Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles.
We thank Drs Mukamal and Lazo for their shrewd commentary (1) on our study of alcohol consumption and initial presentation of 12 cardiovascular diseases (2). In their editorial they note some of the major strengths of our study, such as our use of “big data” and that we are the first to use clinically recorded alcohol consumption data. The latter of which is particularly important as it is not always clear how findings obtained under controlled conditions, such as alcohol consumption recorded using structured surveys, translate to real life, whereas in our study, the data used is directly actionable as it is exactly that which a clinician (and therefore services) would use in day-to-day practice when deciding what treatment and/or advice to provide for a patient.
We also agree with several of the limitations Mukamal and Lazo raised, as acknowledged in the discussion section of our manuscript. However, there are claims, such as that our findings will be difficult to export to settings outside of the UK that we respectfully disagree with and justify our reason for doing so below.
In principle our observations are likely to generalise to the United States and other western countries. All clinicians are expected to record a patient’s alcohol history and update it periodically (3). This is taught from the earliest days in medical school (4). Any and all electronic health record systems would have a means of recording this information (through standard classification coding schemes or via free-text). Our use of Read (5) codes should not be considered parochial as they map onto SNOMED-CT codes (6) and have greater granularity than the international statistical classification of diseases, 10th version. Others have shown that provided that a standard framework is adopted by researchers, the classification system used is largely irrelevant (7).
Mukamal and Lazo also commented on missing data in our study. Missing data in the context of electronic health records is subtly different to that encountered in investigator led studies. We had data available on alcohol consumption for 57% of the eligible sample. There are no large population based cohorts with anywhere close to this coverage. For example, the UK Biobank, a considerably smaller population of just 500k participants, represents only 5.5% of those invited (8). Other studies that have been celebrated for their contribution to the evidence base of alcohol and chronic disease, such as the Nurses’ Health Study II (9), with an initial response rate of 24% (10), also fall noticeably short of what we achieved. Selection biases (11) are a major concern in epidemiological studies and our use of electronic health records actually reduces this bias.
As outlined in our manuscript, we agree that there is potential for misclassification error when recording and classifying clinical alcohol data. The overwhelming majority of participants were classified using data that would be considered orthodox (e.g. categories based on amount of alcohol consumed). Additional codes were incorporated into our algorithm to be as inclusive as possible. The potential of introducing bias through doing this is likely to be minimal given the low frequency of less standard codes underlying drinking category classifications. For instance, taking the example Mukamal and Lazo had apprehensions about, only 0.3% of patients were classified as moderate drinkers using codes related to their GP finding alcohol on their breath. Conscious of imprecise measurement, we were careful to discuss our findings in general terms and not comment on specific levels of consumption associated with a higher or lower risk. Furthermore, as our findings for aggregate endpoints and clinical traits are concordant with the latest meta-analyses of observational studies, we do not consider bias introduced by misclassification errors to be markedly greater in electronic health records.
We also found their conclusion that our study does not offer a materially new view of the association between alcohol consumption and cardiovascular disease somewhat perplexing. Firstly, as outlined in the introduction (and in detail in the supplementary material), we are aware that smaller studies exist for alcohol and several of the cardiovascular diseases we investigated. We clearly noted instances where our findings corroborated existing evidence as well as stated the novel contributions of our study in the discussion section. We also noted that most prior studies had in some manner combined different types of non-drinker and this is a major sticking point in the debate as to whether the protective effects of moderate drinking are genuine, and as such one of the primary aims of our study was to correct for this. In some ways our work represents a “spring cleaning” of the evidence base for alcohol consumption and cardiovascular diseases.
But it is too simplistic to conclude that is all we offered. As Mukamal and Lazo acknowledge, we also examined the association between drinking categories and some cardiovascular phenotypes for the first time. We were also the first study to focus on the initial lifetime presentation of multiple cardiovascular diseases within the same sample. Much of the existing literature has ignored competing risks for other cardiovascular diseases (e.g. studies focussed on myocardial infarction have not accounted for intercurrent diseases such as angina) which can lead to biased estimates of disease risk (12). Building further upon examining several cardiovascular diseases being a major strength of our study; we found that heavy drinkers had a greater risk of first presenting with seven of the 12 cardiovascular diseases we examined compared to moderate drinkers. However, if one were to solely focus on the findings for aggregate cardiovascular endpoints this would have been subtly masked. Heavy drinkers seemed relatively no worse off than moderate drinkers but this appears to be driven by divergent disease associations (figure). This is not a phenomenon commonly discussed in the existing literature.
