Population data that support a public health decision
Surveillance, inequalities, intervention evaluation and geographical analysis: the statistical work behind a defensible health policy.
Population data are rarely comparable as they arrive. An incidence rate depends on the denominator behind it, a survey depends on its sampling design, a geographical difference can disappear after adjusting for age structure, and a historical series can change shape because the notification system changed, not because the population did. Before interpreting a signal, you need to know which part belongs to the phenomenon and which part to the way it is measured.
We work with public health teams, epidemiology departments, health observatories and population research groups. The statistical work consists of turning registries, surveys, time series and geographical or digital signals into comparable estimates, with a clear benchmark for telling what is change, what is expected variation and what uncertainty comes with each reading. The health decision remains with those who know the population and manage the intervention.
It is one of Azahar's two core areas of work, with six published case studies and our own research output in surveillance, infodemiology and population impact evaluation.
We regularly work with
Epidemiology and surveillance departments that collect indicators routinely and need to tell a real deviation from expected variation before triggering a response.
Population health research groups with competitive funding who work with surveys, registries or multiple sources and need to preserve the representativeness and traceability of the analysis.
Health observatories and planning units that need to compare areas or groups without mistaking demographic, social or coverage differences for differences in the phenomenon being studied.
Teams that have already rolled out a population intervention and need to know how much of the change can be attributed to it and what would have happened without it.
Where we usually come into a public health project
In public health, the method depends on what you are trying to tell apart. Before collecting, you may need to design a survey or validate an instrument; with the intervention already under way, to build a comparison that allows its effect to be evaluated; and in surveillance, to separate trend, geography, inequality or digital signal from expected variation. If you would rather start from what you need to do than from your sector, each block leads to the full path for that need.
Design a study
A study is won or lost before the first data point is collected.
Population representativeness through probability sampling designs
View service →Validate the instrument and analyse what it measures: reliability, clinical scales and patient-reported outcomes
View service →I have data and don't know where to start
You already have the data. The question is what to do with them.
From sequencing data to integration with the rest of your study
View service →How patients move between states, where their trajectory changes and where follow-up is lost
View service →Find out what works
Did what you did work? And what is worth changing?
Integrated mortality and morbidity metrics for clinical decisions
View service →We evaluate the observed effect of an intervention already in place against a defensible counterfactual scenario
View service →Monitor populations and places
Understand what is happening to a population before it becomes a headline.
Continuous indicator monitoring and statistical detection of deviations
View service →Identifying geographical patterns
View service →Quantifying health gaps between social groups
View service →Structural factors that shape population health
View service →Complementary epidemiological surveillance based on web search behaviour
View service →Monitoring narratives, misinformation and the spread of health information on digital platforms
View service →The rest of the catalogue, organised by what you need to do
Population problems tend to cut across several methodological layers. Before monitoring, sources may need preparing and harmonising; before talking about an effect, a causal strategy needs defining; and after the analysis, the results need turning into an app or a report that can be updated without redoing the work from scratch. Those services are organised by the need they address, not by the institution that generates the data.
Projects in this sector
What we are usually asked in this sector
Can you work with administrative registry data that were not collected for research purposes?
Yes, but the first job is to understand the process that generated them. An administrative registry serves management needs: its coding, coverage or inclusion criteria can change over time, and a missing value rarely means the same thing in every centre or period. Before modelling, we review which population it represents, which changes break comparability and which variables can be used with a stable definition. Only then do we decide which questions the source can answer and which would require information the registry never collected.
We have a health survey with a complex sampling design. Does it need specific handling?
Yes. Stratification, clusters and weights are part of the design that turns the people surveyed into an estimate for the population. Ignoring them can distort both the estimate and, above all, its variance, producing confidence intervals and tests that do not reflect the study's real precision. The analysis builds in that design from the start and keeps the weights and sampling structure in every estimate, comparison and model that needs them.
What can be claimed about the effect of a population intervention that is already in place?
Only what the available comparison allows. Without random allocation, estimating an effect requires building a defensible counterfactual with the design that best fits the data (for example, difference-in-differences, interrupted time series or synthetic control) and checking the assumptions it depends on. The result is presented as an estimate conditional on those assumptions, with sensitivity analyses where appropriate. Methodology can strengthen a causal reading, but it does not automatically turn any before-and-after comparison into evidence that the intervention caused the change.
Does your work fit better elsewhere?
Shall we talk?
Tell us what you're working on and we'll give you a straight answer on how we can help.