occurred in only 4 patients; there was no anuria or use of KRT. ICU nurses and ICU residents had an AUROC for prediction of AKI 3 of respectively 0.867 (95% CI, 0.780-0.929) and 0.809 (95% CI, 0.716-0.883); for NephroCheck® this was 0.838 (95% CI, 0.750-0.904).

CONCLUSIONS

Healthcare professionals performed poor or fair in predicting CSA-AKI and knowledge of Nephrocheck® result did not improved prediction. No conclusions could be made for prediction of severe AKI, due to limited number of events.

A Machine Learning Model for Accurate Prediction of Sepsis in ICU Patients

Background: Although numerous studies are conducted every year on how to reduce the fatality rate associated with sepsis, it is still a major challenge faced by patients, clinicians, and medical systems worldwide. Early identification and prediction of patients at risk of sepsis and adverse outcomes associated with sepsis are critical. We aimed to develop an artificial intelligence algorithm that can predict sepsis early.

Methods: This was a secondary analysis of an observational cohort study from the Intensive Care Unit of the First Affiliated Hospital of Zhengzhou University. A total of 4,449 infected patients were randomly assigned to the development and validation data set at a ratio of 4:1. After extracting electronic medical record data, a set of 55 features (variables) was calculated and passed to the random forest algorithm to predict the onset of sepsis.

Results: The pre-procedure clinical variables were used to build a prediction model from the training data set using the random forest machine learning method; a 5-fold cross-validation was used to evaluate the prediction accuracy of the model. Finally, we tested the model using the validation data set. The area obtained by the model under the receiver operating characteristic (ROC) curve (AUC) was 0.91, the sensitivity was 87%, and the specificity was 89%.

Conclusions: This newly established machine learning-based model has shown good predictive ability in Chinese sepsis patients. External validation studies are necessary to confirm the universality of our method in the population and treatment practice.

Timing of Kidney Support Therapy in Acute Kidney Injury: What Are We Waiting For?

The optimal timing of kidney support therapy in critically ill patients with acute kidney injury (AKI) without life-threatening complications related to AKI is controversial. Recent multicenter, randomized, controlled studies have questioned the need for earlier initiation of therapy, despite one study showing a benefit in survival and others with no differences in mortality based on the timing of kidney support therapy initiation. These findings reflect the uncertainties in decisions to initiate kidney support therapy, which should ideally be individualized according to the patient's comorbidities, severity of illness, trajectory of kidney function, and urine output as well as requirements for fluid balance and solute removal. A delayed approach could translate into a potentially reduced burden of dialysis dependence in addition to saving health resources. However, we must ascertain what constitutes the waiting period and the benefits and risks associated with this approach. This article reviews the concept of timing of dialysis in AKI, performs a critical assessment of the most important clinical trials in this topic, discusses ongoing research and knowledge gaps, and defines key research issues to address in the future.

Developing an Agnostic Risk Prediction Model for Early AKI Detection in Cancer Patients

Acute kidney injury (AKI) is a common complication among oncology patients associated with lower remission rates and higher mortality. To reduce the impact of this condition, we aimed to predict AKI earlier than existing tools, to allow clinical intervention before occurrence. We trained a random forest model on 597,403 routinely collected blood test results from 48,865 patients undergoing cancer treatment at The Christie NHS Foundation Trust between January 2017 and May 2020, to identify AKI events upcoming in the next 30 days. AKI risk levels were assigned to upcoming AKI events and tested through a prospective analysis between June and August 2020. The trained model gave an AUROC of 0.881 (95% CI 0.878-0.883), when assessing predictions per blood test for AKI occurrences within 30 days. Assigning risk levels and testing the model through prospective validation from the 1st June to the 31st August identified 73.8% of patients with an AKI event before at least one AKI occurrence, 61.2% of AKI occurrences. Our results suggest that around 60% of AKI occurrences experienced by patients undergoing cancer treatment could be identified using routinely collected blood results, allowing clinical remedial action to be taken and disruption to treatment by AKI to be minimised.

Machine learning model for early prediction of acute kidney injury (AKI) in pediatric critical care

BACKGROUND

Acute kidney injury (AKI) in pediatric critical care patients is diagnosed using elevated serum creatinine, which occurs only after kidney impairment. There are no treatments other than supportive care for AKI once it has developed, so it is important to identify patients at risk to prevent injury. This study develops a machine learning model to learn pre-disease patterns of physiological measurements and predict pediatric AKI up to 48 h earlier than the currently established diagnostic guidelines.

