Re:QRISK2 validation by ethnic group
We thank Professor Hippisley-Cox and colleagues for their comments on our paper. We agree that our study has limitations, but consider that we have stated these clearly within the manuscript. We believe that, notwithstanding these limitations, we bring new information regarding the performance of these widely used cardiovascular risk scores in three British ethnic groups. Importantly, these analyses were conducted independently of the authors of QRISK2 and Framingham risk scores and we report ethnicity-specific findings. Our take-home messages are that these risk scores performed differently and only modestly in the SABRE cohort in identifying individuals at high risk who would later go on to suffer events (particularly in South Asian women and in African Caribbean men and women). We suggest that further independent validation in other multi- ethnic datasets would be helpful.
With regards to specific comments, our responses are below: We agree that numbers of events were few in women, and indeed in African Caribbeans of both sexes - for this reason we urged caution in interpretation of the findings in these groups under 'Strengths and limitations'. The SABRE cohort is preponderantly male due to its initial objectives and in men there were 387 events (as stated). We show ethnicity specific % risk in men and women in figure 1, from which numbers of events can be estimated. For information, there were 173, 189 and 25 events in European, South Asian and African Caribbean men, respectively.
With regard to definition of outcome, we attempted to mirror the definition used in the derivation of QRISK2(1). We included participants who underwent coronary revascularisation procedures as having incident coronary heart disease (CHD), because such patients are most likely to be recorded as having CHD according to one of the nationally agreed Read codes used in the quality and outcomes framework for general practice (as reported in the methods for derivation of QRISK2)(1). Space limitations prevented us from reporting numbers of events recorded from different sources and we did not have complete data from all three sources for all participants. In summary, 302 events (65%) were derived from general practice record review(n=219) or death certificates(n=83). A further 31 events were identified solely through participant recall and the remaining 123 events were identified from Hospital Episode Statistics(HES). Inter-reporter agreement for GP record review and participant recall was excellent (kappa=0.79). For the 1308 participants where we could compare events reported from all three sources, the kappa value indicated good agreement( 0.63). We should point out that admission practices have changed and that early HES records (before 1997) are generally regarded as less than complete than more recent records, hence it is highly likely that HES records for the 10 year period following our baseline (1988-91) underestimate rather than overestimate events for that period. Sensitivity analyses restricted to events identified solely by GP record review produced similar findings to those reported in the paper.
We agree that South Asians and African Caribbeans in our cohort are all first generation migrants and we specifically state that 'our findings may not be generalisable to future generations in each ethnic minority group', although the extent to which UK ethnic differences in CHD risk are changing across generations remains to be established.
We agree also that a number of data items were missing for our calculation of the QRISK2 scores, and we noted this limitation under 'strengths and limitations' . We note that the derivation and validation datasets for QRISK and QRISK2 contained less than 1% with rheumatoid arthritis and less than 0.17% with chronic kidney disease; in addition less than 0.6% had atrial fibrillation(ref). Such small amounts of missing data are unlikely to impact on our findings. We agree that data on family history were collected only at follow-up and that this is a key shortcoming, however, as reported, sensitivity analyses which included history of parental CVD at age less than 60 years in a subset(n=1854) did not improve calibration or classification, although it did improve discrimination in African Caribbeans (but not in Europeans or South Asians).
It is a strength of our study that we had complete data for 3674 participants(94% of those followed-up) for key baseline risk factors: blood lipids, smoking , blood pressure and body mass index and that ethnicity in all participants was defined based on self-assignment and country of birth of parents. In contrast, the main analyses for derivation of QRISK2, of necessity, used multiple imputation to replace missing values for systolic blood pressure, cholesterol/HDL ratio (available in <35%), smoking status, and body mass index. (1)
We thank Hippisley-Cox and colleagues for drawing attention to the 2013 online QRISK2 update; we agree that the increase in patient self- assigned ethnicity (now at 50%) will have improved the accuracy of this tool. Nevertheless, given the importance of accurate risk scoring, as recently highlighted by the critical responses to the recently released AHA guidelines(2), further validation of the QRISK2 update as it applies to ethnic groups, would seem warranted.
Therese Tillin, Alun D Hughes, Peter Whincup, Jamil Mayet, Naveed Sattar, Paul M McKeigue, Nish Chaturvedi
1. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, Minhas R, Sheikh A, Brindle P: Predicting cardiovascular risk in England and Wales: prospective derivation and validation of QRISK2. BMJ 336:1475-1482, 2008
2. Ridker PM, Cook NR: New American guidelines for prevention of cardiovascular disease. Lancet 2013, published online 20/11/2013, doi:10.1016/S0140-6736(13)62388-0
Conflict of Interest: