Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
Does QRISK describe a partially treated population?
- Eugene M.G. Milne (6 July 2007)
-
QRISK can NOT be used for treatment decisions
- L S Lewis (8 July 2007)
-
QRisk better than Framingham ?
- Michel J. Romanens (9 July 2007)
-
Doubts about QRISK score: total / HDL cholesterol should be important.
- Richard Peto (13 July 2007)
-
QRISK and health inequalities
- John Macleod, Chris Metcalfe, George Davey Smith (13 July 2007)
-
FRAMINGHAM, ASSIGN and QRISK Cardiovascular Risk Scores
- Hugh Tunstall-Pedoe, Mark Woodward (14 July 2007)
-
QRISK has great potential, but don't underestimate Framingham
- William G Simpson, and Sorrel A Abbott (15 July 2007)
-
QRISK - Methodological limitations?
- Marie Therese Cooney, Alexandra Dudina, Ian Graham - Prof of Preventive Cardiology and Consultant Cardiologist (18 July 2007)
-
Measures of smoking, deprivation and cardiovascular risk
- Sarah H Wild, Marlene Stewart, Jacqueline Price, F.Gerald Fowkes, Gordon Murray. (20 July 2007)
-
QRISK still leaves patients at risk or unnecessarily treated
- Gordon A A Ferns (21 July 2007)
-
Waist to Hip ratio should be included
- Ben D Ewald (23 July 2007)
-
QRISK - authors response
- Julia Hippisley-Cox, Julia Hippisley-Cox, Carol Coupland, Yana Vinogradova, John Robson, Margaret May, Peter Brindle (7 August 2007)
-
Multiple imputation needs to be used with care and reported in detail
- John B Carlin, Jonathan A. C. Sterne, Ian R. White, John B. Carlin, Patrick Royston, Michael G. Kenward, Angela M. Wood and James R. Carpenter (21 August 2007)
-
Setting for risk calculation will affect performance of QRISK
- Richard J McManus, Dr Jonathan Mant (21 August 2007)
-
Powerful stratification tool
- Guy Wilkinson (24 August 2007)
-
QRISK validation study published
- Peter M Brindle (18 October 2007)
-
Are risk calculators answering the wrong question?
- G Alastair Cooke (31 October 2007)
|
|
|||
|
Eugene M.G. Milne, Assistant Regional Director of Public Health Government Office for the North East, Citygate, Newcastle upon Tyne, NE1 4WH
Send response to journal:
|
This is an interesting and important paper which correctly identifies deficiencies in existing approaches to risk identification. (1) However, it contains no mention of the effect of interventions (other than at baseline) upon individuals whose data were used in the derivation and validation, and who may have been clinically identified as being at high risk during the period of the study. As a result, the paper implicitly assumes either that there was no prophylactic use of preventive measures between 1995 and 2007 (which we know to be untrue) or, paradoxically, that prophylactic use of interventions makes no difference to outcome (which would render detection redundant). Two features of the results suggest this may not be a trivial issue. Firstly, there was a clear disadvantage to patients with missing data - a plausible proxy for lack of intervention - in terms of event-free survival. Secondly, the proportional differences in gender risk between Framingham and QRISK could, at least partially, reflect gender differences in the use or efficacy of preventive interventions during the study period. (2, 3) Some of the overestimation of risk arising from other scores may be because they describe pre-prophylactic (in the case of Framingham) or largely pre-prophylactic treatment populations (ASSIGN), whereas the QRISK results describe outcomes in a population that was likely to have been receiving a more substantial level of such intervention. (4, 5) A counter to this may be the observation that there were minimal differences in the baseline survival function between the closed (older) and open (including more recent patients) cohorts. Increasing use of effective prevention might have been expected to increase these differences. It would be extremely interesting, if possible, to see a further analysis of this study with patients divided into those who, subsequent to their date of study entry, received a preventive measure versus those who did not. If a genuine difference existed between these groups, it would not only support at a 'real-world' population level the benefits of risk detection and intervention, but also help us to avoid slipping into under- recognition of high risk individuals by assuming a partially treated risk to be the same as an untreated risk. 1. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study 10.1136/bmj.39261.471806.55. BMJ 2007:bmj.39261.471806.55. 2. Lawlor DA, Bedford C, Taylor M, Ebrahim S. Geographical variation in cardiovascular disease, risk factors, and their control in older women: British Women's Heart and Health Study. J Epidemiol Community Health. 2003;57(2):134-40. 3. Ridker PM, Cook NR, Lee IM, Gordon D, Gaziano JM, Manson JE, et al. A randomized trial of low-dose aspirin in the primary prevention of cardiovascular disease in women. N Engl J Med 2005;352(13):1293-304. 4. Anderson KM, Odell PM, Wilson PW, Kannel WB. Cardiovascular disease risk profiles. Am Heart J 1991;121(1 Pt 2):293-8. 5. Woodward M, Brindle P, Tunstall-Pedoe H. Adding social deprivation and family history to cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;93(2):172-6. Competing interests: None declared |
|||
|
|
|||
|
L S Lewis, GP Surgery, Newport, Pembrokeshire SA42 0TJ
Send response to journal:
|
The paper from Hippisley-Cox et al. provides valuable new data modelling on the vexed question of accurate CVD risk prediction. But the study is fundamentally flawed, if the aim was to define an UNTREATED risk score. The study population has been subject to increasing preventive interventions, and the underlying death rate from CVD has fallen by some 40% in the last 10 years. Thus the QRISK score is calibrated to real population outcomes today, but disregards the fact that the population is partially-treated, as Eugene Milne points out. It is therefore invalid as a tool for deciding who to treat, and more importantly perhaps, who not to treat. As the ‘large numbers of people may have been overtreated’ News headlines testify, these political ethical and rationing questions dominate today’s environment. Notwithstanding the moral questioning of absolute-risk rationing decisions raised by Luc Bonneux (whereby the youngster with high risk- factors goes untreated for years, whilst the smoking 55-year-old male qualifies for a Statin) decisions on who can be most cost-effectively treated must be made. An important acknowledgement of the NICE guideline is that based on current Simvastatin pricing, the NHS can lower the threshold, and cost-effectively treat everyone at an untreated CVD 10-year risk of 20%. The Framingham Score remains well suited to the purpose as a yardstick upon which to move this decision threshold. Family History is accounted for, but deprivation and other variables may demand a local adjustment. QRISK offered the potentially valuable inclusion of Deprivation, but unfortunately only on an area, and not an individual basis. And QRISK perpetuates the categorical rather than continuous risks which ASSIGN and UKPDS attempt to address. As this paper’s Table 7 testifies, the use of the flawed QRISK score will greatly reduce the numbers of people, especially younger people, deemed eligible for preventive treatment. It is hard to see how it can be recalibrated to an UNTREATED UK population, except to those which ASSIGN, FRAMINGHAM and UKPDS risk scores have already been calibrated. In short, the QRISK can NOT be used for treatment decision-making. Competing interests: None declared |