Figure - Multivariable adjusted hazard ratios for 12 cardiovascular diseases in clinically recorded heavy drinkers compared with moderate drinkers, sorted by effect size (adapted from Bell et al. BMJ 2017 DOI: 10.1136/bmj.j909). CHD = coronary heart disease; CVD = cardiovascular disease; SCD = sudden coronary death.
However, as we outline in the manuscript, such heterogeneity is perhaps to be expected. Observationally, the metabolic signature of alcohol consumption appears extremely complex (13), consisting of both adverse and favourable perturbations in multiple cardiometabolic traits. Of course, a perennial concern in any observational study is residual confounding but there is evidence that alcohol is causally associated with a variety of cardiovascular biomarkers (14,15), including non-linear associations (16,17). While many of these traits have been discounted as irrelevant to the aetiology of coronary heart disease (18–20) some are considered causal risk factors (21). Given that even conventional cardiovascular risk factors with large effect sizes, such as blood pressure, have been demonstrated not to have consistently strong associations with the occurrence of all types of cardiovascular disease (22) further highlights the importance of looking beyond aggregated endpoints and towards specific diseases.
Authors and affiliations
Steven Bell,1 Genetic Epidemiologist, Harry Hemingway,2 Professor of Clinical Epidemiology
1 Department of Public Health and Primary Care, University of Cambridge, Strangeways Research Laboratory, Cambridge, CB1 8RN, United Kingdom 2 Farr Institute of Health Informatics Research (London), University College London, 222 Euston Road, London, NW1 2DA, United Kingdom
Competing interests
All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work..
References
1. Mukamal K, Lazo M. Alcohol and cardiovascular disease. BMJ. 2017 Mar 22;356:j1340.
2. Bell S, Daskalopoulou M, Rapsomaniki E, George J, Britton A, Bobak M, et al. Association between clinically recorded alcohol consumption and initial presentation of 12 cardiovascular diseases: population based cohort study using linked health records. BMJ. 2017 Mar 22;356:j909.
3. Moyer VA, on behalf of the U.S. Preventive Services Task Force. Screening and behavioral counseling interventions in primary care to reduce alcohol misuse: U.S. preventive services task force recommendation statement. Ann Intern Med. 2013 Aug 6;159(3):210–8.
4. Frank E, Elon L, Naimi T, Brewer R. Alcohol consumption and alcohol counselling behaviour among US medical students: cohort study. BMJ. 2008 Nov 7;337:a2155.
5. O’Neil M, Payne C, Read J. Read Codes Version 3: a user led terminology. Methods Inf Med. 1995;34(1–2):187.
6. Benson T. The history of the Read codes: the inaugural James Read Memorial Lecture 2011. J Innov Health Inform. 2011;19(3):173–82.
7. Olier I, Springate DA, Ashcroft DM, Doran T, Reeves D, Planner C, et al. Modelling Conditions and Health Care Processes in Electronic Health Records: An Application to Severe Mental Illness with the Clinical Practice Research Datalink. PLOS ONE. 2016 Feb 26;11(2):e0146715.
8. Swanson JM. The UK Biobank and selection bias. The Lancet. 2012;380(9837):110.
9. Mostofsky E, Mukamal KJ, Giovannucci EL, Stampfer MJ, Rimm EB. Key Findings on Alcohol Consumption and a Variety of Health Outcomes From the Nurses’ Health Study. Am J Public Health. 2016 Jul 26;106(9):1586–91.
11. Naimi TS, Stockwell T, Zhao J, Xuan Z, Dangardt F, Saitz R, et al. Selection biases in observational studies affect associations between “moderate” alcohol consumption and mortality. Addiction. 2017;112(2):207–14.
12. Varadhan R, Weiss CO, Segal JB, Wu AW, Scharfstein D, Boyd C. Evaluating Health Outcomes in the Presence of Competing Risks: A Review of Statistical Methods and Clinical Applications. Med Care. 2010;48(6):S96–105.
13. Würtz P, Cook S, Wang Q, Tiainen M, Tynkkynen T, Kangas AJ, et al. Metabolic profiling of alcohol consumption in 9778 young adults. Int J Epidemiol. 2016 Aug 5;45(5):1493–506.
14. Brien SE, Ronksley PE, Turner BJ, Mukamal KJ, Ghali WA. Effect of alcohol consumption on biological markers associated with risk of coronary heart disease: systematic review and meta-analysis of interventional studies. BMJ. 2011;342:d636.
