Re: Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies
CCBYNCOpen access
Rapid response to:
Research
Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies
Re: Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies
Response
We appreciate the thoughtful comments by Timothy Frayling and colleagues on our study entitled ‘Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies’ published in BMJ recently (1). We would like to take the opportunity to respond to their comments.
1. Although residual confounding cannot be excluded in observational analyses of gene-environment interactions, we adjusted for a wide range of potential confounders, and these adjustments did not appreciably alter the results. Also, since we analyzed changes in diet quality vs. changes in weight, this mimics an intervention study and reduces the chances of confounding by socioeconomic status (SES), which is unlikely to change substantially over a relatively short period of time. Because the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) cohorts are more homogeneous in terms of occupation, education, and other SES factors, confounding is much less compared to the general population cohorts. Nevertheless, we performed a sensitivity analysis with additional adjustment for SES indicators available in the NHS, including father’s occupation as the indicator of childhood socioeconomic position and spouse’s educational level as the indicator of adult socioeconomic position (these variables are not available in the HPFS) (2), and our main findings remained virtually the same. Of note, the analysis by Tyrrell et al. was a cross-sectional study, which is highly prone to reverse causation and confounding (3). We have previously shown that this type of analysis can be misleading for studies of dietary factors and body weight (4).
2. Although randomized controlled trials (RCT) are often considered a gold standard, they are not immune to confounding and serious bias (5), especially in dietary intervention trials where compliance is typically poor and dropout rates are high. The relationship between vitamin E supplement use and coronary heart disease (CHD) (6) was cited as an example to dismiss findings from observational studies. However, no one has claimed that a definitive conclusion can be drawn from one study or even one type of studies. Also, it should be kept in mind that long-term observational studies and RCTs of vitamin supplementations may be testing different exposures in terms of dosages, formulation (“natural” vs synthetic), different degrees of adherence, different durations of exposure, and concurrent use of other supplements or medications. A widely cited example of discordance between observational studies and RCT’s is hormone replacement therapy (HRT) and CHD risk. Data from observational studies suggested that HRT reduces risk of CHD, but initial results from the Women's Health Initiative (WHI) indicated an increased risk. Although initially this discrepancy was thought to result from confounding in observational analyses, further analyses determined both analyses are valid and the discrepancies were likely due to the timing of HRT initiation and different age groups of participants in observational studies and WHI (7,8). Different formulations of HRT also had different effects on health outcomes.
3. We agree that ‘main genetic effects are usually protected from confounding and reverse causation, whereas gene by environment interactions are not protected in this way’. This indeed emphasizes the importance of analyzing gene by environment interactions in well-designed, prospective cohorts. In addition, the change in diet vs. change in weight analysis we employed in our study provides additional strengths to minimize potential reverse causation (9). It is worth noting that a dietary pattern very similar to that we used effectively reduced weight in a 6-year randomized trial (10), indicating the validity of our assessment of diet and its relation to the weight outcome.
4. Tyrrell et al. (3) employed a “negative control” method to test the robustness of gene-environment interaction findings. Although this method is interesting, there are still debates on selection of valid negative controls, which is based on assumptions that are often difficult to verify. Sun-protection use was used as negative control variable in Tyrrell et al.’s study. However, sun-protection use is likely to be correlated with vitamin D levels, which may play a role in adiposity (11). Sunscreen use may be correlated with outdoor physical activity levels as well.
5. While we agree with the point about the importance of biological mechanisms in interpreting gene-environment interactions, providing biological evidence typically goes beyond the scope of epidemiologic research. In practice, both hypothesis-driven and hypothesis-free investigations contribute significantly to the discovery of novel factors related to human health. We also agree that at present there is no sufficient evidence to specifically target individuals for improvement in dietary quality on the basis of genetic factors because, even if there is evidence of gene-environment interaction for weight change, most individuals are likely to benefit from such a diet.
6. In our analyses, we carefully controlled for lifestyle characteristics such as physical activity and alcohol intake, following the methods reported before (9). We did not analyze BMI on the transformed scale because the distribution of BMI was not skewed in our cohorts, and it is more complicated to interpret the results from of analysis BMI changes on the transformed scale. Indeed, very few epidemiologic studies analyzed transformed BMI in adults. In our study, we analyzed the dietary factors as both categorical and continuous variables, and the results were consistent.