|||
|
|
|||
|
Michel J. Romanens, MD CH-4600 Olten
Send response to journal:
|
Hello, looking at the ROC results of QRisk and Framingham there are no difference in diagnostic performance. This suggests, that for both men and women a simple correction factor may show, that Framingham and QRisk have identical diagnostic performance. For men e.g. a correction factor of 0.67 may yield similar diagnostic performance. Correction factors have been recommended by IAS for the PROCAM algorithm. Further, I miss the information of sensitivity and specificity of QRisk and Framingham for your cohort. Competing interests: None declared |
|||
|
|
|||
|
Richard Peto, Professor of Medical Statistics and Epidemiology CTSU, Richard Doll Building, Old Road Campus, Oxford, OX3 7LF
Send response to journal:
|
In the report of the QRISK cardiovascular disease risk score the ratio of total to HDL cholesterol appears to be of no relevance to the incidence of disease, despite the large size of the study population. This is very different from what has been reliably found in other studies. The contrast is so sharp that it suggests some serious error in the collection, processing or analysis of the QRISK data. Until more details of the materials and methods of this new QRISK study are made available, the reliable conclusion should be retained from previous studies that the ratio of total to HDL cholesterol (undistorted by lipid-lowering drugs) is strongly predictive of the primary incidence of coronary disease. Competing interests: None declared |
|||
|
|
|||
|
John Macleod, Reader in Clinical Epidemiology and Primary Care Department of Social Medicine, University of Bristol, Bristol, BS8 2PR, Chris Metcalfe, George Davey Smith
Send response to journal:
|
We congratulate Julia Hippisley-Cox and colleagues for their ingenious and innovative use of primary care data in the development of a new cardiovascular risk prediction tool. [1] Their paper and the accompanying editorial raise several important points. [1,2] Risk prediction tools place individuals on a scale of risk (discrimination), and then a threshold on that scale is chosen (calibration) above which individuals are considered at sufficient risk to warrant intervention. Hippisley-Cox and colleagues appear to suggest that they have achieved better discrimination between high and low risk individuals through including novel risk factors, particularly area deprivation, in their new prediction tool compared to other tools such as those based on the Framingham study and the more recently developed Scottish ASSIGN instrument. This is unlikely to be the case. The statistical indices of discrimination for the three different tools (table 5) such as area under the ROC curve are all very similar suggesting similar levels of discrimination amongst the three instruments. [1] Further, both QRISK and ASSIGN include area deprivation, so any difference between the two instruments is unlikely to be due to that factor. We recently investigated the influence on discrimination of consideration of several area and individual measures of material disadvantage in a Scottish population where the independent association of some of these with coronary risk was substantial. In no instance did a measure of material disadvantage make any practically useful contribution to improved risk prediction. [3,4] The closer coincidence of observed with expected events seen with QRISK is most likely to reflect better calibration since the instrument was derived using data from a random two thirds of the QRESEARCH population then validated using data from the remaining third. This is reflected in the similarity of individuals in the two datasets in tables 1 and 2 and is what the authors refer to as “home advantage”. [1] Such home advantage would probably also apply to the use of QRISK in another contemporary UK primary care database such as that compiled by the Health Improvement Network (THIN) as the constituency of contributing practices is very similar. [5] In our own study, recalibration of the Framingham prediction tool to our study population considerably improved its performance; inclusion of material disadvantage in this recalibrated instrument had essentially nothing further to add. [3,4] But if QRISK, for whatever reason, achieves more accurate prediction of cardiovascular risk in a contemporary UK population then what could be the argument against its widespread adoption in the UK, as the authors propose? The difficulty relates to the very problem the originators of QRISK suggest that their instrument will help solve, health inequalities. Inequalities in quality and quantity of life experienced between different social groups in the UK are substantially driven by differences in experience of cardiovascular disease. This is unlikely to be primarily due to doctors over-predicting risk amongst the affluent and under-predicting it amongst the disadvantaged. [2] Risk prediction as a basis for treatment decisions is only a relatively recent innovation and its use is still inconsistent. In the UK, recommended treatment thresholds are based more on economic (what the health service is perceived to be able to afford) than clinical (the trade off between likely benefits from the intervention versus likely iatrogenic harm) considerations. [6] Any effective health technology will tend to increase rather than decrease health inequalities, since effective health technologies are more successfully accessed by the socially advantaged, for many reasons. [7] Consequently there is likely to be a mismatch between individual risk (generally lower in people of higher social position) and receipt of coronary prevention. Better prediction might eventually address this mismatch, were treatment decisions to be more widely based on it. In the meantime however, “over-prediction” by tools such as Framingham or ASSIGN is not necessarily a problem in situations where this lowers the bar for eligibility for prevention in people whose capacity to benefit clearly outweighs their risk of iatrogenic harm. Table 6 of the paper shows that substantially more men from relatively deprived areas would receive treatment based on both Framingham and ASSIGN predictions compared to those of QRISK. For women, Framingham predictions are slightly lower and again those of ASSIGN are substantially higher. It is therefore difficult to see how the widespread adoption of QRISK, in preference to the tools currently available, as the basis for treatment decisions around coronary prevention in the UK “will help to minimise health inequalities”. [1] References 1. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK. A new cardiovascular disease risk score for the UK. BMJ 2007 doi: 10.1136/bmj.39261.471806.55 2. Bonneux L. Cardiovascular risk models. BMJ, doi:10.1136/bmj.39262.643090.47 3. Macleod J, Davey Smith G, Metcalfe C, Hart C. Can consideration of either psychological or material disadvantage improve coronary risk prediction in primary care? Society for Academic Primary Care Annual Scientific Meeting, London July 2007. 