15. Roerecke M, Kaczorowski J, Tobe SW, Gmel G, Hasan OSM, Rehm J. The effect of a reduction in alcohol consumption on blood pressure: a systematic review and meta-analysis. Lancet Public Health. 2017;2(2):e108–20.
16. Silverwood RJ, Holmes MV, Dale CE, Lawlor DA, Whittaker JC, Smith GD, et al. Testing for non-linear causal effects using a binary genotype in a Mendelian randomization study: application to alcohol and cardiovascular traits. Int J Epidemiol. 2014 Dec 1;43(6):1781–90.
17. Vu KN, Ballantyne CM, Hoogeveen RC, Nambi V, Volcik KA, Boerwinkle E, et al. Causal Role of Alcohol Consumption in an Improved Lipid Profile: The Atherosclerosis Risk in Communities (ARIC) Study. PLOS ONE. 2016 Feb 5;11(2):e0148765.
18. Burgess S, Harshfield E. Mendelian randomization to assess causal effects of blood lipids on coronary heart disease: lessons from the past and applications to the future. Curr Opin Endocrinol Diabetes Obes. 2016;23(2):124–30.
19. Keavney B, Danesh J, Parish S, Palmer A, Clark S, Youngman L, et al. Fibrinogen and coronary heart disease: test of causality by “Mendelian randomization.” Int J Epidemiol. 2006 Aug 1;35(4):935–43.
20. Borges MC, Lawlor DA, de Oliveira C, White J, Horta B, Barros AJ. The Role of Adiponectin in Coronary Heart Disease Risk: A Mendelian Randomization Study. Circ Res. 2016 Jun 1;119:491–9.
21. Jansen H, Samani NJ, Schunkert H. Mendelian randomization studies in coronary artery disease. Eur Heart J. 2014 Aug 1;35(29):1917–24.
22. Rapsomaniki E, Timmis A, George J, Pujades-Rodriguez M, Shah AD, Denaxas S, et al. Blood pressure and incidence of twelve cardiovascular diseases: lifetime risks, healthy life-years lost, and age-specific associations in 1·25 million people. The Lancet. 2014;383(9932):1899–911.
Competing interests:
No competing interests
16 August 2017
Steven Bell
Genetic Epidemiologist
Harry Hemingway
University of Cambridge
Department of Public Health and Primary Care, Strangeways Research Laboratory, Cambridge CB1 8RN, United Kingdom
Alcohol and cardiovascular disease: Author’s response
We thank Drs Mukamal and Lazo for their shrewd commentary (1) on our study of alcohol consumption and initial presentation of 12 cardiovascular diseases (2). In their editorial they note some of the major strengths of our study, such as our use of “big data” and that we are the first to use clinically recorded alcohol consumption data. The latter of which is particularly important as it is not always clear how findings obtained under controlled conditions, such as alcohol consumption recorded using structured surveys, translate to real life, whereas in our study, the data used is directly actionable as it is exactly that which a clinician (and therefore services) would use in day-to-day practice when deciding what treatment and/or advice to provide for a patient.
We also agree with several of the limitations Mukamal and Lazo raised, as acknowledged in the discussion section of our manuscript. However, there are claims, such as that our findings will be difficult to export to settings outside of the UK that we respectfully disagree with and justify our reason for doing so below.
In principle our observations are likely to generalise to the United States and other western countries. All clinicians are expected to record a patient’s alcohol history and update it periodically (3). This is taught from the earliest days in medical school (4). Any and all electronic health record systems would have a means of recording this information (through standard classification coding schemes or via free-text). Our use of Read (5) codes should not be considered parochial as they map onto SNOMED-CT codes (6) and have greater granularity than the international statistical classification of diseases, 10th version. Others have shown that provided that a standard framework is adopted by researchers, the classification system used is largely irrelevant (7).
Mukamal and Lazo also commented on missing data in our study. Missing data in the context of electronic health records is subtly different to that encountered in investigator led studies. We had data available on alcohol consumption for 57% of the eligible sample. There are no large population based cohorts with anywhere close to this coverage. For example, the UK Biobank, a considerably smaller population of just 500k participants, represents only 5.5% of those invited (8). Other studies that have been celebrated for their contribution to the evidence base of alcohol and chronic disease, such as the Nurses’ Health Study II (9), with an initial response rate of 24% (10), also fall noticeably short of what we achieved. Selection biases (11) are a major concern in epidemiological studies and our use of electronic health records actually reduces this bias.
