FolderUnderstanding and preventing

Read health data: dates, denominators and uncertainties

A health graph describes what has been observed by a given system over a period and territory. Before comparing two numbers, you need to know their definition, denominator and how they were collected.

Topic updated: 27/10/2026

AI-generated illustration, not evidence of a health event. The quality of a conclusion depends on the definition of the data and their limitations, not only on the shape of the graph.

EssentialShort answer

To be retained

The starting points

To interpret a monitoring data, check the source, date of the event and date of publication, case definition, territory, denominator and method changes. An observed increase may reflect increased circulation, but also more testing or reporting.

Deepen ▼

Putting these benchmarks into practice

Comparing two periods, populations or testing volumes requires different precautions. Try three fictional examples and use a reading checklist.

Open the newsletter reading workshop →

01Reference

Identify what is counted

Two tables can use the same word for different realities.

Definition of cases

Suspected, probable, confirmed or consulted cases may meet separate criteria.

Observed population

A sentinel network, hospitals or all laboratories do not cover the same population.

Denominator

A crude number does not allow the same comparison as a rate compared to a defined population.

Time unit

Week of symptoms, testing, reporting or publication can produce staggered curves.

02Reference

Review quality and delays

The most recent figures are often the most likely to be revised.

Complete

Missing data and late reporting systems may change the balance sheet.

Intensity of screening

A change in access to tests or test criteria affects the cases detected.

Revisions

Deleted duplicates, laboratory corrections or new definitions sometimes change past series.

Comparability

Different methods and capacities limit comparisons between territories or periods.

03Reference

Make a proportionate conclusion

A given question is not always enough to explain a cause.

Search trend

Compare several points, the usual season and the complementary indicators.

Distinguish association and cause

A simultaneous evolution of two series does not prove a causal link.

Keep the margin of uncertainty

Report what the method does wrong, and avoid excessive accuracy.

Return to source

See the methodological notes and corrections of the authority that publishes the bulletin.

LandPutting into context

Situations to be considered

Before comparing figures, you need to know what is counted and how the data was produced.

Increased number of cases

Verify whether testing, case definition, population covered or reporting has changed.

Two territories are compared

Look at population-related rates, ages, periods, and quality of surveillance.

A graph shows a recent fall

Search for a deadline for consolidation or reporting before finding a real decrease.

In practiceObserve and understand

One figure, three different questions

Before comparing, identify the indicator and its denominator. Cases relative to a population, the proportion of positive tests and a percentage change have different meanings.

One figure, three different questions — explanations and limits
Hypothetical example	Calculation	What it describes
50 cases per 100,000 residents	50 / 100,000 × 100,000 = 50.	A rate for the period considered, with its case definition.
40 positives among 400 tests	40 / 400 × 100 = 10%.	Positivity among tests, not prevalence in the population.
From 10% to 15%	+5 percentage points; +50% relative change.	Two different ways to express the same change.

Compare counts, percentage points and percentages

Two hypothetical populations each have 1,000 people. Cases rise from 100 to 150. Change the counts to see why a percentage-point difference and a relative change answer different questions. Each person is counted at most once per period.

Period A10 %10 %

Period B15 %15 %

+5 percentage pointsDifference in proportions (B − A)

+50 %Relative change from A

100 cases become 150: +50 cases in equal populations of 1,000 people per period.

Why do these two results differ?

Proportion = cases ÷ population × 100. Percentage-point difference = proportion B − proportion A. Relative change = (cases B − cases A) ÷ cases A × 100, because the populations have equal size here. If A is zero, relative change is undefined. Calculations precede rounding.

The example does not account for age, recruitment or observation duration. A real health comparison must also check these factors. With different denominators, compare consistently constructed rates or proportions.

Keep in mind

Document method changes, repeat testing and reporting delays before explaining a rise. These examples are not outbreak data.

References for these explanations

CDC — Evaluating surveillance systems: sensitivity and predictive value ↗

Sources consulted on 15/10/2026. Examples and diagrams are educational; the original guidance defines its scope.

QuestionsClarifications

Frequently Asked Questions

Why does the last point of a curve sometimes drop?

Recent data may be incomplete due to the delay between event, confirmation and reporting. Check the consolidation date.

Can we compare two countries directly?

Only with caution: definitions, access to tests, system coverage and reporting dates may vary.

Is a rate always preferable to a gross number?

It helps to compare populations of different sizes, but does not alone correct differences in age, screening or reporting.

Why do the last few weeks change sometimes?

Results can be delayed and re-attributed to their event date; recent series are often provisional.

Institutional sources

External sources can evolve. Check their date and instructions for your territory.

PratiqueCheck before deciding

Éviter trois comparaisons trompeuses

Reading guidance and interpretation limits
Concept or situation	What it describes	Point to watch
Nombre et taux	Un nombre décrit un volume ; un taux utilise un dénominateur.	Comparing populations of different sizes requires specifying the denominator.
Semaine et cumul	Une période courte ne se compare pas à un cumul annuel.	Aligner les dates et signaler les données encore incomplètes.
Détection et incidence	Davantage de tests peut produire davantage de détections.	Check testing, case definitions and changes in methods.