Observational versus randomised trial evidence

Abstract

Debbie Lawlor and colleagues (May 22, p 1724)1Lawlor DA Davey-Smith G Bruckdorfer KR Kundu D Ebrahim S Those confounded vitamins: what can we learn from the differences between observational versus randomised trial evidence?.Lancet. 2004; 363: 1724-1727Summary Full Text Full Text PDF PubMed Scopus (325) Google Scholar compare findings from observational studies and randomised trials, using the EPIC-Norfolk prospective study on plasma ascorbic acid and mortality2Khaw K-T Bingham S Welch A et al.Relation between plasma ascorbic acid and mortality in men and women in EPIC-Norfolk prospective study: a prospective population study.Lancet. 2001; 357: 657-663Summary Full Text Full Text PDF PubMed Scopus (448) Google Scholar as an example. In this article, we stressed that our results indicated that plasma vitamin C was probably a marker of particular foods, and it was not supplements that were protective. We made this point first in the results section, and several times subsequently in the discussion. There is no contradiction between our findings and our interpretation, and the result of the later randomised trial3Heart Protection Study Collaborative GroupMRC/BHF Heart Protection Study of antioxidant vitamin supplementation in 20536 high-risk individuals: a randomised placebo-controlled trial.Lancet. 2002; 360: 23-33Summary Full Text Full Text PDF PubMed Scopus (1361) Google Scholar cited by Lawlor and colleagues, showing no effect of vitamin C supplements on cardiovascular mortality.Confounding is an issue in all observational studies and we discussed the subject extensively. There might be disagreement with our interpretation that “our findings suggest that an increase in dietary intake of foods rich in ascorbic acid might have benefits for cardiovascular disease and all-cause mortality in men and women and add to the large amount of evidence that lends support to the health benefits of fruit and vegetable intake”. However, alternative explanations for the observed association, such as other confounders, also need to be biologically and quantitatively plausible.Nevertheless, based on the misinterpretation of our findings, Lawlor and colleagues suggest that our findings must be the result of confounding. They present data (their table 1) for women from their study, indicating that plasma vitamin C concentrations in adult life are associated with both early life exposures and exposures in adult life, findings we have also reported previously.4Shohaimi S, Bingham S, Welch A, et al. Occupational social class, educational level and area deprivation independently predict plasma ascorbic acid concentration: a cross-sectional population based study in the Norfolk cohort of the European Prospective Investigation into Cancer (EPIC-Norfolk). Eur J Clin Nutr; published online March 31, 2004, DOI: www.10.1038/sj.ejcn.1601979Google Scholar However, the issue is whether these associations could induce substantial confounding to explain our findings.Lawlor and co-workers would have to show that there were pronounced differences in risk associated with the degree of difference in socioeconomic indicators they describe. As an example, they provide data on adult height, adult leg length, and the adult leg-to-trunk ratio as biomarkers of childhood environment. In their data, there is a 1·3 cm difference in height between individuals in the top quarter and those in the bottom quarter of the plasma vitamin C distribution. As in our cohort, the SD of height among women is about 6 cm. In our data, height is only weakly (negatively) associated with cardiovascular mortality (relative risk 0·8 between the bottom and top fifth of the distribution). The largest confounding risk ratio (RR) that these anthropometric measures could induce for cardiovascular mortality lies between 0·95 and 1·0.Similar quantitative considerations apply to the other variables shown. In their data on indicators of childhood socioeconomic status, 91·6% of the lowest vitamin C quartile and 82·1% of the highest quartile were manual social class in childhood. Even if non-manual social class in childhood were 100% protective against later cardiovascular mortality, the confounding risk ratio between the highest and lowest quartiles would be 0·90 (RR 82·1/91·6). For realistic values for the protective effect of early life non-manual social class, the confounding RR would be about 0·95. Our results displayed a relative risk for cardiovascular mortality, comparing the highest quintile of plasma vitamin C with the lowest, of 0·29 in men and 0·41 in women. There is a wide gap between the amount of reduced risk observed and the degree of confounding that could be reasonably ascribed to the variables Lawlor and colleagues propose. They seem to have assumed erroneously that if two variables are associated, they must be capable of substantial mutual confounding. However, they need to assess the potential magnitude of effect of the confounders they cite as essential and provide quantitative justification for the claim that “A more likely explanation is that associations between antioxidants and disease in observational studies are confounded by social and behavioural factors across the life course”. Dietary differences could just as well explain some of the socioeconomic inequalities in health.Randomised trials are often not feasible and we have to rely on observational evidence. All study designs, whether observational or interventional, have limitations, and we agree with John Concato and Ralph Horwitz' view in the accompanying Commentary5Concato J Horwitz RI Beyond randomised versus observational studies.Lancet. 2004; 363: 1660-1661Summary Full Text Full Text PDF PubMed Scopus (115) Google Scholar to Lawlor and colleagues' Viewpoint that there is a need for findings to be assessed in the context of existing evidence and to ensure responsible interpretation of all studies. Debbie Lawlor and colleagues (May 22, p 1724)1Lawlor DA Davey-Smith G Bruckdorfer KR Kundu D Ebrahim S Those confounded vitamins: what can we learn from the differences between observational versus randomised trial evidence?.Lancet. 