Australian researchers have developed cardiovascular risk prediction models that could one day allow GPs to assess a patient’s CVD risk at the click of a button.
The NSW team drew on data from 680 general practices including more than 850,000 patients, including risk factors like depression, COPD, HbA1c, BMI and eGFR alongside traditional predictors to build tools to estimate the five-year risk of cardiovascular disease.
Writing in Heart [link here], the authors said their study had shown the value of using primary care data to estimate CVD risk, with the full 28-paramater model achieving a Harrell’s C score 0.803 for females and 0.772 for males.
The authors also evaluated the performance of a model based on the same variables as the AusCVDRisk calculator – a tool which was adapted from the New Zealand PREDICT-1 equation but had not been validated using Australian data.
They found their full model out-performed the refitted AusCVDRisk model, which had Harrell’s C of 0.78 for females and 0.75 for males.
“Our models diverge from PREDICT, from which AusCVDRisk is derived, by incorporating a broader range of clinical variables,” they said.
The inclusion of a wider range of prediction factors aligned more closely with the PREVENT and QR4 equations developed in the UK and US, the investigators said.
The equations also cast a wider net than existing international models, expanding the range of chronic conditions considered, including COPD and asthma.
While the AusCVDRisk calculator was adapted from New Zealand equations, the models built in this study contained a broader range of biomarkers and conditions and were more directly relevant to Australian clinical data, the authors said.
Door open for automated risk screening
Building these models based on data collected from GP systems opened up the possibility for these tools to one day be used to facilitate automated CVD risk prediction.
“Since our equations were developed using data collected through routine Australian general practice clinical software, they are potentially suitable for incorporation into this software to automate estimation of absolute cardiovascular risk,” the research team wrote.
“Automation of risk prediction obviates the requirement to specifically enter risk factor data, and thus presents the potential to identify patients-at-risk who might otherwise be missed and prompt early initiation of preventive interventions.”
Even if risk scores could only be calculated for one third of patients in the cohort used for this study, that would represent a substantial improvement in Australian clinical practice, given only around 10% of patients have a CVD risk score in their electronic medical record, the authors said.
Former CSANZ president Professor Clara Chow said it made sense to integrate and automate CVD risk prediction into clinical practice software.
“This paper does demonstrate feasibility of such an approach. The question is which CV risk algorithm to integrate into clinical practice software,” Professor Chow told the limbic.
“This paper actually derives a new algorithm, in the process of demonstrating feasibility of the concept, from a NSW linked dataset.
“Their new algorithm does demonstrate good overall fit as measured by the c-statistic. The authors can’t really do a direct comparison to the current AusCVD calculator for several reasons, including not having all the variables for the current calculator and not having the actual algorithm.”
A strength of the algorithm in this study was that it was from an Australian population, albeit a small subset of the population that attended primary care centres in NSW, she said.
“Ideally we probably want to have a system that is continually deriving algorithms, is being continually monitored for performance, and can use a the largest relevant diverse dataset possible e.g. the whole of Australia,” she said.
This study focused on a five-year risk of cardiovascular events, which made direct comparisons difficult with the 10-year predictions created by the PREVENT and QR4 tools.
“However, our study’s focus on 5-year risk may offer more immediate clinical utility for patients, compared with longer term predictions,” the authors said.
The cohort was limited to Australian adults aged 30 to 74, meaning validation of the tool for those aged over 75 was also needed.