A PROJECT FUNDED BY
PRECISION DRIVEN HEALTH INITIATIVE
Health Data Pool (HDP) is the data store of primary health care records and hospital admissions data from secondary care.
View an end-to-end patient record
Assist in making informed decisions
Conduct population health analysis
Develop precision health models
Undertake preventative health measure
Benefits of HDP for healthcare providers
To instantiate the use of HDP we developed a predictive model – Hospital Admission Risk Predictor (HARP).
Predicts the hospital admission risk
Predictive health models can be built leveraging the Health Data Pool.
Precision Health Models
Predicts hospital readmission risk
Have an idea for any future models
Summary of Results
To select a set of patients for potential intervention to avoid admission we selected a 50% “cut-off” probability. This is best done in discussion with stakeholders, taken account of the relative costs and benefit associated with true and false positives.
Assuming, 50% as our cut-off probability, the model returned 27 true positives and 40 false positives. If we plot these set of points: x false positives (1 – Specificity) versus y true positives (sensitivity) for all predictions we can plot a gives the Receiver Operating Characteristic (ROC) curve. The Area Under the Curve (c-statistic) for the model was 0.71.
Logistic regression modelling approach
We adopted the logistic regression modelling approach for HARP predictions. For testing and verification of HARP data of 50,000 patients have been collected/extracted from Whanganui DHB and Whanganui Regional Health Network (PHO). The project demonstrates key interfaces for data acquisition, prediction model development and distribution of risk information to users as responses to single queries or batches.
HARP Methodology & Results
© 2018 Farhaan Mirza, Barry Gribben, Mirza Baig et al