Prevalence Studies and Health Surveys

Overview

A prevalence study estimates how common a health condition, characteristic, exposure or service outcome is in a defined population. A health survey is a structured system for collecting comparable information from a sample or census, often through interviews, questionnaires, examinations, measurements or record linkage. Many health surveys are cross-sectional, but some are repeated cross-sectional or longitudinal.

A defensible prevalence estimate requires more than counting positive responses. The numerator, denominator, target population, reference period, sampling design, measurement method, non-response and weighting all affect what the estimate means.[1–3]

This lecture develops the design of prevalence studies and population health surveys, including repeated surveys, complex sampling, standardisation and interpretation.

Why this matters

Health systems use prevalence estimates to quantify need, plan services, monitor inequalities and evaluate coverage. A biased estimate can misdirect resources even when its confidence interval is narrow.

The same observed percentage can represent very different quantities:

• point prevalence on a specified date;
• period prevalence over a month or year;
• lifetime prevalence;
• proportion ever diagnosed;
• screening positivity;
• treatment coverage; or
• self-reported symptoms.

These are not interchangeable. Their meaning depends on case definition, recall period and ascertainment.

Key concepts

Define the prevalence quantity

Point prevalence is the proportion of the population meeting the case definition at a specified time. Period prevalence includes anyone meeting the definition during a stated interval. Lifetime prevalence depends heavily on recall and survival.

For a binary state:

prevalence = number meeting the case definition ÷ number in the defined population assessed for that state

The denominator must exclude only people who are genuinely outside the target definition. Missing outcome measurements should not be silently counted as negative.

Case definition determines the estimate

Hypertension prevalence may be defined using:

• measured blood pressure at one visit;
• repeated measurements within the visit;
• diagnosis history;
• current antihypertensive treatment;
• a combination of measured and treatment criteria; or
• ambulatory confirmation.

Each definition serves a different purpose. Measurement at one visit can estimate survey-defined raised blood pressure but may not equal a clinical diagnosis.

A survey is a measurement system

A health survey combines:

• coverage of the target population;
• sample selection;
• contact and consent;
• questionnaire and examination procedures;
• field quality assurance;
• data processing;
• weighting and estimation; and
• disclosure and reporting.

When to use this approach

DECISION GUIDE

When a prevalence survey is appropriate

• The decision requires the burden or distribution of an existing state.
• The target population can be defined geographically, institutionally or demographically.
• A sampling frame or multistage sampling process can reach the population.
• The condition can be measured validly within the survey encounter or reference period.
• The required precision is feasible after allowing for clustering and non-response.
• Incidence, progression and individual change are not the primary estimands.

Repeated cross-sectional surveys are appropriate for population trends when independent samples are drawn using comparable definitions and procedures at successive time points. They estimate change at population level but do not show within-person trajectories.

Research question and methodological fit

The primary question is:

What is the age- and sex-specific prevalence of survey-defined hypertension among adults living in District A in 2027?

Related questions include:

• What proportion of adults with hypertension are aware of their condition?
• What proportion report treatment?
• What proportion have controlled blood pressure under the survey definition?
• How does prevalence differ by locality and socioeconomic group?

These are descriptive population questions. A household survey is preferable to a clinic sample because many adults with hypertension may not attend care.

Design, setting and study population

DESIGN PROFILE

Population prevalence survey

Core structure: A sample representing a defined population is assessed during a specified survey period using standardised case definitions and measurements.

Best suited for: Estimating burden, coverage, awareness, treatment, control and population inequalities.

Important limitations: Coverage and non-response may distort estimates; one-occasion measurements may not equal clinical diagnosis; complex designs require design-aware analysis.

The target population might be all usual residents aged 18 years or older. The source population is adults living in sampled households during the field period. The protocol must decide how to treat institutions, migrant workers, temporary residents, people without stable housing and those unable to respond without assistance.

Sampling and recruitment

Large health surveys commonly use stratified multistage cluster sampling. For example:

1. stratify by urban and rural locality;
2. select enumeration areas with known probabilities;
3. list or update households;
4. select households systematically;
5. select one or all eligible adults; and
6. calculate inclusion probabilities and survey weights.