References
1. Wang T, Heianza Y, Sun D, Huang T, Ma W, Rimm EB, Manson JE, Hu FB, Willett WC, Qi L. Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies. BMJ. 2018 Jan 10;360:j5644
2. Lidfeldt J, Li TY, Hu FB, Manson JE, Kawachi I. A prospective study of childhood and adult socioeconomic status and incidence of type 2 diabetes in women. Am J Epidemiol. 2007;165:882-9.
3. Tyrrell J, Wood AR, Ames RM, et al. Gene-obesogenic environment interactions in the UK Biobank study. Int J Epidemiol. 2017;46:559-575.
4. Smith JD, Hou T, Hu FB, et al. A Comparison of Different Methods for Evaluating Diet, Physical Activity, and Long-Term Weight Gain in 3 Prospective Cohort Studies. J Nutr. 2015;145:2527-2534.
5. Frieden TR. Evidence for Health Decision Making - Beyond Randomized, Controlled Trials. N Engl J Med. 2017 Aug 3;377(5):465-475.
6. Rimm EB, Stampfer MJ, Ascherio A, Giovannucci E, Colditz GA, Willett WC. Vitamin E consumption and the risk of coronary heart disease in men. N Engl J Med. 1993;328:1450-1456.
7. Manson JE, Aragaki AK, Rossouw JE, et al. Menopausal Hormone Therapy and Long-term All-Cause and Cause-Specific Mortality: The Women's Health Initiative Randomized Trials. JAMA. 2017;318:927-938.
8. Bhupathiraju SN, Grodstein F, Rosner BA, et al. Hormone Therapy Use and Risk of Chronic Disease in the Nurses' Health Study: A Comparative Analysis With the Women's Health Initiative. Am J Epidemiol. 2017;186:696-708.
9. Mozaffarian D, Hao T, Rimm EB, Willett WC, Hu FB. Changes in diet and lifestyle and long-term weight gain in women and men. N Engl J Med. 2011;364:2392-2404.
10. Schwarzfuchs D, Golan R, Shai I. Four-year follow-up after two-year dietary interventions. N Engl J Med. 2012;367:1373-1374.
11. Vanlint S. Vitamin D and obesity. Nutrients. 2013;5:949-56
Rapid Response:
Re: Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies
Response
We appreciate the thoughtful comments by Timothy Frayling and colleagues on our study entitled ‘Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies’ published in BMJ recently (1). We would like to take the opportunity to respond to their comments.
1. Although residual confounding cannot be excluded in observational analyses of gene-environment interactions, we adjusted for a wide range of potential confounders, and these adjustments did not appreciably alter the results. Also, since we analyzed changes in diet quality vs. changes in weight, this mimics an intervention study and reduces the chances of confounding by socioeconomic status (SES), which is unlikely to change substantially over a relatively short period of time. Because the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) cohorts are more homogeneous in terms of occupation, education, and other SES factors, confounding is much less compared to the general population cohorts. Nevertheless, we performed a sensitivity analysis with additional adjustment for SES indicators available in the NHS, including father’s occupation as the indicator of childhood socioeconomic position and spouse’s educational level as the indicator of adult socioeconomic position (these variables are not available in the HPFS) (2), and our main findings remained virtually the same. Of note, the analysis by Tyrrell et al. was a cross-sectional study, which is highly prone to reverse causation and confounding (3). We have previously shown that this type of analysis can be misleading for studies of dietary factors and body weight (4).
2. Although randomized controlled trials (RCT) are often considered a gold standard, they are not immune to confounding and serious bias (5), especially in dietary intervention trials where compliance is typically poor and dropout rates are high. The relationship between vitamin E supplement use and coronary heart disease (CHD) (6) was cited as an example to dismiss findings from observational studies. However, no one has claimed that a definitive conclusion can be drawn from one study or even one type of studies. Also, it should be kept in mind that long-term observational studies and RCTs of vitamin supplementations may be testing different exposures in terms of dosages, formulation (“natural” vs synthetic), different degrees of adherence, different durations of exposure, and concurrent use of other supplements or medications. A widely cited example of discordance between observational studies and RCT’s is hormone replacement therapy (HRT) and CHD risk. Data from observational studies suggested that HRT reduces risk of CHD, but initial results from the Women's Health Initiative (WHI) indicated an increased risk. Although initially this discrepancy was thought to result from confounding in observational analyses, further analyses determined both analyses are valid and the discrepancies were likely due to the timing of HRT initiation and different age groups of participants in observational studies and WHI (7,8). Different formulations of HRT also had different effects on health outcomes.
