Development of risk models for the prediction of new or worsening acute kidney injury on or during hospital admission: a cohort and nested study

Time Series AI Model for Acute Kidney Injury Detection Based on a Multicenter Distributed Research Network: Development and Verification Study

Background

Acute kidney injury (AKI) is a marker of clinical deterioration and renal toxicity. While there are many studies offering prediction models for the early detection of AKI, those predicting AKI occurrence using distributed research network (DRN)-based time series data are rare.

Objective

In this study, we aimed to detect the early occurrence of AKI by applying an interpretable long short-term memory (LSTM)-based model to hospital electronic health record (EHR)-based time series data in patients who took nephrotoxic drugs using a DRN.

Methods

We conducted a multi-institutional retrospective cohort study of data from 6 hospitals using a DRN. For each institution, a patient-based data set was constructed using 5 drugs for AKI, and an interpretable multivariable LSTM (IMV-LSTM) model was used for training. This study used propensity score matching to mitigate differences in demographics and clinical characteristics. Additionally, the temporal attention values of the AKI prediction model's contribution variables were demonstrated for each institution and drug, with differences in highly important feature distributions between the case and control data confirmed using 1-way ANOVA.

Results

This study analyzed 8643 and 31,012 patients with and without AKI, respectively, across 6 hospitals. When analyzing the distribution of AKI onset, vancomycin showed an earlier onset (median 12, IQR 5-25 days), and acyclovir was the slowest compared to the other drugs (median 23, IQR 10-41 days). Our temporal deep learning model for AKI prediction performed well for most drugs. Acyclovir had the highest average area under the receiver operating characteristic curve score per drug (0.94), followed by acetaminophen (0.93), vancomycin (0.92), naproxen (0.90), and celecoxib (0.89). Based on the temporal attention values of the variables in the AKI prediction model, verified lymphocytes and calcvancomycin ium had the highest attention, whereas lymphocytes, albumin, and hemoglobin tended to decrease over time, and urine pH and prothrombin time tended to increase.

Conclusions

Early surveillance of AKI outbreaks can be achieved by applying an IMV-LSTM based on time series data through an EHR-based DRN. This approach can help identify risk factors and enable early detection of adverse drug reactions when prescribing drugs that cause renal toxicity before AKI occurs.

Acute Kidney Injury: From Diagnosis to Prevention and Treatment Strategies

Acute kidney injury (AKI) is characterized by an acute decrease in renal function that can be multifactorial in its origin and is associated with complex pathophysiological mechanisms. In the short term, AKI is associated with an increased length of hospital stay, health care costs, and in-hospital mortality, and its impact extends into the long term, with AKI being associated with increased risks of cardiovascular events, progression to chronic kidney disease (CKD), and long-term mortality. Given the impact of the prognosis of AKI, it is important to recognize at-risk patients and improve preventive, diagnostic, and therapy strategies. The authors provide a comprehensive review on available diagnostic, preventive, and treatment strategies for AKI.

Artificial Intelligence in Acute Kidney Injury Risk Prediction

Acute kidney injury (AKI) is a frequent complication in hospitalized patients, which is associated with worse short and long-term outcomes. It is crucial to develop methods to identify patients at risk for AKI and to diagnose subclinical AKI in order to improve patient outcomes. The advances in clinical informatics and the increasing availability of electronic medical records have allowed for the development of artificial intelligence predictive models of risk estimation in AKI. In this review, we discussed the progress of AKI risk prediction from risk scores to electronic alerts to machine learning methods.

Nephrotoxin-Related Acute Kidney Injury and Predicting High-Risk Medication Combinations in the Hospitalized Child

BACKGROUND

In the hospitalized patient, nephrotoxin exposure is one potentially modifiable risk factor for acute kidney injury (AKI). Clinical decision support based on nephrotoxin ordering was developed at our hospital to assist inpatient providers with the prevention or mitigation of nephrotoxin-related AKI. The initial decision support algorithm (Algorithm 1) was modified in order to align with a national AKI collaborative (Algorithm 2).

OBJECTIVE

Our first aim was to determine the impact of this alignment on the sensitivity and specificity of our nephrotoxin-related AKI detection system. Second, if the system efficacy was found to be suboptimal, we then sought to develop an improved model.

DESIGN

A retrospective cohort study in hospitalized patients between December 1, 2013 and November 30, 2015 (N = 14,779) was conducted.

INTERVENTIONS

With the goal of increasing nephrotoxin-related AKI detection sensitivity, a novel model based on the identification of combinations of high-risk medications was developed.