The World Health Organization STEPwise approach illustrates a standardised framework for noncommunicable-disease risk-factor surveys while allowing adaptation to national contexts.[2,3]

PROTOCOL CHECKLIST

Population survey sampling plan

• Define strata, clusters, household and participant selection stages.
• Record selection probabilities at every stage.
• Specify household and within-household non-response follow-up.
• Plan fieldwork across days and times to reduce availability bias.
• Determine whether replacement of non-responding households is prohibited.
• Inflate sample size for design effect and anticipated non-response.
• Prespecify base weights, non-response adjustment and calibration.

Replacing a non-responding household with a nearby cooperative household can undermine probability sampling because the replacement does not necessarily have the same selection probability or characteristics.

Variables, constructs and measurements

Hypertension survey measurement framework

illustrative

Table 1. Illustrative survey definitions

Indicator	Numerator	Denominator	Measurement
Raised blood pressure	Adults meeting the survey threshold	Adults with valid blood-pressure measurements	Mean of specified readings under standard protocol
Hypertension	Raised blood pressure or current antihypertensive treatment	Adults with valid measurement and medication data	Measurement, interview and medicine verification
Awareness	Participants with hypertension reporting prior diagnosis	Participants meeting survey hypertension definition	Interview
Treatment	Participants with hypertension reporting current medication	Participants meeting survey hypertension definition	Interview and medicine inspection
Control	Treated participants below specified threshold	Treated participants with valid blood-pressure measurement	Standardised measurement

Note: Thresholds and measurement rules must be specified in the protocol and interpreted as survey definitions.

Measurement quality requires calibrated devices, correct cuff sizes, participant preparation, repeated readings, standardised rest and documented exclusions. Interview items need translation, pretesting and consistent reference periods.

Survey weighting, standardisation and domains

A final survey weight commonly begins with the inverse of the participant’s selection probability. It may then be adjusted for non-response and calibrated to trusted population totals. Each step should be documented with enough detail to reproduce the weight. Extreme weights can increase variance; trimming may improve stability but changes the estimator and requires justification.

Weights cannot guarantee unbiasedness. Non-response adjustment depends on variables that are available for respondents and non-respondents and that explain participation and the outcome. Calibration to age and sex does not address selection related to unmeasured health status, work schedule or access.

When subgroup estimates are planned, the subgroup is a domain of the survey population. Variance should be estimated using survey-domain procedures rather than deleting all observations outside the subgroup when that deletion would lose design information. Prespecified domains also reduce selective reporting of unstable comparisons.

Age standardisation supports comparisons when populations have different age structures. The team should name the standard population, age groups and whether the published values are crude or standardised. Standardised prevalence is a comparison summary rather than the literal observed percentage in the study population.

Cluster sampling can be efficient operationally but usually reduces statistical efficiency because people within clusters resemble one another. The design effect used in sample-size planning and the realised design effect in analysis need not be identical. The sampling and variance-estimation methods should preserve the actual selection stages.[4]

Prevalence-survey reports should follow relevant observational reporting guidance, with explicit sampling, participant flow, variables and analysis details.[5]

Procedures and data collection

Field manuals should cover household approach, participant selection, consent, privacy, measurement order, device checks, referral thresholds and adverse findings. Supervisors should review completeness, timing, unusual distributions and protocol deviations while fieldwork is active.

A survey may need a referral pathway for markedly raised blood pressure or other clinically important findings. The research team must explain that screening measurements are not a complete diagnosis while avoiding false reassurance.

Bias, validity and quality assurance

Potential problems include:

• outdated household lists;
• differential participation by work schedule or health status;
• device drift;
• digit preference or interviewer effects;
• seasonal fieldwork;
• missing examinations among frail participants;
• inaccurate self-report of diagnosis or treatment; and
• post hoc changes to case definitions.

