Target, Source and Study Populations

Overview

Population terminology is not decorative protocol language. It identifies to whom the research question applies, from whom participants can actually be obtained, who enters the study, and whose data contribute to the final result. These groups may overlap, but they are rarely identical.

This lecture distinguishes the target, source, sampled, recruited and analysis populations. It also explains the unit of sampling, observation and analysis, and shows how participant flow affects internal validity and the extent to which findings may be applied beyond the study sample.

Why this matters

A study can use excellent measurements and appropriate statistical methods yet answer the wrong population-level question. Common problems include recruiting only from a tertiary centre while describing the target population as all patients in the community, excluding participants after enrolment without reporting why, or treating multiple records from one person as though they were independent participants.

Population definitions influence:

• whether eligible people had a realistic opportunity to be included;
• whether selection is related to the exposure, intervention or outcome;
• whether the study estimates the intended quantity;
• whether findings can reasonably inform practice in another setting; and
• whether clinicians, reviewers and policymakers can identify the boundaries of inference.

Key concepts

Target population

The target population is the full group about which the researchers intend to make an inference. It should be defined by people, place, time and relevant clinical characteristics. A target population may be broad, but it must remain scientifically meaningful.

Source population

The source population is the population from which the study participants arise. It is constrained by the study setting, period, health system, registers and access arrangements. In a well-designed study, the source population is a defensible operational representation of the target population.

Sampled, recruited and analysis populations

The sampled population consists of the units selected or invited through the sampling procedure. The recruited population consists of people who enrol or whose records are included after eligibility and consent or waiver processes. The analysis population consists of participants whose data contribute to a particular analysis.

Different analyses may have different denominators. For example, 1,200 women may be enrolled, 1,170 may have complete outcome ascertainment, and 1,105 may have all variables required for a complete-case regression model.

Units must be named explicitly

The sampling unit is selected during sampling, the unit of observation is measured, and the unit of analysis contributes an independent analytical unit unless the analysis models dependence.

These units can differ. A survey may sample clinics, then patients within clinics. The clinic is a first-stage sampling unit, the patient is the observation unit, and the analysis must account for patients being clustered within clinics. A pathology study may observe several slides per patient; slides are observations, but they are not necessarily independent patients.

DECISION GUIDE

Define a study population in five questions

1. About whom is the final inference intended?
2. Which real-world population could give rise to eligible participants?
3. What list, facility, community or register identifies potential units?
4. Who was approached, enrolled, followed and analysed?
5. What differences between these groups could alter the estimate?

From research question to population definitions

Population definitions should be derived from the research question rather than copied from a standard template. The sequence is:

1. identify the intended clinical or public-health decision;
2. state the target population precisely;
3. define the accessible source population;
4. identify the sampling frame or recruitment source;
5. state eligibility criteria;
6. describe sampling and recruitment;
7. predefine analysis populations; and
8. document participant flow.

For the illustrative study, a defensible target population might be “women diagnosed with gestational diabetes who deliver in public hospitals in the state and remain resident during the first 12 postpartum weeks.” The source population could be “women with gestational diabetes who delivered at the three participating public hospitals between 1 January and 31 December 2025 and were identifiable through linked maternity and laboratory records.”

The second statement is narrower, but it is more honest. Broader applicability can be discussed only after comparing the study setting and participants with the population of interest.

Population map for the illustrative study

illustrative

Table 1. Population stages

Stage	Illustrative definition	Main concern
Target population	Women with gestational diabetes delivering in public hospitals in the state during 2025	Intended inference
Source population	Eligible women delivering at three participating public hospitals	Coverage of the target population
Sampled population	Women selected from monthly delivery lists	Sampling procedure
Recruited population	Selected women who met criteria and entered the study	Refusal and consent
Follow-up population	Enrolled women with outcome ascertainment by 12 weeks	Loss to follow-up
Analysis population	Women included in each prespecified analysis	Missing variables and exclusions

Note: The table is illustrative. Each denominator should be reported.

Internal validity and applicability

Internal validity concerns whether the study result is valid for the people and circumstances actually studied. Applicability, transportability or generalisation concerns whether the result can inform another population or setting.

A highly selected sample does not always invalidate the internal comparison. For example, a randomised treatment effect may be internally valid among enrolled participants even when restrictive eligibility criteria limit applicability. Conversely, a representative sample does not protect against measurement error, confounding or poor follow-up.

Researchers should therefore avoid treating internal and external validity as a simple trade-off. The relevant questions are:

• Was selection into the study related to factors that affect the exposure–outcome or intervention–outcome relationship?
• Are important effect modifiers distributed differently between the study and target populations?
• Are care pathways, resources, baseline risk and outcome definitions sufficiently similar?
• Is the study purpose descriptive, causal, prognostic or predictive?

Writing population definitions in a protocol

PROTOCOL CHECKLIST

Population information to prespecify

• target population, including place and time;
• source population and participating settings;
• sampling frame or recruitment source;
• sampling, observation and analysis units;
• inclusion and exclusion criteria;
• recruitment and consent pathway;
• follow-up period and outcome-ascertainment sources;
• planned analysis populations and treatment of missing data; and
• participant-flow information to be reported.

Avoid vague expressions such as “all patients attending the hospital” unless the period, service, eligibility conditions and route of identification are specified. Also avoid calling a convenience sample a “random sample” merely because researchers approached patients in no preferred order.