3. We agree that ‘main genetic effects are usually protected from confounding and reverse causation, whereas gene by environment interactions are not protected in this way’. This indeed emphasizes the importance of analyzing gene by environment interactions in well-designed, prospective cohorts. In addition, the change in diet vs. change in weight analysis we employed in our study provides additional strengths to minimize potential reverse causation (9). It is worth noting that a dietary pattern very similar to that we used effectively reduced weight in a 6-year randomized trial (10), indicating the validity of our assessment of diet and its relation to the weight outcome.
4. Tyrrell et al. (3) employed a “negative control” method to test the robustness of gene-environment interaction findings. Although this method is interesting, there are still debates on selection of valid negative controls, which is based on assumptions that are often difficult to verify. Sun-protection use was used as negative control variable in Tyrrell et al.’s study. However, sun-protection use is likely to be correlated with vitamin D levels, which may play a role in adiposity (11). Sunscreen use may be correlated with outdoor physical activity levels as well.
5. While we agree with the point about the importance of biological mechanisms in interpreting gene-environment interactions, providing biological evidence typically goes beyond the scope of epidemiologic research. In practice, both hypothesis-driven and hypothesis-free investigations contribute significantly to the discovery of novel factors related to human health. We also agree that at present there is no sufficient evidence to specifically target individuals for improvement in dietary quality on the basis of genetic factors because, even if there is evidence of gene-environment interaction for weight change, most individuals are likely to benefit from such a diet.
6. In our analyses, we carefully controlled for lifestyle characteristics such as physical activity and alcohol intake, following the methods reported before (9). We did not analyze BMI on the transformed scale because the distribution of BMI was not skewed in our cohorts, and it is more complicated to interpret the results from of analysis BMI changes on the transformed scale. Indeed, very few epidemiologic studies analyzed transformed BMI in adults. In our study, we analyzed the dietary factors as both categorical and continuous variables, and the results were consistent.
References
1. Wang T, Heianza Y, Sun D, Huang T, Ma W, Rimm EB, Manson JE, Hu FB, Willett WC, Qi L. Improving adherence to healthy dietary patterns, genetic risk, and long term weight gain: gene-diet interaction analysis in two prospective cohort studies. BMJ. 2018 Jan 10;360:j5644
2. Lidfeldt J, Li TY, Hu FB, Manson JE, Kawachi I. A prospective study of childhood and adult socioeconomic status and incidence of type 2 diabetes in women. Am J Epidemiol. 2007;165:882-9.
3. Tyrrell J, Wood AR, Ames RM, et al. Gene-obesogenic environment interactions in the UK Biobank study. Int J Epidemiol. 2017;46:559-575.
4. Smith JD, Hou T, Hu FB, et al. A Comparison of Different Methods for Evaluating Diet, Physical Activity, and Long-Term Weight Gain in 3 Prospective Cohort Studies. J Nutr. 2015;145:2527-2534.
5. Frieden TR. Evidence for Health Decision Making - Beyond Randomized, Controlled Trials. N Engl J Med. 2017 Aug 3;377(5):465-475.
6. Rimm EB, Stampfer MJ, Ascherio A, Giovannucci E, Colditz GA, Willett WC. Vitamin E consumption and the risk of coronary heart disease in men. N Engl J Med. 1993;328:1450-1456.
7. Manson JE, Aragaki AK, Rossouw JE, et al. Menopausal Hormone Therapy and Long-term All-Cause and Cause-Specific Mortality: The Women's Health Initiative Randomized Trials. JAMA. 2017;318:927-938.
8. Bhupathiraju SN, Grodstein F, Rosner BA, et al. Hormone Therapy Use and Risk of Chronic Disease in the Nurses' Health Study: A Comparative Analysis With the Women's Health Initiative. Am J Epidemiol. 2017;186:696-708.
9. Mozaffarian D, Hao T, Rimm EB, Willett WC, Hu FB. Changes in diet and lifestyle and long-term weight gain in women and men. N Engl J Med. 2011;364:2392-2404.
10. Schwarzfuchs D, Golan R, Shai I. Four-year follow-up after two-year dietary interventions. N Engl J Med. 2012;367:1373-1374.
11. Vanlint S. Vitamin D and obesity. Nutrients. 2013;5:949-56
Lu Qi, Tiange Wang, Frank Hu, Walter Willett
Competing interests: No competing interests