4. Macleod J, Davey Smith G, Metcalfe C, Hart C. Does consideration of either psychological or material disadvantage improve coronary risk prediction? Prospective observational study of Scottish men. J Epidemiol Community Health 2007 (in press) 5. Ryan R, McManus RJ, Mant J, Macleod J, Hobbs FDR. The prescribing of statins and mortality following a diagnosis of heart failure. Society for Academic Primary Care Annual Scientific Meeting, London July 2007. 6. National Institute for Health and Clinical Excellence. Statins for the prevention of cardiovascular events in patients at increased risk of developing cardiovascular disease or those with established cardiovascular disease. January 2006. www.nice.org.uk/TA094 7. Susser M. The technological paradox of health inequality, and a probe with a practical tool. J Epidemiol Community Health 2000;54:882-3. Competing interests: None declared |
|||
|
|
|||
|
Hugh Tunstall-Pedoe, Emeritus Professor and Senior Research Fellow Cardiovascular Epidemiology Unit, University of Dundee, DD1 9SY, Mark Woodward
Send response to journal:
|
As inventors of the ASSIGN cardiovascular score,(1) featured in this QRISK paper,(2)and despite being named in the acknowledgements, we had not been aware of this manuscript before its very rapid publication. We proposed a collaborative head-to- head comparison of QRISK with ASSIGN, and hope this will still be possible. Incorporating social deprivation and family history, the innovative features of ASSIGN, this partisan assessment nevertheless champions QRISK.It raises a number of technical questions, and is not free of anomalies and error, including even the reference to our ASSIGN paper. (1) Grandiose claims are made for a United Kingdom constituency despite trivial Scottish involvement, and ASSIGN’s adoption for Scottish use in recent SIGN guidelines.(3) Maybe it is the enormous GP database that impresses, explaining publication in the BMJ when ASSIGN was judged insufficiently new and interesting a year ago. But big numbers do not remedy bias. Others will need convincing that mass clinical data are as reliable as those derived from standardized measurement of risk factors by trained observers under controlled conditions, with full ten-year follow-up. QRISK is bizarrely anomalous in finding no significant independent effect of lipids on cardiovascular disease. Risk scores are used to prioritize treatment, ranking individuals by their estimated risk. Overall calibration of the risk estimate is less important than the fairness of the order of ranking — the discrimination of future cases and non-cases. This paper showed discrimination by Framingham, ASSIGN and QRISK to be virtually identical, although far from ideal. We developed ASSIGN because social deprivation and a positive family history were ignored in the Framingham score, but contribute importantly to cardiovascular risk.(1,4) Their omission from scoring was unfair to the socially deprived, and to those with family/ ethnic susceptibility. Who gets treatment is determined through an optimal cutpoint which is given quasi-mystical significance in clinical practice by calling it “predicted absolute risk”. Numbers for treatment can be changed either by moving the cutpoint or recalibrating the score. NICE is going round in a circle, first increasing numbers by lowering the Framingham cutpoint,(5)next taking fright at the implications in terms of workload, then championing a new score because it gives lower values anyway and saves work and money. Recalibration, or moving the cutpoint can both be done without a totally new score each time. Accurate prediction of individual risk is an illusion. “Prediction is very difficult, especially about the future” (Niels Bohr). All scores are derived from past data, yet event rates are changing. Treatment thresholds should be decided with reference to cost-effectiveness, benefits, risks and workload in relation to the current cardiovascular disease burden of the population, subgroup, or age-sex group concerned. Below age 35 few warrant prophylactic medication: from their late 60s virtually everybody is at substantial cardiovascular risk and is arguably in the Polypill situation.(6) Between these ages a cardiovascular score identifies those at substantial medium-term risk. Paradoxically, when many promote risk factor measurement and selective treatment from age 40 onwards, QRISK identifies 0.0% of women below age 55 and only 0.28% of men aged 45-54 for treatment. Framingham and ASSIGN are more intuitively correct for this. There are differences between ASSIGN and QRISK in handling smoking, diabetes and lipids. There are problems in how different measures of deprivation can be interchanged. Manipulation of age differs in the three scores, affecting the gradient in risk estimates with age. These are technical issues for further exploration, but they will have affected the key comparisons of the three scores reported in the QRISK paper. Meanwhile this interesting and important paper cannot be the last word on cardiovascular risk scoring, or on the Framingham and ASSIGN scores. 1. Woodward M, Brindle P, Tunstall-Pedoe H. Adding social deprivation and family history to the cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;92 (2):172-6. 2. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. BMJ 2007;0(2007): bmj.39261.471806.55v1 (5 July), doi:10.1136/bmj.39261.471806.55 3. Scottish Intercollegiate Guidelines Network. Risk estimation and the prevention of cardiovascular disease. A national clinical guideline. Edinburgh: SIGN, 2007 4. Tunstall-Pedoe H, Woodward M. By neglecting deprivation, cardiovascular risk scoring will exacerbate social gradients in disease. Heart 2006;92:307-10. 5. National Institute for Health and Clinical Excellence. Statins for the prevention of cardiovascular events in patients at increased risk of developing cardiovascular disease or those with established cardiovascular disease guidance type: Technology appraisal. In: Excellence NIfHaC, editor, 2006. 6. Wald NJ, Law MR. A strategy to reduce cardiovascular disease by more than 80%. BMJ 2003; 326: 1419-24. Competing interests: Originators of ASSIGN cardiovascular risk score |
|||
|
|
|||
|
William G Simpson, Consultant Chemical Pathologist Clinical Biochemistry Department, Aberdeen Royal Infirmary, Aberdeen AB25 2ZD, and Sorrel A Abbott
Send response to journal:
|
