Why the Target Population Matters
A treatment effect is not a number detached from the people to whom it applies. It summarizes a comparison for a specified population, under specified treatment strategies and outcome definitions. Even when the strategies and outcome stay fixed, changing who is included can change the population-level contrast if treatment effects differ among clinical groups.
The target population is the group about whom the treatment question is intended to draw a conclusion. It may be defined by diagnosis, disease severity, age, comorbidities, prior treatment, care setting, or a combination of criteria. The trial’s eligibility rules determine who could enroll, while actual enrollment determines who contributed data. Neither automatically defines the population to which the results should apply: the target should be stated directly.
The treatment strategies and outcome in this definition must be the ones specified for the estimand. For example, a mean symptom score at week 12 under two defined treatment strategies is not interchangeable with a different follow-up time or a different strategy for handling an intercurrent event. This tutorial focuses on the population component: whom the contrast describes.
Making the Population Operational
A useful population definition can be translated into criteria that a clinician or trial team could apply to an individual at a specified time. State the condition and relevant clinical features, the care setting if it matters, and the key inclusion and exclusion criteria. For instance, “adults with the diagnosis” may be too broad if the intended question concerns adults with moderate disease who have not previously received a particular therapy.
Eligibility criteria serve a practical role in selecting trial participants, but they also set boundaries on the evidence directly generated by the trial. A trial restricted to patients with one severity level gives direct evidence for that group. Applying its result to a broader group requires a reason to expect the contrast to carry over, or additional evidence about how effects vary across groups. A broader target may be clinically relevant even when the trial itself is narrower; the distinction should be made explicit rather than hidden in a phrase such as “patients like those in the trial.”
- Clinical criteria describe the disease, severity, history, or other features that make the question relevant.
- Setting and time clarify where and when the target group is defined, such as patients receiving care in a particular type of clinic at treatment initiation.
- Eligibility rules state who is included and excluded. They should be precise enough to distinguish the intended group from neighboring populations.
- Reach describes how far the trial evidence extends beyond those enrolled. It depends on the target, the trial population, and what is known about differences between them.
Worked Example: A Severity-Defined Target Population
Suppose a hypothetical trial compares a new treatment with usual care on a symptom score at week 12; higher scores are better. The intended target is adults with the condition and either moderate or severe disease at treatment initiation. The treatment strategies and outcome are held fixed. In hypothetical data, the mean scores under treatment and usual care are 70 and 60 for moderate disease, and 52 and 50 for severe disease.
The treatment contrast within moderate disease is \(70-60=10\) score points. Within severe disease it is \(52-50=2\) score points. These are distinct population-specific contrasts, not competing estimates of one universal effect.
Suppose the target population is 80% moderate and 20% severe. The target mean under treatment is \(0.80(70)+0.20(52)=56+10.4=66.4\) points. Under usual care it is \(0.80(60)+0.20(50)=48+10=58\) points. Thus the target-population contrast is \(66.4-58=8.4\) points.
Now consider a different target population that is 30% moderate and 70% severe. Its treatment mean is \(0.30(70)+0.70(52)=21+36.4=57.4\) points. Its usual-care mean is \(0.30(60)+0.70(50)=18+35=53\) points, giving a contrast of \(57.4-53=4.4\) points. The treatment strategies and subgroup effects did not change; the average contrast changed because the target population has different proportions of the two severity groups.
The arithmetic assumes the stated subgroup means apply to the target populations and that the two severity groups form a complete, non-overlapping partition of each population. In a real application, those assumptions would need support. The example illustrates why naming the target population is part of interpreting the contrast.
Why Population Composition Changes an Average
The calculation above is an instance of a general result. Let \(C\) denote a set of mutually exclusive, exhaustive clinical strata, and let \(\pi_c(P)=P(C=c)\) be the proportion of target population \(P\) in stratum \(c\). Let \(\Delta_c\) be the mean treatment contrast within that stratum. On the mean-difference scale, the population contrast is the weighted average of the stratum contrasts.
The justification is the law of total expectation applied separately to each treatment strategy, followed by subtraction. The weights are the target population’s stratum proportions, not necessarily the proportions in the trial. If all stratum contrasts are equal, changing the weights leaves this mean difference unchanged. If contrasts differ, a change in weights can change it.
This weighted-average identity is specific to the mean difference as written. For other effect scales, such as a ratio of means, averaging subgroup effects with population proportions generally does not yield the population-level ratio. State the scale and calculate the population contrast on that scale rather than assuming every effect measure follows the same averaging rule.
Worked Example: Comparing Two Eligibility Definitions
Consider a hypothetical treatment question for adults with the same condition and week-12 outcome. Clinical review suggests two strata: patients with no listed comorbidity and patients with a specified comorbidity. Assume the mean-difference treatment effects are 6 points and 2 points, respectively. These effects are hypothetical and are used only to show how an eligibility definition changes the population contrast.
A narrowly defined target excludes the specified comorbidity. Its population consists only of the first stratum, so its contrast is 6 points. A broader target includes both strata and is 50% in each. Its contrast is \(0.50(6)+0.50(2)=3+1=4\) points.
For a different broad target that is 80% in the no-comorbidity stratum and 20% in the comorbidity stratum, the contrast is \(0.80(6)+0.20(2)=4.8+0.4=5.2\) points. All three contrasts are in week-12 score points, and each describes a different target population. The calculation assumes that the two strata exhaust the stated broad target and that the given stratum effects apply to its members.
