Inappropriate use of GRADE
GRADE was developed to evaluate the quality of interventional studies, and all observational studies will initially be graded as low or very low degree of evidence (10). For studies on dietary patterns, activity and smoking, however, blinded randomised controlled trials are rarely possible, or even desirable. Does that mean that we cannot advise against high alcohol consumption, smoking, inactivity, or unhealthy diets?
When evaluating observational studies with GRADE, there are criteria for upgrading the strength of the evidence, such as dose-response effects and strong effect estimates. For nutritional studies an adapted version, NutriGrade, has been recommended, whereby evidence is upgraded for risk differences of 20 % or more (11). Many of the analyses by NutriRecs showed dose-response effects that were strongly significant and far beyond 20 % for moderately increased consumption (1). However, they did not upgrade the evidence, as the authors considered that red and processed meat may have been correlated with other dietary factors that could have confounded the relationship with morbidities and mortality.
The average Norwegian intake of processed meat compared with a diet without processed meat is associated with an increase in the relative risk of early mortality of about 35 %
The authors could nonetheless have investigated this by performing analyses stratified by whether studies adjusted for various dietary factors, but no such subgroup analyses were undertaken. Both measurement errors and changes in meat consumption during the follow-up period can affect the strength of observed associations (12, 13). For example, when using repeated measurements instead of only one baseline measurement and when also correcting for measurement errors, the relative risk for all-cause mortality per portion per day for red and processed meat increased from 1.05 to 1.11 to 1.25 in the Nurses' Health Study and from 1.08 to 1.14 to 1.83 in the Health Professionals Follow-up Study (and for type 2 diabetes from 1.10 to 1.14 to 1.44). To date, relatively few studies have used repeated measurements of diet and published risk estimates corrected for measurement errors. It is therefore likely that the effect estimates reported in most observational studies are conservative estimates of the true underlying effect.
The use of the GRADE criteria in the new meta-analyses has been criticised by several nutrition researchers. Other systems for classifying strength of evidence, which for example are used in the Nordic nutrition recommendations, also take into account biological plausibility, mechanisms and experimental data. When grading the evidence of the effect of red and processed meat on early mortality with the diet-adapted version NutriGrade or the evidence grading developed by the World Cancer Research Fund, which also incorporates experimental data in the evaluation of evidence (3, 14), the evidence is considered moderate to strong.
Most of the studies included in the analyses of red and processed meat and all-cause mortality had low risk of selection bias and scored high on most of the quality criteria by which they were evaluated (1). Studies which scored high on all quality criteria showed clearly that a 50 g per day increase in consumption of processed meat was associated with a 20 % increase in risk of all-cause mortality (12). The studies which the authors judged to have the highest quality came to opposite conclusions to those of the NutriRecs authors. Previous meta-analyses have found that high red and processed meat intake increases the risk of all-cause mortality (3), type 2 diabetes (13), colorectal cancer (15, 16), and cardiovascular disease (17).