METHODS

EHR data from 16,863 pediatric critical care patients between 1 month to 21 years of age from three independent institutions were used to develop a single machine learning model for early prediction of creatinine-based AKI using intelligently engineered predictors, such as creatinine rate of change, to automatically assess real-time AKI risk. The primary outcome is prediction of moderate to severe AKI (Stage 2/3), and secondary outcomes are prediction of any AKI (Stage 1/2/3) and requirement of renal replacement therapy (RRT). Predictions generate alerts allowing fast assessment and reduction of AKI risk, such as: "patient has 90% risk of developing AKI in the next 48 h" along with contextual information and suggested response such as "patient on aminoglycosides, suggest check level and review dose and indication".

RESULTS

The model was successful in predicting Stage 2/3 AKI prior to detection by conventional criteria with a median lead-time of 30 h at AUROC of 0.89. The model predicted 70% of subsequent RRT episodes, 58% of Stage 2/3 episodes, and 41% of any AKI episodes. The ratio of false to true alerts of any AKI episodes was approximately one-to-one (PPV 47%). Among patients predicted, 79% received potentially nephrotoxic medication after being identified by the model but before development of AKI.

CONCLUSIONS

As the first multi-center validated AKI prediction model for all pediatric critical care patients, the machine learning model described in this study accurately predicts moderate to severe AKI up to 48 h in advance of AKI onset. The model may improve outcome of pediatric AKI by providing early alerting and actionable feedback, potentially preventing or reducing AKI by implementing early measures such as medication adjustment.

Inverse Correlation Between Incidence and Mortality of Acute Kidney Injury in Critically Ill Patients: A Systematic Review

BACKGROUND

The reported incidence and mortality of acute kidney injury (AKI) in patients in intensive care units (ICUs) is remarkably different even with standardized AKI criteria. The aim of this study was to investigate the correlation between the incidence and mortality of patients with AKI in ICUs.

METHODS

We systematically reviewed clinical studies regarding adult ICU patients with AKI using Kidney Disease: Improving Global Outcomes-equivalent criteria from 2004 to May 1, 2018. We searched MEDLINE, EMBASE, and Cochrane Library to investigate the correlation between the incidence and mortality of patients with AKI in each cohort. Studies with small number of participants (less than 500) were excluded. The correlation between the incidence of AKI and mortality of patients was evaluated using a regression model.

RESULTS

Our review yielded 76 cohorts, comprising 564,455 patients in ICU (median age, 60.5 years; men, 59.5%). The mortality of all patients did not correlate with the incidence of AKI in each cohort; however, the mortality of patients with AKI significantly decreased [squared correlation coefficient (R) = 0.18, regression coefficient (β) = -0.25, P < 0.001] as the incidence of AKI increased. This correlation was also observed in a subgroup analysis limited to the clinical setting of general ICUs, and among patients with mild or severe AKI.

CONCLUSIONS

An inverse correlation between the incidence of AKI and the mortality of patients with AKI may indicate an advantage of frequent AKI occurrence, possibly because of increased awareness and larger exposure to AKIs; further study is needed, however, to confirm the causality.

TRIAL REGISTRATION

The protocol was registered in PROSPERO database (CRD 42019129322).

Acute kidney injury in the critically ill: an updated review on pathophysiology and management

Acute kidney injury (AKI) is now recognized as a heterogeneous syndrome that not only affects acute morbidity and mortality, but also a patient’s long-term prognosis. In this narrative review, an update on various aspects of AKI in critically ill patients will be provided. Focus will be on prediction and early detection of AKI (e.g., the role of biomarkers to identify high-risk patients and the use of machine learning to predict AKI), aspects of pathophysiology and progress in the recognition of different phenotypes of AKI, as well as an update on nephrotoxicity and organ cross-talk. In addition, prevention of AKI (focusing on fluid management, kidney perfusion pressure, and the choice of vasopressor) and supportive treatment of AKI is discussed. Finally, post-AKI risk of long-term sequelae including incident or progression of chronic kidney disease, cardiovascular events and mortality, will be addressed.

