15.4 Critical Appraisal of Nutrition Research & Evidence

15.4 Critical Appraisal of Nutrition Research & Evidence

Clinical Core: Evidence-based nutrition support is the integration of the best available research, clinical expertise, and patient values—not the mechanical adoption of an abstract conclusion. Appraisal asks three questions: Are the methods valid? How large and precise is the effect? Does it apply to this patient, product, dose, route, and setting?

Frame the Question Before Searching

Use PICO: patient or population, intervention, comparator, and outcome. Add time and setting when they matter. “Is fish oil useful?” is too broad. “In mechanically ventilated adults receiving EN during the first ICU week, does a specific omega-3–enriched formula versus an isocaloric standard high-protein formula reduce ventilator-free days or infection without increasing intolerance?” is testable.

Predefine patient-important outcomes. A change in a laboratory surrogate may not improve survival, function, infection, quality of life, or time at home. Distinguish efficacy under controlled conditions from effectiveness in routine care.

Match Design to the Question

Randomization balances known and unknown prognostic factors on average, but it does not repair high attrition, unequal co-interventions, unblinded subjective outcomes, or selective reporting. Check whether the primary analysis followed intention-to-treat principles and whether the trial was registered before outcomes were known.

Observational studies may be the best feasible evidence for uncommon harms or long-term outcomes. Adjustment reduces measured confounding but cannot eliminate unmeasured confounding or reverse causation. A large sample does not convert a biased comparison into a randomized one.

Effect Size and Precision

Suppose infection occurs in 20% of the control group and 12% of the intervention group.

$\text{Relative risk} = \frac{0.12}{0.20} = 0.60$

$\text{Absolute risk reduction} = 0.20 - 0.12 = 0.08\;(8\%)$

$\text{Number needed to treat} = \frac{1}{0.08} = 12.5 \rightarrow 13$

The relative risk reduction is 40%, but the absolute effect—8 fewer infections per 100 treated—communicates baseline risk and is often more useful for decisions. Number needed to treat is rounded up to the next whole patient and should be presented with the time horizon and confidence interval.

A confidence interval describes the range of effects compatible with the data under the model. For a ratio measure, an interval crossing 1 includes no difference; for a mean difference, crossing 0 includes no difference. A narrow interval around a trivial effect may be precise but clinically unimportant. A wide interval may include meaningful benefit and harm even when the p value exceeds 0.05. A p value is not the probability that the null hypothesis is true and does not measure effect magnitude.

Diagnostic Evidence

Sensitivity is the proportion of people with the target condition who test positive; specificity is the proportion without it who test negative. Positive and negative predictive values depend strongly on prevalence. A screening tool validated in an ICU may perform differently in an ambulatory clinic because disease spectrum and prevalence change.

Likelihood ratios combine sensitivity and specificity and can update pretest odds to post-test odds. Appraise whether all participants received the same independent reference standard and whether test readers were blinded to the reference result. A nutrition assessment tool should not be declared diagnostic merely because it correlates with another imperfect tool.

Systematic Reviews and Heterogeneity

A systematic review should publish a reproducible question, eligibility criteria, comprehensive search, duplicate study selection, risk-of-bias assessment, and transparent synthesis. Meta-analysis is appropriate only when populations, interventions, comparators, and outcomes are sufficiently compatible.

Statistical heterogeneity measures such as I² describe inconsistency but do not explain it. Examine dose, timing, formula composition, baseline nutrition risk, route, setting, and outcome definitions. A pooled estimate can be misleading when clinically different products are treated as one intervention. Funnel-plot asymmetry may suggest publication bias but also has other explanations and is unreliable with few studies.

Certainty and Recommendation Strength

GRADE considers risk of bias, inconsistency, indirectness, imprecision, and publication bias, with possible upgrading of observational evidence for factors such as a large effect under defined circumstances. Certainty reflects confidence in an effect estimate. Recommendation strength additionally incorporates the balance of benefits and harms, patient values, resources, feasibility, acceptability, and equity. Therefore, certainty and recommendation strength are related but not identical.

Applying Evidence to Nutrition Support

Before applying a result, compare the study with the patient:

• Was the patient population similar in age, disease, severity, organ function, and baseline nutrition risk?
• Is the exact product, nutrient dose, route, timing, comparator, and background care available?
• Are outcomes patient-important and measured over a relevant duration?
• Do benefits outweigh access, infection, metabolic, aspiration, medication, cost, and burden risks?
• Can the intervention be delivered reliably in this setting, and does it worsen inequity or caregiver burden?
• Do the patient’s goals and preferences support the tradeoff?

Document uncertainty. If evidence is indirect or rapidly evolving, a monitored time-limited trial with predefined success and stop criteria may be more defensible than an indefinite intervention. Reassess when new evidence, safety alerts, products, or patient goals change.