Evidence Based Functional Medicine: Grading Evidence and Communicating Uncertainty

An earlier article introduced IIFM’s evidence labels and the principles of grading evidence. This article goes deeper, into the concepts that most often cause misinterpretation of research and the skills needed to apply population evidence to an individual patient. These skills matter in all of medicine; in integrated practice, where claims about interventions vary widely in their support, they matter even more.

Study designs and their weaknesses

Surrogate markers and real outcomes

Many studies measure surrogate markers, such as cholesterol, blood pressure, HbA1c or inflammatory markers, because real outcomes such as heart attacks and deaths take longer to observe. Surrogates are useful when changes in them reliably predict changes in outcomes. Sometimes they do not. The drug torcetrapib raised HDL cholesterol substantially, yet increased mortality in a large trial. Lowering homocysteine with B vitamins did not reduce heart attacks. Antioxidant supplements lowered some markers of oxidative stress without improving survival. Evidence based on surrogate changes alone should be interpreted with caution.

Absolute and relative risk

Relative risk reductions often sound impressive: “a 25 percent reduction in heart attacks”. But the meaning depends on the baseline risk. If the risk of a heart attack over 10 years is 8 percent, a 25 percent relative reduction lowers it to 6 percent, an absolute reduction of 2 percentage points. For someone whose baseline risk is 2 percent, the same relative reduction lowers it to 1.5 percent, an absolute reduction of half a point.

The number needed to treat, the reciprocal of the absolute risk reduction, expresses how many people need an intervention for one to benefit. In the first example it is 50; in the second, 200. Presenting absolute risks and numbers needed to treat helps patients and practitioners judge whether an intervention is worthwhile, particularly when it has costs or side effects.

Confidence intervals and significance

A result is often reported with a confidence interval, the range of values consistent with the data. A wide interval means uncertainty. A statistically significant result may be too small to matter clinically; a non significant result may reflect a small study rather than no effect. The clinical question is always whether the plausible range of effects is large enough to matter to patients.

Heterogeneity of treatment effect

An average effect in a trial conceals variation between individuals. Some people benefit more, some less, and some not at all. Subgroup analyses can suggest who benefits most but are prone to false positive findings, especially when many subgroups are tested. Where patient characteristics clearly change baseline risk, such as age or disease severity, they can guide the application of trial results even when relative effects are similar.

Biases in the evidence base

• Publication bias: studies with positive results are more likely to be published, inflating apparent effects. Funnel plots and trial registries help detect it.
• Funding bias: industry funded studies, including in nutrition and supplements, are more likely to report favourable conclusions.
• Outcome switching: reporting outcomes other than those planned.
• Small study effects: small trials tend to report larger effects than large trials of the same intervention.
• Spin: abstracts and press releases may overstate findings.

From populations to a person

Applying evidence to an individual involves several steps.

1. Is the evidence relevant? Were the study participants similar to this patient?
2. What is this patient’s baseline risk? This determines absolute benefit.
3. What are the likely benefits and harms in absolute terms?
4. What does the patient value? The same numbers may lead to different decisions for different people.
5. Can the effect be tested in this person? For symptomatic treatments with rapid effects, a structured trial may help.

N of 1 approaches

For interventions aimed at symptoms that fluctuate and respond quickly, such as pain, sleep or digestive symptoms, formal N of 1 trials alternate an intervention and a comparison in a planned sequence, ideally blinded, with outcomes measured consistently. A blinded N of 1 design was used in a trial of statin related muscle symptoms, showing that most symptoms occurred on placebo too. Informal versions of this logic, changing one thing at a time, measuring before and after and stopping what does not help, are valuable in everyday practice.

Communicating uncertainty

• Use natural frequencies: “out of 100 people like you, about 8 would have a heart attack in 10 years without treatment and about 6 with it” is clearer than percentages and relative risks.
• Use visual aids, such as icon arrays, where helpful.
• Be explicit about confidence: distinguish strong, moderate and weak evidence, as the IIFM labels do.
• Avoid false precision: round numbers and ranges are more honest than exact figures when evidence is uncertain.
• Acknowledge unknowns: “we don’t yet know” is a legitimate and trust building answer.

A worked example

A 52 year old woman asks whether she should take a supplement that a study reported reduced cardiovascular events by 20 percent. The practitioner looks at the evidence. The study was a single trial in people with established heart disease, funded by the manufacturer, using a surrogate composite outcome, with a wide confidence interval. The woman has no heart disease and a 10 year cardiovascular risk of about 4 percent. Even if the relative effect applied to her, the absolute reduction would be less than one percentage point, a number needed to treat of over 100, and the evidence may not apply to people without heart disease at all. Meanwhile, her blood pressure is raised and she is inactive, both with stronger evidence for intervention. She decides to focus on those instead, with the supplement question revisited if better evidence emerges.

Evidence, applied

Evidence based practice is not about reciting study results. It is about understanding their strengths and limits, translating them into terms that fit an individual patient, and communicating uncertainty honestly. These skills allow integrated practitioners to evaluate interventions from every system on equal terms.