This paper has set out a proposal for a more contemporary and locally valid approach to risk estimation (1). The approach shows great promise, but as it stands has some fatal flaws, in particular the inclusion of patients on treatment. An attempt is made to address this flaw with regard to antihypertensive medication, but there is no mention of the circa 30% reduction of risk anticipated with statin treatment. It is no surprise therefore that cholesterol did not appear to contribute to risk in the entire QRISK population, whereas it is obviously a major risk factor if untreated. Reanalysis, grouping patients into treated vs untreated, as suggested by Milne in an earlier rapid response (2) initially sounds attractive, possibly even offering an estimate of the benefits of treatment in 'real life'. Unfortunately, this would also be flawed as the groups would not be comparable. As an example, the untreated group is likely to include 'self -selected' individuals who choose not to attend for screening and are less likely to adhere to advice (and hence likely to have a higher risk, as suggested in the original paper, where the risk estimates for individuals with missing data were almost double those with complete data). Another factor contributing to the apparent difference in estimate of risk is the duration of follow up. Extrapolating from the data provided, the mean follow up in the QRISK derivation dataset was around 5 years, and this appears to have simply been extrapolated to give a 10 year risk assessment. Duration of follow up was however embedded in the Framingham risk analysis, and so the equations allow for calculation of risk over different time periods ranging from 4 to 12 years (3). Using the Framingham risk calculator in this way reveals that the risk during a 10 year period is most heavily influenced by a higher risk in the latter years, hence follow up of a shorter duration (as in QRISK) would be expected to show a lower risk. In view of the longer follow up and less confounding from the use of effective risk-reducing therapies, Framingham has to remain the risk engine of choice. 1. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study 10.1136/bmj.39261.471806.55. BMJ 2007:bmj.39261.471806.55. 2. Milne EMG. Does QRISK describe a partially treated population? BMJ 2007 (Rapid response to reference (1)) 3. Anderson KM, Odell PM, Wilson PW, Kannel WB. Cardiovascular disease risk profiles. Am Heart J 1991; 121: 293-8. Competing interests: None declared |
|||
|
|
|||
|
Marie Therese Cooney, Research Fellow in Cardiology Adelaide Meath Hospital, Tallaght, Dublin 24, Ireland, Alexandra Dudina, Ian Graham - Prof of Preventive Cardiology and Consultant Cardiologist
Send response to journal:
|
Hippisley-Cox et al recently described a new risk estimation system, QRISK, which was derived from and tested on a UK population [1]. They demonstrated superior performance in terms of the estimation of absolute risk when the system was tested in the validation cohort, compared to Framingham [2] and ASSIGN [3]. Given, as the authors acknowledge, that the validation cohort came from the same population as the derivation cohort, this is perhaps not surprising. We acknowledge the obvious merits of the project, particularly the very substantial amount of recent data used, the diversity of those included in terms of ethnicity and social background and the demonstration of the important risk gradient associated with decreasing social class. However, we have noted several methodological limitations which may have caused inconsistencies and may lead to difficulty when applying the risk function to other populations. Firstly, QRISK is not based on a cohort study with randomly selected participants; rather the participants in QRISK were patients of selected GPs, so its representativeness of the total UK population is uncertain. Although mean risk factor levels correlate well with those of the British population in 1995, the correlation with mortality and event rates for the general British population at this time was not described. The risk factors were measured at varying times relative to the date of study entry and not specifically for the purposes of the study. No information is given on standardisation of measurement of the risk factors. This is in contrast to SCORE [4], and many other risk estimation systems. Importantly, a very substantial number of the participants do not have information on risk factor levels, for example, total cholesterol (TC)/high density lipoprotein (HDL) cholesterol ratio was not available in approximately 70% of the sample. Secondly, the hazard ratio for total cholesterol/high density lipoprotein cholesterol is completely inconsistent with numerous previous studies. It is reported as 1.001 for each increase of 1 unit in men and women, with confidence intervals including 1 in each. According to these figures, the hazard ratio associated with a substantial increase in TC/HDL from 2 to 10 would be 1.005, which is trivial regardless of the baseline risk. It is possible that part of the reason for this trivial hazard ratio is the method used to deal with the problem of missing data. For approximately 70% of participants the TC/HDL levels were calculated by a computerised regression technique based on values of other predictor variables, which was repeated 5 times. This may explain why certain other variables for example, body mass index and level of social deprivation have assumed a greater importance relative to cholesterol measures. Lastly, the authors indicated a significant difference between the survival rate in those with measured TC/HDL ratio and the survival rate in those with missing data. (10 year observed risks of CV event were 4.0% and 4.9% in women and men respectively who had a recorded TC/HDL ratio compared to 7.9% and 10.9% for women and men who did not have a recorded TC/HDL ratio.) No information was given on the usage of statins in the population. Presumably GPs are more likely to measure total cholesterol/HDL ratio in patients being considered for or already receiving statin therapy. Therefore, given that the ~70% of participants without a measured TC/HDL ratio were assigned the age-band and gender specific mean value from the derivation dataset , the values of TC/HDL assigned to those without a recorded value in the validation dataset were definitely those of a much lower risk group and were possibly on-treatment values. It may be reasonable that in a population in which ~70% of participants have been given such an artificial value for TC/HDL a risk estimation system in which changes in TC/HDL ratio cause a negligible effect on risk would function well. However, how well such a system would estimate risk in people with measured values of TC/HDL ratio is unknown. The results of the second validation may provide some useful information. 1. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and valiation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. BMJ 2007: 10.1136/bmj.39261.471806.55. 2. Anderson K, Wilson PW, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation 1991;83:356-362. 3. Woodward M, Brindle P, Tunstall-Pedoe H, for the SIGN group on risk estimation. Adding social deprivation and family history to cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;93:172–176. 4. Conroy R, Pyörälä K, Fitzgerald AP, Sans S, Menotti A, De Backer G, De Bacquer D, Ducimetière P, Jousilahti P, Keil U, Njølstad I, Oganov RG, Thomsen T, Tunstall-Pedoe H, Tverdal A, Wedel H, Whincup P, Wilhelmsen L, Graham IM; SCORE project group. Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur heart J 2003;24:987-1003. Competing interests: None declared |
|||
|
|
|||
|
Sarah H Wild, Senior Lecturer University of Edinburgh, Marlene Stewart, Jacqueline Price, F.Gerald Fowkes, Gordon Murray.
