36-fold higher estimate of deaths attributable to red meat intake in GBD 2019: is this reliable? – Author's reply

Abstract

Alice Stanton and colleagues1Stanton AV Leroy F Elliott C Mann N Wall P De Smet S 36-fold higher estimate of deaths attributable to red meat intake in GBD 2019: is this reliable?.Lancet. 2022; (published online Feb 25.)https://doi.org/10.1016/S0140-6736(22)00311-7Summary Full Text Full Text PDF PubMed Scopus (2) Google Scholar raise concerns about the large increase in estimated deaths due to red meat intake in the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 20192GBD 2019 Risk Factors CollaboratorsGlobal burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019.Lancet. 2020; 396: 1223-1249Summary Full Text Full Text PDF PubMed Scopus (1221) Google Scholar results compared with the GBD 20173GBD 2017 Risk Factor CollaboratorsGlobal, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017.Lancet. 2018; 392: 1923-1994PubMed Scopus (2081) Google Scholar results. As they note, the reasons for the change were estimation of new risk functions, updated systematic reviews, and a change in the theoretical minimum risk exposure level (TMREL). Ioannidis has raised doubts about the credibility of much of the published literature on diet risk–outcome pairs that are based primarily on observational data.4Ioannidis JPA The challenge of reforming nutritional epidemiologic research.JAMA. 2018; 320: 969-970Crossref PubMed Scopus (172) Google Scholar The GBD estimates reflect uncertainty from the estimated risk functions, uncertainty in risk exposure, and uncertainty in the TMREL. But many commentators have noted that their concern about the diet risk–outcome pairs extends beyond uncertainty intervals and is about the potential for residual confounding in observational studies on diet. These doubts have been echoed by the Independent Advisory Committee for the GBD, an oversight body that reviews the work of the GBD every 6 months. Based on these concerns, we embarked on a major revision to our approach to all risk–outcome pairs beginning in 2017; these revisions influenced the GBD 2019 results but will be more fully reflected in the forthcoming GBD 2020 revisions. Published systematic reviews for some diet risk–outcome pairs are highly contradictory. For example, in the American Journal of Clinical Nutrition, three systematic reviews on the link between nut consumption and cardiovascular disease and diabetes were published at nearly the same time yet with markedly different conclusions.5Afshin A Micha R Khatibzadeh S Mozaffarian D Consumption of nuts and legumes and risk of incident ischemic heart disease, stroke, and diabetes: a systematic review and meta-analysis.Am J Clin Nutr. 2014; 100: 278-288Crossref PubMed Scopus (312) Google Scholar, 6Luo C Zhang Y Ding Y et al.Nut consumption and risk of type 2 diabetes, cardiovascular disease, and all-cause mortality: a systematic review and meta-analysis.Am J Clin Nutr. 2014; 100: 256-269Crossref PubMed Scopus (151) Google Scholar, 7Zhou D Yu H He F et al.Nut consumption in relation to cardiovascular disease risk and type 2 diabetes: a systematic review and meta-analysis of prospective studies.Am J Clin Nutr. 2014; 100: 270-277Crossref PubMed Scopus (92) Google Scholar A set of meta-analyses of red meat intake generated considerable public debate about the bases of the analyses and their interpretation.8Han MA Zeraatkar D Guyatt GH et al.Reduction of red and processed meat intake and cancer mortality and incidence: a systematic review and meta-analysis of cohort studies.Ann Intern Med. 2019; 171: 711-720Crossref PubMed Scopus (76) Google Scholar, 9Vernooij RWM Zeraatkar D Han MA et al.Patterns of red and processed meat consumption and risk for cardiometabolic and cancer outcomes: a systematic review and meta-analysis of cohort studies.Ann Intern Med. 2019; 171: 732-741Crossref PubMed Scopus (65) Google Scholar, 10Zeraatkar D Han MA Guyatt GH et al.Red and processed meat consumption and risk for all-cause mortality and cardiometabolic outcomes: a systematic review and meta-analysis of cohort studies.Ann Intern Med. 2019; 171: 703-710Crossref PubMed Scopus (95) Google Scholar Following standardised guidelines such as the Preferred Reporting Items for Systematic Reviews and Meta-Analyses is not sufficient to generate the same results; there is considerable scope for variation across groups in how studies are included or meta-analysed. In order to improve standardisation and comparability, beginning in GBD 2019, we have undertaken highly standardised systematic reviews, which are documented in GBD 2019.2GBD 2019 Risk Factors CollaboratorsGlobal burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019.Lancet. 2020; 396: 1223-1249Summary Full Text Full Text PDF PubMed Scopus (1221) Google Scholar We have also become increasingly aware that many studies in the diet field assume a log-linear dose–response relationship, but there are many reasons to expect that these relationships are not linear and indeed many examples where the data show they are not. To relax the assumption of linearity, we developed a new form of dose–response meta-regression11Zheng P Barber R Sorensen RJD Murray CJL Aravkin AY Trimmed constrained mixed effects models: formulations and algorithms.J Comput Graph Stat. 2021; 30: 544-556Crossref Scopus (19) Google Scholar that allows the estimation of risk functions that better follow the available data. The new meta-regression methods also address other limitations in existing methodology, including the common problem in diet studies in which relative risks are reported for groups with consumption levels ranging from the lowest to the highest quartile. Ongoing simulation studies under review confirm that these new dose–response meta-regression tools are better able to capture non-linear relationships when present compared with existing methods, and correctly return log-linear relationships when the underlying relationship has that functional form. To aid in interpretability of the strength of the evidence supporting our analyses, we are introducing a five-star rating system in GBD 2020. The idea is to define the risk function closest to the null that is compatible with