Evolution of CVD Risk assessment in New Zealand.
The first version of PREDICT CVD (aka PROMPT) was created based on the 1999 guidelines and implemented the Framingham (Anderson) equation with some additional ‘Clinically High’ classifications to cater for the exclusions which applied to certain sub-groups.
The first version of PREDICT CVD/Diabetes was originally based on the New Zealand Cardiovascular and Type 2 Diabetes Guidelines (both published in Dec 2003), and utilised the Framingham risk assessment equations as a base, with further NZ specific adjustments applied (5% upward adjustments, minimum of 15% classifications and Clinically High considerations).
Predict was updated in 2004, 2009, 2012, 2016 and 2018 to maintain the currency of decision support content and align with new guidance.
Throughout this period, the Vascular Informatics Using Epidemiology and the Web (VIEW) / (Predict research group), within the University of Auckland were collecting and linking Primary and Secondary Care Predict data, with the main objective of creating a more locally tailored and NZ specific risk prediction equation.
In 2018, VIEW’s research was published in the Lancet journal and was then ready for use and further implementation within New Zealand.
In preparation for this Enigma created ‘Your Heart Engine‘ as a centralised implementation of the risk prediction algorithms; making the group’s work available in a quality-assured manner for wider, NZ usage.
Versions of algorithms implemented within YHE:
- Updated version based on handbook update, metabolic syndrome no longer considered an uplifting factor
- CKD a clinically high consideration
- First draft proofing versions of VIEW equations.
- Revised VIEW equations.
- Largely based on AUS primary care equations, but with some specific handling for certain rail-standards-based quirks.
- Revised VIEW equations.
- Revised VIEW equations. – As published in Lancet Journal, never released into a production product.
- Revised VIEW equations with BMI coefficient included. – Intended to be the initial release algorithm for Predict CVD Diabetes with 2018 consensus statement updates.
The algorithm published within the Lancet Journal was the first of the VIEW-derived (Predict-based) NZ Primary Prevention equations. This was almost immediately superseded by a further update from the VIEW team which included BMI as a new coefficient. The first implementation of the 2018 Consensus Statement within PREDICT uses the “NZ Primary Prevention Equations” – which more specifically implement the “YHE2018-BMI” enabled equation (not the version published in the Lancet).
From here, routine (regular) algorithm updates are expected, not only with new algorithms being introduced, but with existing coefficient values being further tweaked as the base data collections continue to grow both in terms of their size, but also the follow-years contained within them.
Expected future revisions of YHE algorithms:
- SMI specific enhancements created by Dr Ruth Cunningham, in conjunction with VIEW data linkage.
- Diabetes-specific enhancements of core VIEW outputs.
- The very first, NZ based secondary prevention equations (sourced from VIEW outputs).
- These strings are to be seen as Enigma’s codes – which relate back to, and identify which YHE algorithm implementation was used to generate the risk.
- These codes should not be used by other tool vendors as ‘equivalents’ to YHE – as their implementation may well vary from YHE’s.
- Other tool vendors who use YHE’s APIs are eligible to use these codes within their writeback of the generated CVR value.
- Enigma continues to use the CVR screening term which we introduced into Medtech when Predict was installed in practices from 2005 and beyond.
- This version-code is being pushed back into ‘Slot 4’ of that CVR code into a measurement titled ‘RA Algorithm’.
Other tool vendors are free to utilise that same measurement slot in an existing CVR screening term, however, we would politely request that they do not re-use our specific codes as detailed above as that would potentially misrepresent how the risk was calculated.
Why is this important?
With the various versions of algorithms, the level of CV Risk predicted for an individual has changed significantly.
When considering what to do for an individual, clinical guidance has been offered, paired with the risk algorithm used; treatment thresholds have been set according to the risk level which is expected *from that algorithm*. Accordingly, recall periods were also set in conjunction with the risk algorithm used. – When implementing recalls or KPI analysis against stored data, it will be important for the tools which do this job to consider the context of the risk, and how it ought to have been treated *at the time which it was created*.
For example, a patient who was risk assessed in 2017, using the NZ2012 algorithm, who had a 6% risk, would have expected a recall for a follow-up risk assessment in 2 years, and any tool which observes whether or not that recall was performed, or when that patient is due, ought to continue to observe that 2 year period. However, if that same patient’s risk was 6% generated using the YHE2018-BMI equations, then their recall period would be 5 years. It’s important that the risk level can be considered in the appropriate context, and so the algorithm reference needs to be stored along side the recorded risk.