As outlined in our manuscript, we agree that there is potential for misclassification error when recording and classifying clinical alcohol data. The overwhelming majority of participants were classified using data that would be considered orthodox (e.g. categories based on amount of alcohol consumed). Additional codes were incorporated into our algorithm to be as inclusive as possible. The potential of introducing bias through doing this is likely to be minimal given the low frequency of less standard codes underlying drinking category classifications. For instance, taking the example Mukamal and Lazo had apprehensions about, only 0.3% of patients were classified as moderate drinkers using codes related to their GP finding alcohol on their breath. Conscious of imprecise measurement, we were careful to discuss our findings in general terms and not comment on specific levels of consumption associated with a higher or lower risk. Furthermore, as our findings for aggregate endpoints and clinical traits are concordant with the latest meta-analyses of observational studies, we do not consider bias introduced by misclassification errors to be markedly greater in electronic health records.
We also found their conclusion that our study does not offer a materially new view of the association between alcohol consumption and cardiovascular disease somewhat perplexing. Firstly, as outlined in the introduction (and in detail in the supplementary material), we are aware that smaller studies exist for alcohol and several of the cardiovascular diseases we investigated. We clearly noted instances where our findings corroborated existing evidence as well as stated the novel contributions of our study in the discussion section. We also noted that most prior studies had in some manner combined different types of non-drinker and this is a major sticking point in the debate as to whether the protective effects of moderate drinking are genuine, and as such one of the primary aims of our study was to correct for this. In some ways our work represents a “spring cleaning” of the evidence base for alcohol consumption and cardiovascular diseases.
But it is too simplistic to conclude that is all we offered. As Mukamal and Lazo acknowledge, we also examined the association between drinking categories and some cardiovascular phenotypes for the first time. We were also the first study to focus on the initial lifetime presentation of multiple cardiovascular diseases within the same sample. Much of the existing literature has ignored competing risks for other cardiovascular diseases (e.g. studies focussed on myocardial infarction have not accounted for intercurrent diseases such as angina) which can lead to biased estimates of disease risk (12). Building further upon examining several cardiovascular diseases being a major strength of our study; we found that heavy drinkers had a greater risk of first presenting with seven of the 12 cardiovascular diseases we examined compared to moderate drinkers. However, if one were to solely focus on the findings for aggregate cardiovascular endpoints this would have been subtly masked. Heavy drinkers seemed relatively no worse off than moderate drinkers but this appears to be driven by divergent disease associations (figure). This is not a phenomenon commonly discussed in the existing literature.
Figure - Multivariable adjusted hazard ratios for 12 cardiovascular diseases in clinically recorded heavy drinkers compared with moderate drinkers, sorted by effect size (adapted from Bell et al. BMJ 2017 DOI: 10.1136/bmj.j909).
CHD = coronary heart disease; CVD = cardiovascular disease; SCD = sudden coronary death.
However, as we outline in the manuscript, such heterogeneity is perhaps to be expected. Observationally, the metabolic signature of alcohol consumption appears extremely complex (13), consisting of both adverse and favourable perturbations in multiple cardiometabolic traits. Of course, a perennial concern in any observational study is residual confounding but there is evidence that alcohol is causally associated with a variety of cardiovascular biomarkers (14,15), including non-linear associations (16,17). While many of these traits have been discounted as irrelevant to the aetiology of coronary heart disease (18–20) some are considered causal risk factors (21). Given that even conventional cardiovascular risk factors with large effect sizes, such as blood pressure, have been demonstrated not to have consistently strong associations with the occurrence of all types of cardiovascular disease (22) further highlights the importance of looking beyond aggregated endpoints and towards specific diseases.
Authors and affiliations
Steven Bell,1 Genetic Epidemiologist, Harry Hemingway,2 Professor of Clinical Epidemiology
1 Department of Public Health and Primary Care, University of Cambridge, Strangeways Research Laboratory, Cambridge, CB1 8RN, United Kingdom
2 Farr Institute of Health Informatics Research (London), University College London, 222 Euston Road, London, NW1 2DA, United Kingdom
Competing interests
All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work..
References
1. Mukamal K, Lazo M. Alcohol and cardiovascular disease. BMJ. 2017 Mar 22;356:j1340.
2. Bell S, Daskalopoulou M, Rapsomaniki E, George J, Britton A, Bobak M, et al. Association between clinically recorded alcohol consumption and initial presentation of 12 cardiovascular diseases: population based cohort study using linked health records. BMJ. 2017 Mar 22;356:j909.
3. Moyer VA, on behalf of the U.S. Preventive Services Task Force. Screening and behavioral counseling interventions in primary care to reduce alcohol misuse: U.S. preventive services task force recommendation statement. Ann Intern Med. 2013 Aug 6;159(3):210–8.
