Denominator Problems in Sub-Saharan Surveillance Data

Where denominators go wrong

In many parts of sub-Saharan Africa, the most recent census may be a decade old. The population has grown, shifted between urban and rural areas, and been affected by migration, conflict, or economic change since the last count. Health facilities still use those old numbers because there is nothing better available.

This means a district that reports 200 malaria cases looks very different depending on whether the denominator says 50,000 people or 80,000 people. The first gives you a rate of 400 per 100,000. The second gives you 250 per 100,000. Same cases, different story.

Catchment area estimates add another layer of uncertainty. A clinic may officially serve a population of 30,000, but if a nearby facility closed, the real catchment could be double that. Nobody updated the number.

Cross-border movement complicates things further. In border districts of Nigeria, Ghana, and Cameroon, populations shift seasonally for trade and agriculture. A district health team planning for 40,000 residents may actually serve 60,000 during harvest season and 30,000 during the dry months. The denominator is not just wrong; it is wrong by a different amount every quarter.

Urban informal settlements present a particularly acute version of this problem. Population density changes rapidly as people arrive seeking employment. A settlement that housed 15,000 people when the last survey was conducted may have 45,000 residents three years later, but the health post still plans around the old figure. Disease incidence rates calculated against the outdated number look three times worse than they actually are.

The downstream effects

Bad denominators distort comparisons between districts, between time periods, and between programmes. A district that looks like it has rising disease burden may simply have a shrinking denominator because people moved away and the census was not updated.

Resource allocation follows these numbers. Funding formulas, staffing models, and supply chain planning all depend on population estimates. When the estimate is wrong, the allocation is wrong.

This is not a theoretical problem. It happens routinely. Programme managers know their numbers are approximate, but the dashboards rarely say so.

Consider a concrete example from malaria programme planning. A state allocates insecticide-treated nets based on population projections from a decade-old census, extrapolated at 2.8% annual growth. But one local government area experienced net out-migration due to flooding, while another absorbed those displaced people. The first area receives too many nets and warehouses the surplus. The second area runs out within weeks. Both areas report the same distribution rate because the denominators are equally wrong in opposite directions.

International comparisons suffer the same distortion at a larger scale. When global health bodies publish disease burden estimates for countries with outdated census data, the rates may be off by 20-40%. A country that appears to have a higher malaria mortality rate than its neighbour may simply have a worse denominator. Policy decisions, funding priorities, and global health narratives are all shaped by numbers that carry this hidden uncertainty.

• Census data may be 5-15 years old in some districts
• Migration and urbanisation shift populations faster than surveys can track
• Catchment area boundaries rarely match actual patient flow
• Rate calculations inherit every error in the denominator

Practical approaches to denominator correction

Several methods exist to improve denominator estimates between census rounds, though none are perfect. Satellite imagery combined with building footprint analysis can estimate population density in areas where ground surveys are impractical. Gridded population estimate projects produce figures that are updated more frequently than national censuses, though they carry their own modelling assumptions that should be stated alongside any rates derived from them.

Facility-based denominators offer another approach. Instead of using census projections, some programmes estimate their catchment population from utilisation data. If a facility vaccinates 95% of children under one year, and it vaccinated 3,200 children last year, the total under-one population is approximately 3,400. This method works well for programmes with high coverage but breaks down for diseases where care-seeking behaviour varies significantly across communities.

Community health worker registers, where they exist, provide household-level data that can serve as a cross-check on census projections. In parts of Ethiopia, Rwanda, and Nigeria, health extension workers maintain household registers that are updated quarterly. These registers are not perfect because they miss mobile populations and may be incomplete in hard-to-reach areas, but they offer a denominator that is months old rather than years old.

What honest analytics looks like

The answer is not to avoid rate calculations. It is to be transparent about the uncertainty. Reporting a range instead of a single number, flagging districts where the denominator is older than five years, and showing data completeness alongside the trend line all help.

We build population analytics that label data quality explicitly. If a denominator is uncertain, the output says so. If reporting completeness dropped, the dashboard shows it alongside the trend instead of hiding it. That is not a weakness in the product. It is how trustworthy analytics works.

In practice, this means every rate calculation on our platform carries metadata: the denominator source, the year it was last updated, the estimation method, and a confidence qualifier. A district report might show an incidence rate of 320 per 100,000 with a footnote reading 'denominator: 2019 census projection at 2.8% growth, confidence: low.' That footnote changes how a programme manager reads the number, and it should.

We also support side-by-side comparisons using alternative denominator sources where available. If gridded population estimates and census projections disagree by more than 15%, the dashboard shows both rates and flags the discrepancy. This does not solve the denominator problem, but it prevents anyone from treating an uncertain number as though it were precise.