Send response to journal:
|
The effect of socio-economic deprivation on cardiovascular risk has not been explained by differences in the prevalence of risk factors traditionally used to estimate a person’s risk of developing cardiovascular disease1. Two new risk scores, ASSIGN and QRISK, have recently been developed to include measures of deprivation 2 3 in an attempt to improve the prediction of risk using such scoring systems. One potential factor contributing to the value of risk scores including deprivation is that current smoking status (used in conventional scoring systems) may not adequately reflect differences in lifelong smoking habit in people of different socio-economic status. We have therefore explored whether levels of lifelong smoking differ by socio-economic status among current smokers. Analyses were based on data for 3350 people aged 50-75 years participating in a primary prevention trial of aspirin for asymptomatic atherosclerosis who had no history of clinical cardiovascular disease but who were at moderately increased risk of cardiovascular disease with an ankle brachial index (ratio of systolic blood pressure in the ankle to that in the arm) of 0.95 or less4. Socio-economic status was assigned on the basis of postcode and was defined by quintiles of the Scottish Index of Multiple Deprivation (SIMD. SIMD is derived from Census data on income, employment, housing, health, education and skills/training/access to services and telecommunications 5. We used linear regression modelling to investigate whether lifelong exposure to smoking as measured by pack-years of smoking (number of cigarettes smoked per year multiplied by years of smoking/20) varied by deprivation quintile after adjusting for age. Data for pack years of smoking were square-root transformed in the analysis. The mean age of the population studied was 62 years; 22% were in the least deprived national quintile for deprivation and 27% were in the most deprived quintile. Prevalence of current smoking increased from 13% for the least deprived quintile to 36% to the most deprived quintile. Among current smokers age-adjusted pack years of smoking increased from 30 to 34 (p=0.029 for trend) for the least to the most deprived quintile. These findings suggest that current smoking status does not adequately reflect differences in lifelong exposure to cigarette smoking associated with deprivation. Risk scores such as the Framingham risk score 6 that only include current risk factor status may therefore not identify the cardiovascular risk associated with deprivation appropriately 3;7;8. As a consequence making treatment decisions based on scores that do not include deprivation may result in a widening of health inequalities. Our results support the importance of validating and refining scores that include deprivation to address inequalities in risk of cardiovascular disease. Reference List 1. van Rossum CT, Shipley MJ, van de MH, Grobbee DE, Marmot MG. Employment grade differences in cause specific mortality. A 25 year follow up of civil servants from the first Whitehall study. J.Epidemiol.Community Health 2000;54:178-84. 2. Woodward M, Brindle P, Tunstall-Pedoe H. Adding social deprivation and family history to cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;93:172-6. 3. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. BMJ 2007. 4. Heald CL, Fowkes FG, Murray GD, Price JF. Risk of mortality and cardiovascular disease associated with the ankle-brachial index: Systematic review. Atherosclerosis 2006;189:61-9. 5. Office of the Chief Statistician. Scottish Index of Multiple Deprivation 2004: Summary Technical Report. 2004. Edinburgh, Scottish Executive. 6. Anderson KM, Odell PM, Wilson PW, Kannel WB. Cardiovascular disease risk profiles. Am.Heart J. 1991;121:293-8. 7. Brindle PM, McConnachie A, Upton MN, Hart CL, Davey SG, Watt GC. The accuracy of the Framingham risk-score in different socioeconomic groups: a prospective study. Br.J.Gen.Pract. 2005;55:838-45. 8. Tunstall-Pedoe H,.Woodward M. By neglecting deprivation, cardiovascular risk scoring will exacerbate social gradients in disease. Heart 2006;92:307-10. Competing interests: None declared |
|||
|
|
|||
|
Gordon A A Ferns, Professor Postgraduate Medical School, University of Surrey, Guildford, Surrey GU2 7WG
Send response to journal:
|
Although the use of ROC analysis has been criticised as a means for assessing the utility of models that predict future risk, or stratify individuals into risk categories(1), it is clear that applying QRISK (2) does not greatly improve the discrimination of individuals who will or will not develop cardiovascular disease. The marginal improvement in the ROC statistic still leaves more than 20% of patients in the wrong category. There appears to be an urgent need for more direct measures of vascular health, perhaps coronary calcium score (3), before initiating life-long treatment with statins, or other agents unnecessarily. 1. Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 2007;115:928-35. 2. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and vlidation of QRISK, a new cardiovascular diseases risk score for the United Kingdom: a prospective open cohort study. BMJ 2007;335: 136-141. 3. Greenland P, LaBree L, Azen SP, Doherty TM, Detrano RC. Coronary artery calcium score combined with Framingham score for risk prediction in asymptomatic individuals. Journal of the American Medical Association 2004;291:210-5. Competing interests: None declared |
|||
|
|
|||
|
Ben D Ewald, general practitoner and epidemiologist Newcastle NSW 2300
Send response to journal:
|
Hippisley-Cox has recast cardiovascular risk assessment but has ignored one of the strongest and most easily measured items. In Australian data the waist-hip ratio is a much stronger predictor of cardiovascular mortality or total mortality than BMI. This was shown in 11 year survival follow up of the 1989 risk factor prevalence survey. The hazard ratio for a 1 standard deviation increase in BMI was 1.23 while the same measure for waist-hip ratio in men was 1.63. It will be fascinating when girth or waist-hip ratio is included in cardiovascular prediction models to see which other risk factors have reduced hazard ratios. Reference Timothy A Welborn, Satvinder S Dhaliwal and Stanley A Bennett. Waist–hip ratio is the dominant risk factor predicting cardiovascular death in Australia. MJA 2003; 179 (11/12): 580-585 Competing interests: None declared |
|||
|
|
|||
|
Julia Hippisley-Cox, professor of clinical epidemiology University of Nottingham, Julia Hippisley-Cox, Carol Coupland, Yana Vinogradova, John Robson, Margaret May, Peter Brindle
Send response to journal:
|
