Early Identification of Acute Kidney Injury Using Deep Recurrent Neural Nets, Presented with Probable Etiology

Abstract SignificanceIn this SBIR project we propose to develop Previsea novelsoftware based clinical decision supportCDSsystem for predicting acute kidney injuryAKIand attributing AKI to one of several causal mechanismsetiologiesPrevise will use machine learning methods and information drawn from the electronic health recordEHRto identify the early signs of acute kidney injuryBy doing so before the clinical syndrome of AKI is fully developedPrevise will give clinicians the time to intervene with the goals of preventing further kidney damageand decreasing the sequelae of AKICombining this prediction module with a second module that suggests the underlying causes responsible for an incipient or full AKIPrevise will enable clinicians to make earlier and better informed treatment decisions for AKI patientsResearch QuestionCan a machinelearning based CDS predict the development and progression of AKI in hospitalized patientshours in advance of KDIGO stageorwith performance providing an area under the receiver operating characteristic curveAUROCof at leastIs it possible to use a Bayesian model to infer the cause of AKI with high accuracyAUROCPrior workWe have developed a prototype version of the Previse system which predicts AKI up tohours in advance of KDIGO stageorcriteriawith an AUROC nearWe have previously developed machine learning based predictive tools for sepsisin hospital mortalityand other adverse patient events with performance significantly improved over commonly used rules based scoring systemsSpecific AimsTo predict the onset of chart abstracted KDIGO stageorAKI in retrospective datahours in advanceAimto use data drawn from the EHR to identify the cause of AKI at time of onset with high accuracyand to present this causal inferenceits likelihoodand relevant evidence supporting it in a human interpretable fashionAimMethodsWe will predict the onset of AKI using a deeprecurrent neural networkRNNThis expressivenonlinear classifier will incorporate time series information in the qualitative portions of the EHR and will also incorporate features derived from text componentssuch as radiology reportsLabeling AUROC ofor higher athours pre KDIGO AKI will constitute success in AimIn Aimwe will train a dynamic Bayesian network to identify the cause of AKIWe will train this system using semi supervised methodswhere the causes of a set of AKI examples will be hand annotated by clinician expertsthese examples will be split into two groupswith some used for training and the remainder for testingAimwill be successful if this training results in etiology identification accuracy of at leastin the test setFuture DirectionsFollowing the proposed workthe combined Previse system will be deployed for prospective studies at partner hospitals Narrative Acute kidney injuryAKIaffects approximatelyof all hospitalized patients each year in the United Statescausing short term and lasting harmincluding increased risk of mortalityHoweverthe existing tools for acute kidney injury detection have failed to offer significant anticipation of full blown AKIor to provide the insight into the AKI s root causewhich are needed to make an impact on patient outcomesWe will develop Previsea machine learning based prediction system that continuously monitors for incipient AKI and offers clinicians probable root causes of AKI