The narrow eligibility definition may provide a more focused answer for a clinically specific group, while the broader definition may be more relevant to routine care. Neither is inherently the correct population for every question. The choice follows the intended clinical decision.
Generalizability: From Trial Participants to a Target
A trial estimate directly describes the trial population under the conditions and estimand used in the analysis. Generalizing it to a different target population is a separate interpretive step. It is most plausible when the target resembles the trial population in features that affect the treatment contrast, and when the trial includes enough information to assess those features.
A measured characteristic can matter for generalizability if its distribution differs between trial and target populations and if the treatment contrast varies with it. Such a characteristic is an effect modifier for the contrast under consideration. A difference in age distributions, for example, does not by itself prove that the trial result will fail to apply; it raises a question about whether age is associated with differing effects and whether relevant ages were represented.
For a direct population-weighted calculation, the stratum-specific contrasts must be relevant to the target, the strata must be defined consistently in both populations, and the target proportions must be available. If a clinically important target stratum was absent from the trial, its contrast cannot be learned from trial observations in that stratum. Applying an observed contrast there requires additional assumptions or evidence; it is not justified merely by having a large overall trial.
Worked Example: A Target Group Outside Trial Eligibility
A hypothetical randomized trial enrolls adults aged 18 to 65 with a specified diagnosis. Its target for a proposed clinical claim is all adults with that diagnosis, including those older than 65. Suppose the target population is 70% aged 18 to 65 and 30% older than 65, but no trial participants are older than 65.
Assume, for illustration, that the trial-supported mean-difference contrast for ages 18 to 65 is 5 points. If the contrast among adults older than 65 were known to be 3 points, then the all-adult target contrast would be \(0.70(5)+0.30(3)=3.5+0.9=4.4\) points.
But the trial contains no observations in the older stratum, so the value of 3 points is not estimated from these trial data. If the older-stratum contrast were instead 1 point, the same target weights would give \(0.70(5)+0.30(1)=3.5+0.3=3.8\) points. The trial result alone does not distinguish these possibilities. The value 4.4 is a calculation conditional on the assumed older-stratum effect, not a result established by the trial.
This example checks the boundary of generalizability: the target includes a group outside the trial’s eligibility criteria, and the missing stratum-specific effect prevents a data-supported weighted contrast for the full target. Evidence from other sources or a clearly defended assumption would be needed to extend the claim.
Diagnostics, Alternatives, and Limitations
Before interpreting a result beyond the enrolled participants, check the population definition and its relationship to the trial. A practical review asks whether eligibility criteria are operational, whether important target groups were excluded or sparsely represented, whether clinical effect modifiers were measured comparably, and whether target-group proportions are known. A large mismatch or an absent target stratum is a warning against unqualified generalization.
There are several distinct choices that are sometimes confused:
| Approach | Population described | Interpretive point |
|---|---|---|
| Use the trial population as the target | People represented by the trial’s enrolled population | Directly tied to trial participation, but may not answer a broader clinical question. |
| Define a clinically restricted target | A specified subgroup, such as patients meeting a severity criterion | May sharpen the clinical question; the trial must provide relevant evidence for that group. |
| Define a broader target | A population including additional clinical groups | Can improve relevance to practice, but requires support for groups not represented or for differences in effects. |
A subgroup analysis is not automatically a generalizability analysis. Reporting a contrast within a subgroup can describe that subgroup, but transporting an estimate to a population with different subgroup proportions also requires the target proportions and relevant subgroup contrasts. Likewise, a trial that includes diverse participants does not guarantee generalizability to every population: some groups may still be absent, and effects may vary in ways the trial cannot assess.
The main limitations are substantive. Eligibility criteria can exclude people who are common in practice, and broadening the target can require assumptions about treatment effects in less-studied groups. Clinical characteristics used to define strata may be measured inconsistently, and target-population proportions may be uncertain or unavailable. Finally, a population average can conceal important variation: two targets can have the same average contrast while their patients’ subgroup effects differ.
Common Pitfall
A careful conclusion names both the contrast and its population. For example: “Among adults meeting the stated moderate-or-severe disease criteria, the mean week-12 symptom-score difference was 8.4 points under the specified treatment strategies.” This is more informative than calling 8.4 points simply “the treatment effect.”
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A trial enrolls only adults with moderate disease. What additional information is needed to claim its contrast applies to adults with both moderate and severe disease?
- Two strata have mean-difference treatment effects of 8 and 2 points. A target population is 25% in the first stratum and 75% in the second. What is its average contrast?
- Explain why a trial can have many participants and still provide no direct estimate for a target group excluded by its eligibility criteria.
- If all stratum-specific effects are equal on the mean-difference scale, what happens to the population contrast when target stratum proportions change?
Solutions: (1) The target’s stratum proportions and relevant effects in both strata are needed. The trial must provide evidence for the severe group, or an additional assumption or source of evidence must justify the contrast there. (2) The weighted contrast is \(0.25(8)+0.75(2)=2+1.5=3.5\) points. (3) Sample size adds information about represented participants but cannot supply observations from a stratum with no trial participants; its treatment contrast is not directly estimated. (4) The average remains the common stratum effect, because a weighted average of equal values is unchanged by the weights.