Introduction
Why we need more RCTs, and why RCTs are difficult to conduct in the nutrition field?
RCTs | NRSs | |
---|---|---|
Theoretical | ||
Certainty of the evidence regarding causality | Higher | Lower |
Confounding | Unlikely for large RCTs | Adjustment for known and measured confounders possible; residual confounding likely |
Levels of exposure | Few; often relatively high differences in intervention groups | Broad range, possibility of stratifying by exposure level |
Follow-up time of study | Short or limited | Long |
Empirical | ||
Number of participants | Usually < 1000 | Some > 10,000 |
Representativeness for general population | Often limited | Generally good |
Outcome measures | Often risk factor, occasionally morbidity/ mortality | Usually morbidity/ mortality |
Why NRSs provide sometimes potentially less trustworthy results, and how we can identify plausible results?
What tools should be used to address risk of bias?
Bias in RCTs | Action that protect against biases | Bias in NRSs | ROBINS-I | Newcastle–Ottawa Scale (max. 9 points) | |||||
---|---|---|---|---|---|---|---|---|---|
(Cochrane RoB tool) | Domains | Rating (n = studies, %) | Domains | Rating (n = studies, %) | |||||
Bias arising from randomization process | Random sequence generation | Confounding | Bias due to Confounding | Moderate RoB | 20/20 (100%) | Comparability of cohorts on the basis of the design or analysis | High quality | 20/20 (100%) | |
Allocation concealment | Low quality | 0/20 (0%) | |||||||
Selection | Bias in selection of participants into the study | Low RoB | 16/20 (80%) | Representativeness of the exposed cohort | High quality | 13/20 (65%) | |||
Low quality | 7/20 (35%) | ||||||||
Serious RoB | 4/20 (20%) | Selection of the non-exposed cohort | High quality | 20/20 (100%) | |||||
Low quality | 0/20 (0%) | ||||||||
Demonstration that outcome of interest was not present at start of study | High quality | 20/20 (100%) | |||||||
Low quality | 0/20 (0%) | ||||||||
Misclassification | Bias in classification of exposure | Low RoB | 16/20 (80%) | Ascertainment of exposure | High quality | 18/20 (90%) | |||
Low quality | 2/20 (10%) | ||||||||
Moderate RoB | 2/20 (10%) | ||||||||
Serious RoB | 2/20 (10%) | ||||||||
Bias due to deviations from intended intervention | Blinding of participants and personnel | Performance | Bias due to deviations from intended exposure | Low RoB | 1/20 (5%) | N/A | |||
Moderate RoB | 19/20 (95%) | ||||||||
Bias due to missing outcome data | Complete outcome data | Attrition | Bias due to missing outcome data | Low RoB | 15/20 (75%) | Adequacy of follow-up of cohorts | High quality | 19/20 (95%) | |
Moderate RoB | 5/20 (25%) | Low quality | 1/20 (5%) | ||||||
Bias in measurement of the outcome | Blinding of outcome assessment | Detection | Bias in measurement of the outcome | Low RoB | 19/20 (95%) | Assessment of outcome | High quality | 20/20 (100%) | |
Serious RoB | 1/20 (5%) | ||||||||
Low quality | 0/20 (0%) | ||||||||
Was follow-up long enough for outcomes to occur? | High quality | 20/20 (100%) | |||||||
Low quality | 0/20 (0%) | ||||||||
Bias in the selection of the reported results | Avoid selective reporting | Reporting | Bias in the selection of the reported results | Low RoB | 14/20 (70%) | N/A | |||
Moderate RoB | 3/20 (15%) | ||||||||
Serious RoB | 3/20 (15%) | ||||||||
Overall RoB judgementa | Low RoB | 0/20 (0%) | Overall study qualityb | High quality (7–9) | 20/20 (100%) | ||||
Moderate RoB | 13/20 (65%) | Lower quality (0–6) | 0/20 (0%) | ||||||
Serious RoB | 7/20 (35%) |
Examples of critical risk of bias in certain domains in a dietary Cochrane Review
Why is risk of bias so important to evaluate the credibility of study results?
Why can rating the certainty of evidence improve the trustworthiness of findings?