2004; 363: 1724-1727Summary Full Text Full Text PDF PubMed Scopus (325) Google Scholar compare findings from observational studies and randomised trials, using the EPIC-Norfolk prospective study on plasma ascorbic acid and mortality2Khaw K-T Bingham S Welch A et al.Relation between plasma ascorbic acid and mortality in men and women in EPIC-Norfolk prospective study: a prospective population study.Lancet. 2001; 357: 657-663Summary Full Text Full Text PDF PubMed Scopus (448) Google Scholar as an example. In this article, we stressed that our results indicated that plasma vitamin C was probably a marker of particular foods, and it was not supplements that were protective. We made this point first in the results section, and several times subsequently in the discussion. There is no contradiction between our findings and our interpretation, and the result of the later randomised trial3Heart Protection Study Collaborative GroupMRC/BHF Heart Protection Study of antioxidant vitamin supplementation in 20536 high-risk individuals: a randomised placebo-controlled trial.Lancet. 2002; 360: 23-33Summary Full Text Full Text PDF PubMed Scopus (1361) Google Scholar cited by Lawlor and colleagues, showing no effect of vitamin C supplements on cardiovascular mortality. Confounding is an issue in all observational studies and we discussed the subject extensively. There might be disagreement with our interpretation that “our findings suggest that an increase in dietary intake of foods rich in ascorbic acid might have benefits for cardiovascular disease and all-cause mortality in men and women and add to the large amount of evidence that lends support to the health benefits of fruit and vegetable intake”. However, alternative explanations for the observed association, such as other confounders, also need to be biologically and quantitatively plausible. Nevertheless, based on the misinterpretation of our findings, Lawlor and colleagues suggest that our findings must be the result of confounding. They present data (their table 1) for women from their study, indicating that plasma vitamin C concentrations in adult life are associated with both early life exposures and exposures in adult life, findings we have also reported previously.4Shohaimi S, Bingham S, Welch A, et al. Occupational social class, educational level and area deprivation independently predict plasma ascorbic acid concentration: a cross-sectional population based study in the Norfolk cohort of the European Prospective Investigation into Cancer (EPIC-Norfolk). Eur J Clin Nutr; published online March 31, 2004, DOI: www.10.1038/sj.ejcn.1601979Google Scholar However, the issue is whether these associations could induce substantial confounding to explain our findings. Lawlor and co-workers would have to show that there were pronounced differences in risk associated with the degree of difference in socioeconomic indicators they describe. As an example, they provide data on adult height, adult leg length, and the adult leg-to-trunk ratio as biomarkers of childhood environment. In their data, there is a 1·3 cm difference in height between individuals in the top quarter and those in the bottom quarter of the plasma vitamin C distribution. As in our cohort, the SD of height among women is about 6 cm. In our data, height is only weakly (negatively) associated with cardiovascular mortality (relative risk 0·8 between the bottom and top fifth of the distribution). The largest confounding risk ratio (RR) that these anthropometric measures could induce for cardiovascular mortality lies between 0·95 and 1·0. Similar quantitative considerations apply to the other variables shown. In their data on indicators of childhood socioeconomic status, 91·6% of the lowest vitamin C quartile and 82·1% of the highest quartile were manual social class in childhood. Even if non-manual social class in childhood were 100% protective against later cardiovascular mortality, the confounding risk ratio between the highest and lowest quartiles would be 0·90 (RR 82·1/91·6). For realistic values for the protective effect of early life non-manual social class, the confounding RR would be about 0·95. Our results displayed a relative risk for cardiovascular mortality, comparing the highest quintile of plasma vitamin C with the lowest, of 0·29 in men and 0·41 in women. There is a wide gap between the amount of reduced risk observed and the degree of confounding that could be reasonably ascribed to the variables Lawlor and colleagues propose. They seem to have assumed erroneously that if two variables are associated, they must be capable of substantial mutual confounding. However, they need to assess the potential magnitude of effect of the confounders they cite as essential and provide quantitative justification for the claim that “A more likely explanation is that associations between antioxidants and disease in observational studies are confounded by social and behavioural factors across the life course”. Dietary differences could just as well explain some of the socioeconomic inequalities in health. Randomised trials are often not feasible and we have to rely on observational evidence. All study designs, whether observational or interventional, have limitations, and we agree with John Concato and Ralph Horwitz' view in the accompanying Commentary5Concato J Horwitz RI Beyond randomised versus observational studies.Lancet. 2004; 363: 1660-1661Summary Full Text Full Text PDF PubMed Scopus (115) Google Scholar to Lawlor and colleagues' Viewpoint that there is a need for findings to be assessed in the context of existing evidence and to ensure responsible interpretation of all studies. Observational versus randomised trial evidenceAuthors' reply Full-Text PDF