4. Frank E, Elon L, Naimi T, Brewer R. Alcohol consumption and alcohol counselling behaviour among US medical students: cohort study. BMJ. 2008 Nov 7;337:a2155.
5. O’Neil M, Payne C, Read J. Read Codes Version 3: a user led terminology. Methods Inf Med. 1995;34(1–2):187.
6. Benson T. The history of the Read codes: the inaugural James Read Memorial Lecture 2011. J Innov Health Inform. 2011;19(3):173–82.
7. Olier I, Springate DA, Ashcroft DM, Doran T, Reeves D, Planner C, et al. Modelling Conditions and Health Care Processes in Electronic Health Records: An Application to Severe Mental Illness with the Clinical Practice Research Datalink. PLOS ONE. 2016 Feb 26;11(2):e0146715.
8. Swanson JM. The UK Biobank and selection bias. The Lancet. 2012;380(9837):110.
9. Mostofsky E, Mukamal KJ, Giovannucci EL, Stampfer MJ, Rimm EB. Key Findings on Alcohol Consumption and a Variety of Health Outcomes From the Nurses’ Health Study. Am J Public Health. 2016 Jul 26;106(9):1586–91.
10. Nurses’ Health Study. Nurses Health Study: History [Internet]. [cited 2017 Apr 18]. Available from: http://www.nurseshealthstudy.org/about-nhs/history
11. Naimi TS, Stockwell T, Zhao J, Xuan Z, Dangardt F, Saitz R, et al. Selection biases in observational studies affect associations between “moderate” alcohol consumption and mortality. Addiction. 2017;112(2):207–14.
12. Varadhan R, Weiss CO, Segal JB, Wu AW, Scharfstein D, Boyd C. Evaluating Health Outcomes in the Presence of Competing Risks: A Review of Statistical Methods and Clinical Applications. Med Care. 2010;48(6):S96–105.
13. Würtz P, Cook S, Wang Q, Tiainen M, Tynkkynen T, Kangas AJ, et al. Metabolic profiling of alcohol consumption in 9778 young adults. Int J Epidemiol. 2016 Aug 5;45(5):1493–506.
14. Brien SE, Ronksley PE, Turner BJ, Mukamal KJ, Ghali WA. Effect of alcohol consumption on biological markers associated with risk of coronary heart disease: systematic review and meta-analysis of interventional studies. BMJ. 2011;342:d636.
15. Roerecke M, Kaczorowski J, Tobe SW, Gmel G, Hasan OSM, Rehm J. The effect of a reduction in alcohol consumption on blood pressure: a systematic review and meta-analysis. Lancet Public Health. 2017;2(2):e108–20.
16. Silverwood RJ, Holmes MV, Dale CE, Lawlor DA, Whittaker JC, Smith GD, et al. Testing for non-linear causal effects using a binary genotype in a Mendelian randomization study: application to alcohol and cardiovascular traits. Int J Epidemiol. 2014 Dec 1;43(6):1781–90.
17. Vu KN, Ballantyne CM, Hoogeveen RC, Nambi V, Volcik KA, Boerwinkle E, et al. Causal Role of Alcohol Consumption in an Improved Lipid Profile: The Atherosclerosis Risk in Communities (ARIC) Study. PLOS ONE. 2016 Feb 5;11(2):e0148765.
18. Burgess S, Harshfield E. Mendelian randomization to assess causal effects of blood lipids on coronary heart disease: lessons from the past and applications to the future. Curr Opin Endocrinol Diabetes Obes. 2016;23(2):124–30.
19. Keavney B, Danesh J, Parish S, Palmer A, Clark S, Youngman L, et al. Fibrinogen and coronary heart disease: test of causality by “Mendelian randomization.” Int J Epidemiol. 2006 Aug 1;35(4):935–43.
20. Borges MC, Lawlor DA, de Oliveira C, White J, Horta B, Barros AJ. The Role of Adiponectin in Coronary Heart Disease Risk: A Mendelian Randomization Study. Circ Res. 2016 Jun 1;119:491–9.
21. Jansen H, Samani NJ, Schunkert H. Mendelian randomization studies in coronary artery disease. Eur Heart J. 2014 Aug 1;35(29):1917–24.
22. Rapsomaniki E, Timmis A, George J, Pujades-Rodriguez M, Shah AD, Denaxas S, et al. Blood pressure and incidence of twelve cardiovascular diseases: lifetime risks, healthy life-years lost, and age-specific associations in 1·25 million people. The Lancet. 2014;383(9932):1899–911.
Competing interests: No competing interests