Clinical Risk Assessments and Calculators

Clinical Risk Assessments and Calculators

Recent studies have indicated that the current risk calculators are out of date and not specific to local contexts and population. This project is testing the hypothesis:

Can we build risk calculators that are more accurate for the local population and context?

The focus of this research is to validate the current clinical risk of readmission assessment (the LACE score) and compare it with the “patient at risk for re-hospitalisation” (PARR) score used in acute care settings in NZ. This project involves designing and developing a predictive model for clinical readmission risk assessment in order to test the prediction accuracy within a short period of time such as 30, 60, 90-day readmission risk from the date of discharge.

The data variables used include, for LACE: length of stay (days), acute admission, the Charlson Comorbidity Index which includes (age, diabetes mellitus, liver disease, solid tumor, AIDS, moderate to severe CKD, CHF, myocardial infarction, COPD, peripheral vascular disease, CVA or TIA, dementia, hemiplegia, connective tissue disease, leukaemia, malignant lymphoma, peptic ulcer disease) and number of ED visits within six months.

For PARR: gender, age, race (Maori, Pacific, Asian, others), cost weight of last admission, code for last submission, diagnoses for last admission, and number of acute admissions in the previous 90 days, 180 days and 2 years.


  • The risk of hospital readmission model achieved AUROC of 0.727±0.003 and F1 Score of 0.339±0.004 when tested with the entire dataset (generic model).
  • The medicine-specific model has achieved an AUROC of 0.717±0.005 and F1 Score of 0.360±0.003
  • The surgery-specific model has achieved an AUROC of 0.732±0.005 and F1 Score of 0.309±0.005.

This study shows how ineffective the two models for risk of hospital readmission (LACE and PARR) are when applied to the New Zealand population and local context. Further research is required to measure their full impact on admissions and overall clinical acceptance. This also highlights that there is a high need for risk prediction and risk adjustment models to become more accurate, and to be utilised in hospitals for incentives or penalties.

A follow-on project is underway to investigate the use of these prediction models in clinical practice, develop a day 2 model and validate and evaluate this in clinical practice.


Read the published journal article here.

We presented our research at:

  • Health Informatics Conference (HIC), Sydney, Australia, August, 2018
  • 35th International Conference of the International Society for Quality in Health Care (ISQua’ 2018), Kuala Lumpur, Malaysia, 23 – 26 September 2018
  • Precision Driven Health Research Day, Auckland, September, 2018
  • Health Informatics New Zealand (HINZ) Wellington, November 2018
  • Waitemata clinical group, i3, North Shore Hospital, WDHB, December 2018


Dr Mirza Baig – Principal Investigator – Health Informatics Specialist, Orion Health

Reece Robinson, Principal Engineer, Orion Health

Dr Edmond Zhang, Senior Data Scientist, Orion Health

May Lin Tye, Graduate Business Analyst, Orion Health

Professor Rod Jackson, professor of epidemiology in the Section of Epidemiology and Biostatistics, UoA

Dr Robyn Whittaker, Public Health Physician, Waitemata DHB

Delwyn Armstrong, Health Intelligence Manager, Waitemata DHB