A Better Way to Decide Under Uncertainty

Influenza public health and social measures cannot be reduced to one effectiveness number, and that is not a failure of measurement. It is a consequence of what is being measured: different measures act through different pathways, pursue different objectives, are studied in different populations and packages, and carry different burdens. The final judgment is therefore not to find a hidden universal percentage. It is to state a conditional claim clearly enough that people can decide, monitor, and revise it.

Why one number fails—and what remains knowable

A universal effectiveness number would have to make unlike things comparable. It would need to combine delaying importation with reducing community transmission, lowering total infections with flattening a peak, preventing hospital overload with preventing deaths, and epidemiological benefit with economic and social value. The sources do not supply such a conversion, and there is no reason to assume that one exists without substantial additional judgments.

The problem begins with objectives. An early travel measure may be judged by how much time it buys before a virus is seeded. Isolation may be judged by whether it reduces contact with a symptomatic or infected person. Ventilation may be judged by whether it lowers exposure in an indoor environment. School measures may be judged by how they change age-specific mixing while preserving education and care. A community mask recommendation may be intended mainly as source control in crowded indoor settings. These are not different attempts to estimate the same outcome. They are different interventions aimed at different links and decisions.

The outcome then changes what an apparent success means. A measure may slow transmission without preventing every later infection. It may delay a peak and allow hospitals to prepare without reducing the cumulative burden by the same amount. It may reduce infections but not show a proportionate reduction in deaths if severe disease is concentrated in a different group. It may have a modest direct effect while improving the value of a package. The number attached to one outcome cannot be moved to another without evidence or assumptions that must be stated.

Study design adds another boundary. Randomized trials can make comparisons more interpretable, but community PHSM are difficult to isolate and implement under stable conditions. Observational studies show what happened in real populations but may capture several measures, changing behavior and changing risk at once. Simulations can explore early timing, severe scenarios and combinations that cannot be tested directly, but their projections depend on assumptions about transmission, adherence, costs and health-system capacity. GRADE helps express confidence in the evidence; it does not turn these designs into interchangeable measurements.

The influenza mask estimate makes the point precisely. Twelve community randomized trials involving more than 20,000 participants produced a pooled point estimate suggesting a 12% relative reduction in laboratory-confirmed influenza, with a risk ratio of 0.88 and a 95% confidence interval from 0.61 to 1.27. The interval means that the data were compatible with little or no effect in that comparison, so chance could not be excluded. The result supports a modest possible effect under the studied conditions; it does not establish a universal 12% reduction, and it cannot be translated into an absolute number of infections without the relevant baseline risk.

The recommendation categories do not repair this mismatch by hiding it. WHO strongly recommends hand hygiene with moderate-certainty evidence, conditionally recommends mask use in defined circumstances with moderate-certainty evidence, strongly recommends increased ventilation despite low-certainty evidence, and makes more conditional judgments about school measures, isolation mandates and travel restrictions. These categories reflect not only evidence certainty but also benefits and harms, resources, equity, human rights, acceptability and feasibility. Strong does not mean “largest effect,” and conditional does not mean “ineffective.” A recommendation against a broad measure may reflect poor value or limited usefulness in the relevant phase rather than a claim that the measure can never change any outcome.

Context completes the explanation. WHO identifies transmission rate, population density, health-care infrastructure, cultural norms, socioeconomic conditions and inequality as factors that affect implementation and effectiveness. Timing, duration, coverage and adherence are not annotations to an intervention; they help determine what intervention was actually delivered. A policy that cannot be followed because people lack income support or safe housing is not equivalent to the same text implemented with practical support.