Dear Editor Re: Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. Thank you for giving us the opportunity to respond to letters on QRISK. We have tried to address all the substantive points in full and also present additional analyses as requested. 1.1 Additional analyses on statin usage and cholesterol Milne, Lewis, Simpson and Cooney all raise concerns about the inclusion of patients on statins in our analysis and ask whether this might have biased our results. Peto, Cooney and Tunstall-Pedoe all highlight the low hazard ratio for cholesterol ratio. Several people also contacted us individually to request additional analyses to investigate this in more detail. We found the discussions helpful and have now conducted those analyses. Of the 1,283,174 patients in the original derivation cohort, only 14,005 patients (1.09%) had already been prescribed statins at baseline. After removing these patients from the cohort, there were a further 118,009 (9.3%) patients who were prescribed statins during the ten year study period. However the vast majority of these patients started treatment towards the end of the observation period. For example, of the total 8,182,404 person years of observation, only 256,387 person years (3.13%) were exposed to statin treatment. This means that more than 96% of our study observation time was uncontaminated by statin usage and so our population was largely statin naïve. Table 1 shows the adjusted hazard ratios for model A (the additional models to appear in the BMJ print version or from julia.hippisley- cox@nottingham.ac.uk) • Model A includes all patients apart from those prescribed statins at baseline. Patients who began statins during follow up are included. This is our preferred model. • Model B is the same as model A except it adjusts for statin usage as a time varying covariate. The issue with this approach is how such a model with time varying covariates could be applied to a real prospective population. • Model C is the same as model A except the outcome is coronary heart disease • Model D is the same as model A except that the outcome is stroke/TIA Table: Comparison of hazard ratios for cardiovascular disease using four different models. All models exclude statin users at baseline.
We have improved our implementation of multiple imputation for missing data. Our revised model for imputation includes more variables in addition to those originally included. We have added binary variables for diagnosis of hypertension and incident diabetes and continuous variables for the number of prescriptions for aspirin, statins and antihypertensive treatments for each patient during the study period. When the revised imputed datasets were used in the main analysis, the estimated hazard ratios for the cholesterol ratio term in males and females increased to more biologically plausible values given our combined outcome of cardiovascular disease. We think this addresses the concerns raised by Peto, Cooney and Tunstall-Pedoe and improves the analysis. Multiple imputation has rarely been used in derivation of risk scores and has a distinct advantage in terms of using a larger and more representative sample than the analysis of only those subjects who had complete records of all data including serum cholesterol. The adjusted hazard ratios for the ratio of cholesterol/HDL in the complete case analysis was 1.20 (95% CI 1.17 to 1.22) in females and 1.25 (95% CI 1.23 to 1.27) in males. The analysis using only subjects with complete records is available from the authors on request. There was little difference between models for any of the hazard ratios for cholesterol/HDL ratio for males or females which suggests that use of statins has not significantly biased our hazard ratios. Models D shows the hazard ratios for coronary heart disease as an outcome and demonstrates that these are broadly similar to those obtained in other studies for prediction of coronary heart disease mortality1 including Framingham. The calibration statistics and the predictions of those at risk by age, sex and deprivation were very similar to those in our original analysis (results available from the authors). Overall, the exposure to statin was low (3% of the person years) and the results from each of the different models are very similar suggesting that this is not an important source of bias in our analysis. Model A is our preferred final model which will be used in our risk calculator http://www.qrisk.org 1.2 Discrimination In their own research MacLeod et al. found that the addition of material deprivation didn’t improve discrimination of Framingham and consider the ROC curve areas for QRISK and Framingham are similar. Romanens and Ferns make the same point. However, the deficiencies of the ROC statistic are well known2 which is why we presented the D statistic which is more appropriate for use with survival data3. This does demonstrate a significant improvement for QRISK compared with the other scores. The performance of a risk prediction equation will always be specific to contexts of time, place and means and definitions of data collection and outcomes. Our aim is to provide a means to equitably implement a cardiovascular disease prevention programme in the context of contemporary UK practice, rather than to elucidate the aetiology of cardiovascular disease. QRISK gives us a means for deriving, implementing and if necessary updating an ‘in-house’ equation based on general practitioners contemporary data which is at least as good if not better than Framingham in it’s discrimination and better calibrated to identify major cardiovascular events in the United Kingdom. It will also address problems with Framingham which under- estimates risk in people from deprived backgrounds4 5, those on anti-hypertensive treatment and people with positive family histories. 1.3 Equity and over-prediction Lewis has misunderstood – we modelled Townsend scores as a continuous variable in the risk score and only presented the results by fifth for ease of interpretation. We used area based deprivation scores derived at output area (125 households) since there is no generally accepted and reliably recorded individual measure of deprivation. MacLeod et al. consider that over-prediction by Framingham is not a problem because the inclusion of more people is likely to result in more benefit than harm. The bar for intervention has been fixed nationally at a 20% 10 year cardiovascular disease risk by the statin Health Technology Assessment http://guidance.nice.org.uk/TA94#documents . This threshold effectively sets the additional national resource (workload, advice, treatment) that will be allocated to people at increased risk. The issue becomes how best to allocate this resource on an equitable basis taking account of health needs – in this case indicated by cardiovascular risk – which is determined by age, sex, blood pressure and social deprivation amongst other things. We agree all new technologies run the risk of exacerbating relative inequalities, but this is not a reason for not seeking to reduce this by our social interventions. We disagree with Macleod et al who consider that simply including more people, including socially deprived men, will reduce inequality. Equity is a relative not an absolute issue. In the most deprived compared to the least deprived fifth of social deprivation, there is a three-fold increase in the proportion of women identified at high risk and a 30% excess among men. In the most affluent fifth, QRISK indicates that Framingham overestimates risk by a factor of 2. These are substantial and important differences particularly for women and they will be to some extent ameliorated by a prevention programme driven by QRISK but are likely to be further exacerbated if Framingham is retained. QRISK ensures a more equitable allocation of the national resource than Framingham. QRISK prioritises the more socially deprived, particularly women, older ages and people with on antihypertensive treatment more equitably than Framingham. 