Ethical and governance considerations

Interface with statistical analysis

The analysis must preserve the actual data structure. Researchers should state whether the inferential unit is a person, visit, specimen, household or facility. Clustered or repeated observations require methods that acknowledge dependence.

Population definitions also determine the denominator. A prevalence estimate may require sampling weights if inclusion probabilities differ. A treatment-effect analysis may distinguish intention-to-treat, per-protocol and safety populations. A complete-case analysis estimates a quantity for the complete cases unless additional assumptions justify inference to the enrolled population.

See the related Biostatistics lecture on planning a statistical analysis for alignment of the population, estimand, variables and analytical method.

Transporting an estimate to a target populationMMed extension

Transportability methods can reweight or model study results using measured characteristics that differ between the study sample and target population. These methods require a clearly defined target population, comparable variable definitions, adequate overlap and assumptions about effect modifiers. Statistical adjustment cannot repair an undefined target or unmeasured differences that materially modify the effect.

Worked example

WORKED EXAMPLE

Correcting an overbroad population claim

Initial statement: “All Malaysian women with gestational diabetes formed the study population.”

Available data: Three public hospitals in one state, deliveries during 2025, linked maternity and laboratory records, and follow-up through 12 weeks postpartum.

Correction:

• Target population: women with gestational diabetes delivering in public hospitals in the state during 2025.
• Source population: women with gestational diabetes delivering at the three participating hospitals during the study period.
• Sampled population: eligible records selected from monthly delivery lists.
• Recruited population: selected women who consented to prospective follow-up.
• Analysis population: recruited women with ascertainable postpartum testing status.

Interpretation: The study can directly describe the participating-hospital source population under its sampling and missing-data assumptions. Applying the result to private hospitals, other states or national practice requires comparison of service organisation, patient characteristics and testing access.

Suggested protocol or reporting wording

Evidence basis and critical appraisal

Foundational health-research guidance treats population definition as an essential link between the question, sampling plan and interpretation [1]. Appraisal should therefore begin by reconstructing the population pathway rather than by looking only at the final sample size.

Randomised trials frequently use eligibility and recruitment patterns that narrow applicability, so readers should ask whether the enrolled participants resemble the patients for whom the intervention will be considered [2]. Reviews of major medical trials have shown how commonly eligibility criteria exclude substantial proportions of routine-care patients [3].

For internal validity, the important issue is not simply whether the sample is representative. Selection can distort an association when entry into, or retention in, the analysis depends on variables connected to the exposure and outcome [4]. For applicability, researchers should compare measured effect modifiers, care settings and baseline risks between the study sample and the target population. Formal transportability methods can extend an estimate only after both populations and the necessary assumptions are defined [5].

A critical reader should therefore ask:

• Is the target population named independently of the available dataset?
• Could every member of the source population enter the frame?
• Were all selected units given a comparable opportunity to enrol?
• Did follow-up or missing-data exclusions materially redefine the population?
• Are applicability claims supported by setting and participant comparisons?

Common mistakes

Knowledge check

A study recruits adults with hypertension from one specialist clinic. What is the most defensible source population?

Answer and reasoning

Adults with hypertension who could be identified and recruited through that specialist clinic during the stated study period, subject to the eligibility criteria. “All adults with hypertension” would be an unsupported target-to-source leap.

Why can the analysis population differ between outcomes?

Answer and reasoning

Different outcomes may have different ascertainment, follow-up or missing-variable patterns. The denominator for each analysis should therefore be stated rather than assuming one universal analysis sample.

A household survey samples districts, villages and then adults. Which units matter for analysis?

Answer and reasoning

Districts and villages are sampling clusters, while adults are observation units. The analysis must retain information about the multistage design because adults within the same village or district may be correlated and may have unequal inclusion probabilities.

Does a representative sample guarantee an unbiased exposure–outcome association?

Answer and reasoning

No. Representativeness does not remove confounding, measurement error, reverse causation or selection during follow-up. It mainly addresses how the selected sample relates to the population frame and intended descriptive inference.

Key take-home messages

SUMMARY

Key take-home messages

• Define the target, source, sampled, recruited and analysis populations separately.
• Name the sampling, observation and analysis units.
• Population definitions must match the actual setting, period and recruitment source.
• Internal validity and applicability are related but distinct.
• Report participant flow and every analytical denominator.
• Statistical methods cannot repair an undefined population or an incoherent selection process.

References

1. World Health Organization Regional Office for the Western Pacific. Health research methodology: a guide for training in research methods. 2nd ed. Manila: WHO Regional Office for the Western Pacific; 2001. https://iris.who.int/items/0765fa3d-af1d-43b3-bb36-6576e88a7871
2. Rothwell PM. External validity of randomised controlled trials: “to whom do the results of this trial apply?” Lancet. 2005;365(9453):82–93. https://doi.org/10.1016/S0140-6736(04)17670-8
3. Van Spall HGC, Toren A, Kiss A, Fowler RA. Eligibility criteria of randomized controlled trials published in high-impact general medical journals: a systematic sampling review. JAMA. 2007;297(11):1233–1240. https://doi.org/10.1001/jama.297.11.1233
4. Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15(5):615–625. https://doi.org/10.1097/01.ede.0000135174.63482.43
5. Dahabreh IJ, Hernán MA. Extending inferences from a randomized trial to a target population. Eur J Epidemiol. 2019;34(8):719–722. https://doi.org/10.1007/s10654-019-00533-2