A validation study comparing existing prediction models of acute kidney injury in patients with acute heart failure

Acute kidney injury (AKI) is a common complication in acute heart failure (AHF) and is associated with prolonged hospitalization and increased mortality. The aim of this study was to externally validate existing prediction models of AKI in patients with AHF. Data for 10,364 patients hospitalized for acute heart failure between 2008 and 2018 were extracted from the Chang Gung Research Database and analysed. The primary outcome of interest was AKI, defined according to the KDIGO definition. The area under the receiver operating characteristic (AUC) curve was used to assess the discrimination performance of each prediction model. Five existing prediction models were externally validated, and the Forman risk score and the prediction model reported by Wang et al. showed the most favourable discrimination and calibration performance. The Forman risk score had AUCs for discriminating AKI, AKI stage 3, and dialysis within 7 days of 0.696, 0.829, and 0.817, respectively. The Wang et al. model had AUCs for discriminating AKI, AKI stage 3, and dialysis within 7 days of 0.73, 0.858, and 0.845, respectively. The Forman risk score and the Wang et al. prediction model are simple and accurate tools for predicting AKI in patients with AHF.

The importance of the urinary output criterion for the detection and prognostic meaning of AKI

Most reports on AKI claim to use KDIGO guidelines but fail to include the urinary output (UO) criterion in their definition of AKI. We postulated that ignoring UO alters the incidence of AKI, may delay diagnosis of AKI, and leads to underestimation of the association between AKI and ICU mortality. Using routinely collected data of adult patients admitted to an intensive care unit (ICU), we retrospectively classified patients according to whether and when they would be diagnosed with KDIGO AKI stage ≥ 2 based on baseline serum creatinine (Screa) and/or urinary output (UO) criterion. As outcomes, we assessed incidence of AKI and association with ICU mortality. In 13,403 ICU admissions (62.2% male, 60.8 ± 16.8 years, SOFA 7.0 ± 4.1), incidence of KDIGO AKI stage ≥ 2 was 13.2% when based only the SCrea criterion, 34.3% when based only the UO criterion, and 38.7% when based on both criteria. By ignoring the UO criterion, 66% of AKI cases were missed and 13% had a delayed diagnosis. The cause-specific hazard ratios of ICU mortality associated with KDIGO AKI stage ≥ 2 diagnosis based on only the SCrea criterion, only the UO criterion and based on both criteria were 2.11 (95% CI 1.85–2.42), 3.21 (2.79–3.69) and 2.85 (95% CI 2.43–3.34), respectively. Ignoring UO in the diagnosis of KDIGO AKI stage ≥ 2 decreases sensitivity, may lead to delayed diagnosis and results in underestimation of KDIGO AKI stage ≥ 2 associated mortality.

Comparison of machine learning and logistic regression models in predicting acute kidney injury: A systematic review and meta-analysis

INTRODUCTION

We aimed to assess whether machine learning models are superior at predicting acute kidney injury (AKI) compared to logistic regression (LR), a conventional prediction model.

METHODS

Eligible studies were identified using PubMed and Embase. A total of 24 studies consisting of 84 prediction models met inclusion criteria. Independent samples t-test was performed to detect mean differences in area under the curve (AUC) between ML and LR models. One-way ANOVA and post-hoc t-tests were performed to assess mean differences in AUC between ML methods.

RESULTS

AUC data were similar between ML (0.736 ± 0.116) and LR (0.748 ± 0.057) models (p = 0.538). However, specific ML models, such as gradient boosting (0.838 ± 0.077), exhibited superior performance at predicting AKI as compared to other ML models in the literature (p < 0.05). Creatinine and urine output, standard variables assessed for AKI staging, were classified as significant predictors across multiple ML models, although the majority of significant predictors were unique and study specific.

CONCLUSIONS

These data suggest that ML models perform equally to that of LR, however ML models exhibit variable performance with some ML models displaying exceptional performance. The variability in ML prediction of AKI can be attributed, in part, to the specific ML model utilized, variable selection and processing, study and subject characteristics, and the steps associated with model training, validation, testing, and calibration.

How will artificial intelligence and bioinformatics change our understanding of IgA Nephropathy in the next decade?