the available data, inclusive of between-study heterogeneity. This conservative risk function is not used for the GBD estimates of burden but can be used to generate a star rating. Risk functions that are very close to the null receive a lower strength-of-evidence star rating than risk functions that are far from the null. Each of the GBD 2020 risk–outcome pairs will be star-rated with the exception of occupational hazards, which will be published in a future GBD cycle. These major methodological changes have been sequentially implemented in GBD 2019 and the forthcoming GBD 2020, as the work required has been extraordinarily extensive. For GBD 2019, we adopted new systematic reviews and dose–response meta-regression methods, and used these to estimate the TMREL for red meat intake. For GBD 2019, we included estimation of risk functions for ischaemic stroke and haemorrhagic stroke combined because some studies reported on total stroke and some studies reported for specific stroke outcomes. We did this using an indicator variable for studies addressing one stroke outcome versus another. We also included a monotonicity constraint in our analytical model, which aids in estimation of dose–response curves that are biologically plausible (eg, do not reverse direction in some parts of the exposure domain). For GBD 2020, we have undertaken separate analyses for ischaemic stroke and for haemorrhagic stroke. Based on the meta-regression of available studies, there is a clear protective relationship between red meat intake and haemorrhagic stroke, which will be reflected in the GBD 2020 findings. This protective relationship was not identified in the GBD 2019 analysis because of the pooled approach to analysing ischaemic and haemorrhagic stroke data in one meta-regression. The methodological changes in estimating risk functions did not account for all of the increase in risk associated with red meat intake noted by Stanton and colleagues; much of the increase was related to changing the TMREL to zero in GBD 2019. This decision was made to standardise methods across all risk–outcome pairs to eliminate inconsistencies. As Stanton and colleagues note, we adopt the 85th percentile of exposure in the available studies as the TMREL for protective risk factors. In GBD 2017, we set the TMREL for harmful risks to the 15th percentile of exposure based on the available data. But this approach was not consistently applied to other risks. For example, in GBD 2017, for all occupational exposures including injuries, the TMREL was set to zero even though there were no data sources reporting zero occupational injuries. In GBD 2019, to enhance comparability, if the risk functions were increasing from zero exposure, we switched to a TMREL of zero. This approach was applied to several risks, including for residential radon, processed meat, and red meat.2GBD 2019 Risk Factors CollaboratorsGlobal burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019.Lancet. 2020; 396: 1223-1249Summary Full Text Full Text PDF PubMed Scopus (1221) Google Scholar Because our new methods for dose–response risk estimation included integrating over different exposure intervals, we were able to estimate that the risk increased as red meat intake increased from zero. Including the estimated protective effect of red meat on haemorrhagic stroke in GBD 2020 will change the logic and estimates of the TMREL for GBD 2020. We expect that estimates of attributable deaths for red meat will be reduced based on this forthcoming analysis. The star-rating evaluation of the evidence on red meat consumption suggests that once between-study heterogeneity is taken into account, the strength of evidence regarding the relationship between red meat and various outcomes—including ischaemic heart disease—is relatively weak. Stanton and colleagues also claim that detailed citations are not available for the 92 sources used in the analysis of red meat. This is incorrect. Overall, GBD uses hundreds of thousands of sources. To make it tractable to meet the requirements of the Guidelines for Accurate and Transparent Health Estimates Reporting in an efficient manner, we note in the papers that GBD sources are available in the GBD 2019 Data Input Sources Tool, which provides detailed citations. Stanton and colleagues also raise questions about the value of meat as a source of protein, particularly for children. For a governing body developing dietary recommendations, these are important considerations. The role of GBD, however, is not to balance the totality of considerations but rather to follow the science on what level of exposure is associated with the lowest level of risk. Our diet analysis explicitly addresses the effect of red meat consumption in people over the age of 25 years. We estimate zero deaths attributable to red meat below that age. The results over age 25 years have no bearing on dietary recommendations for children. For the GBD 2019 Data Input Sources Tool see http://ghdx.healthdata.org/gbd-2019/data-input-sources For the GBD 2019 Data Input Sources Tool see http://ghdx.healthdata.org/gbd-2019/data-input-sources I declare no competing interests. 36-fold higher estimate of deaths attributable to red meat intake in GBD 2019: is this reliable?We wish to compliment the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) Collaborators for their important contributions to global health metrics over the past 30 years. Their standardised and comprehen-sive estimates of the global burden of diseases, injuries, and risk factors have been used by researchers, govern-ment officials, and non-governmental organisations to make comparisons among populations, to track changes over time, and to monitor progress towards key policy targets, such as the Sustainable Development Goals. Full-Text PDF Global burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019Overall, the record for reducing exposure to harmful risks over the past three decades is poor. Success with reducing smoking and lead exposure through regulatory policy might point the way for a stronger role for public policy on other risks in addition to continued efforts to provide information on risk factor harm to the general public. Full-Text PDF Open Access