APPRAISAL CHECKLIST

Appraising a prevalence survey

• Is the target population explicit and adequately covered by the frame?
• Were inclusion probabilities known or reconstructable?
• Was the case definition valid, standardised and consistently applied?
• Were response patterns and missing measurements described?
• Did the analysis incorporate weights, strata and clusters?
• Are subgroup estimates sufficiently precise and prespecified?

Ethical and governance considerations

Interface with statistical analysis

Report weighted prevalence with confidence intervals that reflect stratification and clustering. Present unweighted sample counts so readers can understand the observed data. Standardisation may be required when comparing populations with different age structures.

For subgroup estimates, report denominators and uncertainty. Small domains may require cautious presentation or planned oversampling. Repeated surveys should assess whether changes in weighting, measurement or population coverage could explain apparent trends.

Design-based and model-based inferenceMMed extension

Design-based inference uses the known sampling process to estimate population quantities and uncertainty. Model-based approaches use statistical assumptions to predict or adjust. They can complement each other, but a model does not erase an unknown or severely biased selection mechanism. The inferential framework should be declared rather than mixed implicitly.

Worked example

WORKED EXAMPLE

From district map to prevalence estimate

The district is stratified into urban, rural accessible and rural remote areas. Enumeration areas are selected with probability proportional to size, followed by household listing and systematic household selection. One adult is randomly selected per household. Survey weights combine the inverse inclusion probability, non-response adjustment within weighting classes and calibration to current age-sex totals.

Blood pressure is measured three times after rest using validated automated devices. The analysis uses the mean of the final two readings. The report presents weighted prevalence, age-standardised comparisons, response disposition at each stage and sensitivity analyses excluding participants with incomplete medication information.

Suggested protocol or reporting wording

Common mistakes

Knowledge check

Why can a very large convenience sample still give a biased prevalence estimate?

Answer and reasoning

Large size reduces random error around the sample result but does not make excluded or self-selected people comparable with the target population. Systematic coverage or response differences can persist.

What is the difference between repeated cross-sectional and longitudinal surveys?

Answer and reasoning

Repeated cross-sectional surveys sample population snapshots at several times, usually with different people. Longitudinal surveys follow the same participants, allowing within-person change to be studied.

Why should unweighted counts accompany weighted prevalence estimates?

Answer and reasoning

Weights describe population representation, whereas unweighted counts show the actual number of observations supporting each estimate. Both are needed to judge precision and data sparsity.

A participant has no valid blood-pressure reading. Should they be counted as not hypertensive?

Answer and reasoning

No. Their outcome is missing unless another prespecified component of the case definition establishes status. Counting missing measurements as negative can bias prevalence downward.

Key take-home messages

SUMMARY

Key take-home messages

• Prevalence requires an explicit population, time, case definition and denominator.
• Population health surveys are integrated sampling and measurement systems.
• Complex sampling and non-response must be reflected in estimation and uncertainty.
• Large samples do not repair coverage or measurement bias.
• Repeated cross-sectional surveys estimate population trends, not individual trajectories.

References

1. World Health Organization. WHO STEPS surveillance manual: the WHO STEPwise approach to noncommunicable disease risk factor surveillance. Geneva: World Health Organization; 2024. https://www.who.int/docs/default-source/ncds/ncd-surveillance/steps/steps-manual.pdf
2. World Health Organization. STEPwise approach to NCD risk factor surveillance (STEPS). Geneva: World Health Organization. https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/steps
3. Riley L, Guthold R, Cowan M, Savin S, Bhatti L, Armstrong T, et al. The World Health Organization STEPwise approach to noncommunicable disease risk-factor surveillance: methods, challenges, and opportunities. Am J Public Health. 2016;106(1):74–78. https://doi.org/10.2105/AJPH.2015.302962
4. Bennett S, Woods T, Liyanage WM, Smith DL. A simplified general method for cluster-sample surveys of health in developing countries. World Health Stat Q. 1991;44(3):98–106.
5. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. PLoS Med. 2007;4(10):e296. https://doi.org/10.1371/journal.pmed.0040296