1.4 Home advantage We agree with Macleod et al that QRISK has a home advantage but one which is likely to be shared by other EMIS practices which together cover 60% of the practices in England and Wales. If QRISK performs as well in a similar database from another major computer system (THIN) it will demonstrate generalisability to most of the national population. MacLeod et al suggest that THIN is also home territory, it is, but only in the sense that primary care is the very territory that a cardiovascular disease risk score is meant to be used in. 1.5 Scotland We included ASSIGN in our analysis as its use of social deprivation in risk estimation led us to consider whether it was suitable for use in England and Wales. The reason why ASSIGN may overestimate risk in England and Wales is because the Scottish ASSIGN cohort represents a higher risk population. It had a mean serum cholesterol of 6.2mmols/l which is higher than the mean of 5.4mmols/l in the Health Survey for England and Wales (1998) and 5.5mmols/l in the Health Survey for Scotland (1998) and 5.7mmols/l in the QResearch dataset. Similarly the proportion of smokers was much higher in ASSIGN (42.4% of men) compared to 34% and 30% respectively in the Health Surveys for England and Scotland and 28% for QResearch. ASSIGN may well be representative of the Scottish Heart Health Extended Cohort population – a cohort recruited in the 1980’s because Scotland had particularly high cardiovascular mortality rates. However it may not represent the whole population of modern-day Scotland since most of the participants came from urban areas of North Glasgow and Edinburgh, with baseline measurements collected between 1986-2001 compared with 1995 to 2007 in our study. Acknowledgements We thank Professors Richard Peto and Patrick Royston for advice on additional analyses presented here and their interpretation. References 1. Diverse Populations Collaborative Group. Prediction of mortality from coronary heart disease among diverse populations: is there a common predictive function? Heart 2002;88:222-228. 2. Cook N. Use and Misuse of the Receiver Operating Characteristic Curve in Risk Prediction. Circulation 2007;115:928-935. 3. Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Statistics in Medicine 2004;23:723-748. 4. Brindle P, McConnachie A, Upton M, Hart C, Davey-Smith G, Watt G. The accuracy of the Framingham risk-score in different socioeconomic groups: a prospective study. British Journal of General Practice 2005;55(520):838- 45. 5. Tunstall-Pedoe HW, M for the SIGN group on risk estimation. By neglecting deprivation, cardiovascular risk scoring will exacerbate social gradients in disease. Heart 2006;92:307-310. Competing interests: The authors of the original paper |
|||
|
|
|||
|
John B Carlin, Professor & Director, Clinical Epidemiology & Biostatistics Unit Murdoch Children's Research Institute & University of Melbourne, Jonathan A. C. Sterne, Ian R. White, John B. Carlin, Patrick Royston, Michael G. Kenward, Angela M. Wood and James R. Carpenter
Send response to journal:
|
As other correspondents have noted1, the published results describing the QRISK prediction model for cardiovascular risk2 indicated that cardiovascular risk is unrelated to cholesterol level (coded as the ratio of total to HDL cholesterol). This result cast doubt on the validity of the reported analyses. The analysis was performed using the method of multiple imputation to allow for missing data. We believe that flawed application of the method caused the anomalous findings on cholesterol. The authors have subsequently clarified that when their analysis was performed in the usual way by restricting to individuals with complete information (no missing data) there was a clear association between cholesterol and cardiovascular risk. Furthermore, a similar result appears to have been obtained after using a revised, improved, imputation procedure3.
Briefly, multiple imputation aims to allow for the uncertainty about
missing data by creating multiple copies of the dataset under analysis
in which the missing values are replaced by imputed values drawn using
a statistical prediction based on the observed data. Standard
statistical analyses are performed on each of the imputed datasets and
the results are combined in an appropriate way to obtain final
conclusions. Valid inferences are obtained because the method accounts
for the distribution of the missing data, under the technical
assumption that the data are There is clear potential for multiple imputation to improve the validity of medical research. However, the validity of results depends on the imputation modelling being done carefully and appropriately, and it is not possible to automate the procedure. A number of pitfalls can arise. In particular, when an analysis explores the association between one or more predictors and an outcome, but some of the predictors have missing values, the outcome carries information about the missing values of the predictors, and this information must be used in the imputation procedure5. For example, when imputing missing cholesterol values, individuals who go on to develop cardiovascular disease should have larger imputed cholesterol values, on average, than those who remain disease-free. Failure to allow for the cardiovascular outcome in imputing the missing cholesterol values would falsely weaken the association between cholesterol and cardiovascular disease, as appears to have happened in the QRISK analysis. In a separate longer article that we have submitted for publication, we provide a brief non-technical review of the reasons that missing data may lead to bias in epidemiological and clinical research, and discuss the circumstances in which multiple imputation may reduce such biases. We suggest guidelines for the conduct and reporting of analyses based on multiple imputation, which aim to enhance the benefits that it potentially offers while reducing the risk of errors in its application.