RESULTS

Application of the algorithms to our nephrotoxin use and AKI data resulted in sensitivities of 46.9% (Algorithm 1) and 43.3% (Algorithm 2, P = .22) and specificities of 73.6% and 89.3%, respectively (P < .001). Our novel AKI detection model was able to deliver a sensitivity of 74% and a specificity of 70%.

CONCLUSIONS

Modifications to our AKI detection system by adopting Algorithm 2, which included an expanded list of nephrotoxins and equally weighting each medication, did not improve our nephrotoxin-related AKI detection. It did improve our system's specificity. Sensitivity increased by >50% when we applied a novel algorithm based on observed data with identification of key medication combinations.

Development and validation of a prediction model for the probability of responding to placebo in antidepressant trials: a pooled analysis of individual patient data

BACKGROUND

Identifying potential placebo responders among apparent drug responders is critical to dissect drug-specific and nonspecific effects in depression.

OBJECTIVE

This project aimed to develop and test a prediction model for the probability of responding to placebo in antidepressant trials. Such a model will allow us to estimate the probability of placebo response among drug responders in antidepressants trials.

METHODS

We identified all placebo-controlled, double-blind randomised controlled trials (RCTs) of second generation antidepressants for major depressive disorder conducted in Japan and requested their individual patient data (IPD) to pharmaceutical companies. We obtained IPD (n=1493) from four phase II/III RCTs comparing mirtazapine, escitalopram, duloxetine, paroxetine and placebo. Out of 1493 participants in the four clinical trials, 440 participants allocated to placebo were included in the analyses. Our primary outcome was response, defined as 50% or greater reduction on Hamilton Rating Scale for Depression at study endpoint. We used multivariable logistic regression to develop a prediction model. All available candidate of predictor variables were tested through a backward variable selection and covariates were selected for the prediction model. The performance of the model was assessed by using Hosmer-Lemeshow test for calibration and the area under the ROC curve for discrimination.

FINDINGS

Placebo response rates differed between 31% and 59% (grand average: 43%) among four trials. Four variables were selected from all candidate variables and included in the final model: age at onset, age at baseline, bodily symptoms, and study-level difference. The final model performed satisfactorily in terms of calibration (Hosmer-Lemeshow p=0.92) and discrimination (the area under the ROC curve (AUC): 0.70).

CONCLUSIONS

Our model is expected to help researchers discriminate individuals who are more likely to respond to placebo from those who are less likely so.

CLINICAL IMPLICATIONS

A larger sample and more precise individual participant information should be collected for better performance. Examination of external validity in independent datasets is warranted.

TRIAL REGISTRATION NUMBER

CRD42017055912.

A Predictive Model for Assessing Surgery-Related Acute Kidney Injury Risk in Hypertensive Patients: A Retrospective Cohort Study

Background

Acute kidney injury (AKI) is a serious post-surgery complication; however, few preoperative risk models for AKI have been developed for hypertensive patients undergoing general surgery. Thus, in this study involving a large Chinese cohort, we developed and validated a risk model for surgery-related AKI using preoperative risk factors.

Methods and Findings

This retrospective cohort study included 24,451 hypertensive patients aged ≥18 years who underwent general surgery between 2007 and 2015. The endpoints for AKI classification utilized by the KDIGO (Kidney Disease: Improving Global Outcomes) system were assessed. The most discriminative predictor was selected using Fisher scores and was subsequently used to construct a stepwise multivariate logistic regression model, whose performance was evaluated via comparisons with models used in other published works using the net reclassification index (NRI) and integrated discrimination improvement (IDI) index.

Results

Surgery-related AKI developed in 1994 hospitalized patients (8.2%). The predictors identified by our Xiang-ya Model were age, gender, eGFR, NLR, pulmonary infection, prothrombin time, thrombin time, hemoglobin, uric acid, serum potassium, serum albumin, total cholesterol, and aspartate amino transferase. The area under the receiver-operating characteristic curve (AUC) for the validation set and cross validation set were 0.87 (95% CI 0.86–0.89) and (0.89; 95% CI 0.88–0.90), respectively, and was therefore similar to the AUC for the training set (0.89; 95% CI 0.88–0.90). The optimal cutoff value was 0.09. Our model outperformed that developed by Kate et al., which exhibited an NRI of 31.38% (95% CI 25.7%-37.1%) and an IDI of 8% (95% CI 5.52%-10.50%) for patients who underwent cardiac surgery (n = 2101).

Conclusions/Significance

We developed an AKI risk model based on preoperative risk factors and biomarkers that demonstrated good performance when predicting events in a large cohort of hypertensive patients who underwent general surgery.