Sankalp KHANNA, Derek IRELAND , David ROLLS and Justin BOYLE CSIRO Australian e-Health Research Centre, Brisbane, Australia July 2018 Implementing a Risk Stratification Algorithm HEALTH AND BIOSECURITY www.csiro.au
Sankalp KHANNA , Derek IRELAND , David ROLLS and Justin BOYLECSIRO Australian e-Health Research Centre, Brisbane, Australia
July 2018
Implementing a Risk Stratification Algorithm
HEALTH AND BIOSECURITYwww.csiro.au
Patient Flow @ CSIRO AEHRC
Implementing a Risk Stratification Algorithm
Aims : • Improving public hospital
performance through efficiency improvements
• Creating an evidence base to support policy and decision making
Science Areas : • Visualisation• Statistical Modelling• Machine Learning• Stochastic Optimisation• Distributed Constraint Reasoning• Discrete Event Simulation
2 |
Implementing a Risk Stratification Algorithm3 |
Risk Stratification for Hospital AvoidancePrimary Care Setting
Predictive Algorithm Driven Risk Stratification to inform Recruitment
Employed across up to 200 GP Practices and Aboriginal Health Services starting Dec 2017
Acute Care Setting
Predictive Algorithm Driven Risk Stratification to inform Discharge Planning
12 Month trial at Queensland Metropolitan Hospital commenced April 2018
Implementing a Risk Stratification Algorithm4 |
Project Brief• To develop, implement and evaluate a web-based risk
stratification algorithm that can be used in-hospital to identify chronic disease patients with a high risk of re-hospitalisation.
• What are we predicting
• Unplanned re-admission within 30 days of discharge from hospital
• Unplanned ED re-presentation within 30 days of discharge from hospital
• Timeline
Chronic Disease Patient Admitted to Hospital
Risk Score generated overnight
Risk score used by care teams for appropriate interventions and
care/discharge planning
Next morning
Trial Apr 2018 to Mar 2019
Post-trial evaluation Apr 2019 to Jun 2019
Implementing a Risk Stratification Algorithm5 |
Study Design
Patients who, over a 5 year period, as an Emergency or Admitted Patient :• Attended Logan Hospital, and• Had at least one Chronic Disease visit
(any QLD hospital)
Patient Cohort
• Emergency Data (EDIS/FirstNet)• Inpatient Data (QHAPDC/ePADT)• Mortality Data (Death Registry)• Pharmacy dispensing information (eLMS)• Pathology test results (AUSLAB)
Data Used for Modelling & Validation
• Patients stays in previous 180 days• ED presentations in previous 180 days• Marital status• Age• Indigenous status• SEIFA• Admission source• Admission unit• Care type• Elective status• Planned same day status• Binary flags for routine dialysis• Number of medication records• Binary flags for medication• Binary flags for abnormal pathology results
Predictor Variables in Final Models
Implementing a Risk Stratification Algorithm6 |
Trial MethodologyData Flows • Algorithm hosted on QLD Health Virtual Machine
• Risk profiling once per day (1 AM)
• Risk score provided for each patient with a chronic
disease history that is staying overnight
Evaluation Research QuestionsDoes the algorithm:
1. Improve the process of identifying patients at high risk
of unplanned re-hospitalisation?
2. Reduce re-hospitalisation rates?
3. Provide information to staff not readily available at the
time of discharge planning?
Implementing a Risk Stratification Algorithm7 |
The Web-based Decision Support Tool - Screenshots
Implementing a Risk Stratification Algorithm8 |
The Web-based Decision Support Tool - Screenshots
Implementing a Risk Stratification Algorithm9 |
The Web-based Decision Support Tool - Screenshots
Implementing a Risk Stratification Algorithm
Questions ?
For more information, please contact : Sankalp KhannaSenior Research Scientist
t +61 7 3253 3629e [email protected] www.aehrc.com
10 |