|Seminar by Professor Rod Jackson
Professor of Epidemiology, University of Auckland
Monday 18th September 2017 at 1pm
@ the Seminar Room, Strangeways Research Laboratory, Department of Public Health and Primary Care
Most cardiovascular risk prediction equations need to be updated or replaced
External validation of existing equations and development of new equations using linked electronic records of 400,000 primary care patients.
The majority of cardiovascular disease (CVD) risk prediction equations are derived from cohorts established last century that included participants at higher risk, on less treatment and who were less socio-economically and ethnically diverse, than patients the equations are now applied to. The validity of most of these equations is uncertain, largely because the relevant cohorts required to evaluate them are rare. Approximately 90% of middle-aged New Zealanders have had quantitative CVD risk assessments using a Framingham Heart Study equation (FHSE), since risk assessment was made a national primary healthcare priority in 2008. We piggybacked on this initiative to recruit a nationally representative primary care cohort to: i. assess the validity of the 2013 Pooled Studies equations (PCEs) that recently replaced FHSEs in the US and elsewhere; and ii. develop new equations if justified.
The PREDICT cohort study automatically recruits participants when general practitioners utilising PREDICT decision support software, calculate patients’ 5-year CVD risk using a modified 1991 FHSE. Baseline CVD risk factors were prospectively linked to ICD-coded CVD hospitalisations and deaths in national databases. To determine if new equations are justified, firstly the calibration of the 2013 US PCEs were assessed in the PREDICT cohort, and secondly, Cox models including prognostic indices from the PCEs were updated by adding new variables available in PREDICT. Finally, new equations including all variables were developed using Cox models and the performance of PCE and new equations compared.
Approximately 90% of eligible patients were recruited and only 80 participants had incomplete risk factor data. The 401,752 participants aged 30-74 years, recruited between 2002 and 2015, experienced 15,386 first major CVD events (9.8% were fatal and 55% met the PCE definition of atherosclerotic CVD ) during 1,685,521 person-years (mean 4•2 years) follow-up. The PCEs over-predicted 5-year CVD risk by more than 50%, which could not be explained by new treatment initiated after baseline. Additional variables representing Polynesian, South Asian and other Asian ethnicities and an area-based social deprivation variable identified substantial groups of patients with predicted CVD risk between 30% lower and 60% higher than based on either the FHSE and PCE scores alone. New equations including additional predictors performed significantly better on all model assessment metrics.
We demonstrate that a contemporary cohort representing typical patients that prediction equations are applied to, can be recruited using decision support integrated with routine linked electronic patient records. We show that new US risk prediction equations substantially overestimate observed CVD risk and this is not explained by new drug treatment. Recalibration would be insufficient to improve the performance of the PCEs and simple variables representing common ethnicities and socio-economic deprivation should be incorporated into current equations. New equations are provided that could be implemented in the 15-20 high-income countries with relatively similar CVD events rates and healthcare systems to New Zealand.
This talk will be chaired by Dr Angela Wood, Senior University Lecturer in Biostatistics, Cardiovascular Epidemiology Unit, Department of Public Health and Primary Care