Competing interests: None declared |
|||
|
|
|||
|
Richard J McManus, Clinical Senior Lecturer Department of Primary Care and General Practice, University of Birmingham, Edgbaston, Birmingham B15, Dr Jonathan Mant
Send response to journal:
|
Dear Sir We congratulate Julia Hippisley-Cox and colleagues on this important and innovative article and have followed with interest the subsequent responses regarding the derivation and validation of a new risk score – QRISK.(1) One aspect of QRISK that has not been discussed is the use of such a score in daily practice. There are broadly two ways in which scores such as this or the Framingham equation are likely to be used – firstly population screening using a practice clinical system to trawl for individuals likely to be at high risk of cardiovascular disease and secondly on an individual basis where a practitioner and his or her patient decide on the most appropriate form of primary prevention. In the former case, any risk estimation tool needs to take into account likely missing data and provide a reasonable assessment of risk prior to calling up an individual for the latter consultation and accurately assessing risk with a full dataset of risk factors from history, examination and investigation. QRISK has been developed and validated on a dataset that provides good evidence of its utility in the case of the kind of data routinely available in EMIS practices and we agree with Macleod et al that this is likely to be similar in practices using other clinical systems.(2) However, in the case of an individual for whom full information is available about current risk factors (blood pressure, cholesterol/HDL ratio etc) we do not believe that the superiority of QRISK has been proven. This is exemplified by the issue of cholesterol/HDL ratio for which data were available on only approximately 30% of the cohort and which despite recalculation of the model may still be undervalued as a risk factor by QRISK.(3)(4) Furthermore, QRISK used the first measurement of each risk factor available after the age of 35 for each individual but it seems more appropriate to use currently measured risk factors to assess current risk in the setting of a face to face consultation. Whilst QRISK appears an excellent tool for a strategy of population level CVD screening using routinely collected clinical data, it is likely that other scores derived from complete measurements (such as Framingham or ASSIGN) will be more appropriate for use with individual patients. Yours faithfully Dr Richard McManus & Dr Jonathan Mant References 1 Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Derivation and validation of QRISK, a new cardiovascular diseases risk score for the United Kingdom: a prospective open cohort study. BMJ 2007;335: 136-141. 2 Macleod J, Metcalfe C, Davey Smith G. QRISK and Health Inequalities. BMJ Rapid response. http://www.bmj.com/cgi/eletters/335/7611/136 [accessed 21 8 2007] 3 Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, May M, Brindle P. Authors Response. BMJ rapid response. http://www.bmj.com/cgi/eletters/335/7611/136 [accessed 21 8 2007] 4 Ingelsson E, Schaefer EJ, Contois JH, McNamara JR, Sullivan L, Keyes MJ, Pencina MJ, Schoonmaker C, Wilson PW, D'Agostino RB, Vasan RS. Clinical utility of different lipid measures for prediction of coronary heart disease in men and women. JAMA. 2007 Aug 15;298(7):776-85. Competing interests: None declared |
|||
|
|
|||
|
Guy Wilkinson, General Practitioner Glossop, Derbyshire, SK13 8PS, UK
Send response to journal:
|
With the Department of Health moving UK Primary Care into population CVD risk stratification, any tool that allows practitioners to focus resource to need is valued. Modelling of potential savings if this tool were adopted for this process would be welcomed. On a practical level within the consultation, the tool would be most valuably integrated within the practice based clinical system, so that deprivation indices down to a postcode or individual level could be used rather than practice aggregates. Competing interests: None declared |
|||
|
|
|||
|
Peter M Brindle, General practitioner and R&D lead for the Avon Primary Care Research Collaborative, Bristol PCT. The Wellspring Surgery, Bristol. BS5 9QY
Send response to journal:
|
Readers may be interested to know that the performance of QRISK compared with Framingham has been evaluated in the THIN database - a sample of about 1.1 million primary care patients from practices that use the In Practice Software system. The paper has just been published in Heart and can be found at: http://heart.bmj.com/cgi/rapidpdf/hrt.2007.134890v2? Competing interests: I am a co-author of QRISK |
|||
|
|
|||
|
G Alastair Cooke, Consultant Cardiologist Doncaster Royal Infirmary, South Yorkshire DN2 5LT
Send response to journal:
|
Qrisk in common with the other risk calculators calculates an individuals 10 year risk of cardiovascular disease. In all these algorithms the biggest determinate of the risk is the age of the patient. The 10 year risk of a 35 year old patient is much lower than the 10 year risk of a 75 year old patient with otherwise identical risk factors. NICE says that patient with a 10 year risk of above 20 % should get a statin. This means that large numbers of elderly patients get a statin but very few patients in their 30s or 40s get a statin. Is this logical? We generally give statin therapy for life. While, we do not know what happens if you take a statin for decades, it seems likely that the longer you take the statin the more you will benefit. The recently published long term follow up of the WOSCOPS study supports the long term benefits of taking a a statin. Conversely, if a patient is never going to develop CHD then they are not going to benefit from taking a statin. We need to know what an individual’s lifetime risk of CHD is, not their 10 year risk. We then need to know what happens, if those patients at high risk, take statins for several decades. Competing interests: I have received lecture fees and educational sponsorship from several statin manufacturers |
|||