Macro-level | Micro-level | GRADE | 3rd WCRF/AICR Reporta | USDA DGA 2015–2020b | |||
---|---|---|---|---|---|---|---|
Y/N/U | Explanation | Y/N/U | Explanation | Y/N/U | Explanation | ||
A priori assumption | Body of evidence (RCTs and cohort studies) begins as high certainty | N | Using NOS | Y | N/A | Y | N/A |
Y | Using ROBINS-I | ||||||
Certainty criteria | Risk of bias (aka study quality) | Y | Cochrane RoB tool for RCTs; NOS or ROBINS-I for NRS | U | No tool; some criteria reported: confounding, measurement error and selection bias (but assessment unclear) | Y | Tool adapted from Cochrane: selection bias, performance bias, detection bias, attrition bias |
Inconsistency | Y | Similarity of point estimates, extent of overlap of 95% CI, statistical criteria including tests of heterogeneity, and I2 | Y | I2 | U | Description vague: e.g., consistent in direction and size of effect or degree of association and statistical significance with very minor exceptions | |
Indirectness | Y | According to PICO criteria | N | N/A | Y | According to PICO criteria | |
Imprecision | Y | Examination of 95% CI, OIS | N | N/A | U | Large sample size | |
Publication bias | Y | Funnel plot | N | N/A | N | N/A | |
Large effect | Y | RR: > 2 or < 0.5 (large effect) RR: > 5 or < 0.2 (very large effect) | Y | RR: > 2 (large effect) | U | Description vague: e.g., clinically meaningful effect size | |
Dose–response | Y | Linear, non-linear | Y | Linear, non-linear | N | N/A | |
Plausible residual confounding | Y | Plausible confounders would decrease an effect, or would create a spurious effect when results suggest no effect | N | N/A | N | N/A | |
Transparent Summary | Summary of Findings Table | Y | Certainty rating (and criteria) for each outcome and the estimate of effect | N | N/A | N | N/A |
Overall rating | Certainty of evidence | High, Moderate, Low, Very low | Convincing, Probable, Limited-suggestive, Limited-no conclusion, Substantial effect on risk unlikely | Strong, Moderate, Limited, Grade not assignable |
What are current approaches for making dietary recommendations?
Macro-level | Micro-level | GRADE | 3rd WCRF/AICR Reporta | USDA DGA 2015–2020b | |||
---|---|---|---|---|---|---|---|
Y/N/U | Explanation | Y/N/U | Explanation | Y/N/U | Explanation | ||
Conflict of interest from committee members | Intellectual and financial conflicts | Y | Should be reported | N | Not reported | U | Per Federal Advisory Committee Act rules, Advisory Committee members were thoroughly vetted for conflicts of interest before they were appointed to their positions and were required to submit a financial disclosure form annually |
EtDs framework | Problem | Y | Problem priority definition? | U | N/A | U | N/A |
Benefit and harms | Y | How substantial are beneficial/harmful effects? | U | N/A | N | N/A | |
Certainty of evidence | Y | see Table 2 | Y | see Table 2 | Y | see Table 2 | |
Values | Y | How much people value the main outcomes? | N | N/A | N | N/A | |
Balance of effects | Y | Balance between desirable and undesirable effects? | N | Panel has sometimes not made recommendations despite strong evidence; because of potentially adverse effects (dairy and prostate cancer) on one cancer despite evidence of protection for another (e.g., dairy and colorectal cancer) | N | N/A | |
Resources required | Y | How large are the costs? | N | N/A | N | N/A | |
Cost effectiveness | Y | Cost effectiveness of the intervention? | N | N/A | N | N/A | |
Equity | Y | Impact on health equity? | N | N/A | N | N/A | |
Acceptability | Y | Option acceptable to key stakeholders? | N | N/A | N | N/A | |
Feasibility | Y | Implementation feasible? | U | Sometimes not feasible | N | N/A | |
Recommendation | Strength of recommendation | Y | Strong, weak (conditional, discretional, or qualified) | Y | Recommendations were made only when the CUP Panels judged the evidence sufficiently strong (when exposure was convincingly/probably or causally linked to cancer risk) | U | The grades used for conclusion statements also fall into one of four categories (as the certainty of evidence evaluation): strong, moderate, limited, and grade not assignable |