IgA nephropathy (IgAN) is the most common glomerulonephritis. It is characterized by the deposition of immune complexes containing immunoglobulin A (IgA) in the kidney’s glomeruli, triggering an inflammatory process. In many patients, the disease has a progressive course, eventually leading to end-stage kidney disease. The current understanding of IgAN’s pathophysiology is incomplete, with the involvement of several potential players, including the mucosal immune system, the complement system, and the microbiome. Dissecting this complex pathophysiology requires an integrated analysis across molecular, cellular, and organ scales. Such data can be obtained by employing emerging technologies, including single-cell sequencing, next-generation sequencing, proteomics, and complex imaging approaches. These techniques generate complex “big data,” requiring advanced computational methods for their analyses and interpretation. Here, we introduce such methods, focusing on the broad areas of bioinformatics and artificial intelligence and discuss how they can advance our understanding of IgAN and ultimately improve patient care. The close integration of advanced experimental and computational technologies with medical and clinical expertise is essential to improve our understanding of human diseases. We argue that IgAN is a paradigmatic disease to demonstrate the value of such a multidisciplinary approach.

A customizable deep learning model for nosocomial risk prediction from critical care notes with indirect supervision

OBJECTIVE

Reliable longitudinal risk prediction for hospitalized patients is needed to provide quality care. Our goal is to develop a generalizable model capable of leveraging clinical notes to predict healthcare-associated diseases 24-96 hours in advance.

METHODS

We developed a reCurrent Additive Network for Temporal RIsk Prediction (CANTRIP) to predict the risk of hospital acquired (occurring ≥ 48 hours after admission) acute kidney injury, pressure injury, or anemia ≥ 24 hours before it is implicated by the patient's chart, labs, or notes. We rely on the MIMIC III critical care database and extract distinct positive and negative cohorts for each disease. We retrospectively determine the date-of-event using structured and unstructured criteria and use it as a form of indirect supervision to train and evaluate CANTRIP to predict disease risk using clinical notes.

RESULTS

Our experiments indicate that CANTRIP, operating on text alone, obtains 74%-87% area under the curve and 77%-85% Specificity. Baseline shallow models showed lower performance on all metrics, while bidirectional long short-term memory obtained the highest Sensitivity at the cost of significantly lower Specificity and Precision.

DISCUSSION

Proper model architecture allows clinical text to be successfully harnessed to predict nosocomial disease, outperforming shallow models and obtaining similar performance to disease-specific models reported in the literature.

CONCLUSION

Clinical text on its own can provide a competitive alternative to traditional structured features (eg, lab values, vital signs). CANTRIP is able to generalize across nosocomial diseases without disease-specific feature extraction and is available at https://github.com/h4ste/cantrip.

Convolutional Neural Network Model for Intensive Care Unit Acute Kidney Injury Prediction

Introduction

Acute kidney injury (AKI) is common among hospitalized patients and has a significant impact on morbidity and mortality. Although early prediction of AKI has the potential to reduce adverse patient outcomes, it remains a difficult condition to predict and diagnose. The purpose of this study was to evaluate the ability of a machine learning algorithm to predict for AKI as defined by Kidney Disease: Improving Global Outcomes (KDIGO) stage 2 or 3 up to 48 hours in advance of onset using convolutional neural networks (CNNs) and patient electronic health record (EHR) data.

Methods

A CNN prediction system was developed to use EHR data gathered during patients’ stays to predict AKI up to 48 hours before onset. A total of 12,347 patient encounters were retrospectively analyzed from the Medical Information Mart for Intensive Care III (MIMIC-III) database. An XGBoost AKI prediction model and the sequential organ failure assessment (SOFA) scoring system were used as comparators. The outcome was AKI onset. The model was trained on routinely collected patient EHR data. Measurements included area under the receiver operating characteristic (AUROC) curve, positive predictive value (PPV), and a battery of additional performance metrics for advance prediction of AKI onset.

Results

On a hold-out test set, the algorithm attained an AUROC of 0.86 and PPV of 0.24, relative to a cohort AKI prevalence of 7.62%, for long-horizon AKI prediction at a 48-hour window before onset.

Conclusion

A CNN machine learning-based AKI prediction model outperforms XGBoost and the SOFA scoring system, revealing superior performance in predicting AKI 48 hours before onset, without reliance on serum creatinine (SCr) measurements.