Formulating an Exposure-Effect Question

From a Broad Idea to an Answerable Question

“Does exposure affect response?” expresses a useful scientific motivation, but it does not yet specify an answer. The answer could differ depending on which patients are considered, how treatment is given, which pharmacodynamic outcome is recorded, when it is measured, and what exposure values are compared. A well-formulated question fixes those choices before data are analyzed.

This is not merely a writing exercise. Each component defines the scope of the scientific claim. If the population or observation time changes, the question changes too. If the exposure contrast is not stated, there is no single comparison to estimate or interpret.

The Components That Make the Question Specific

Start by naming the target population: the people or other individuals to whom the answer is intended to apply. “Patients with the disease” may be too broad if the intended population is adults with a particular disease severity, treatment history, or eligibility profile. State relevant boundaries without listing characteristics that do not affect the intended scope.

Next, specify the dosing conditions relevant to the question. These may include the treatment regimen, duration on treatment, and any protocol conditions that must be consistent when the outcome is observed. This helps distinguish the intended setting from, for example, a single-dose assessment and an assessment after sustained treatment. The conditions should be detailed enough that a reader can tell which treatment context the answer concerns.

Name the pharmacodynamic outcome in observable terms. “Response” or “clinical benefit” is usually too vague: identify what will be recorded and its units or scale when applicable. Also state the observation time, such as a scheduled hour after treatment or a specified study visit. “After treatment” is not a precise time if measurements could be taken at different points.

Finally, specify the exposure variable and the contrast of interest. State what exposure quantity is to be used and how its values or groups are identified in the planned data. The comparison might be between two prespecified values, two defined ranges, or a specified increment. A question that says only “higher exposure” leaves open what counts as higher and what it is being compared with.

A compact way to express a descriptive comparison is to define \(Y(t)\) as the outcome recorded at time \(t\), and \(X\) as the named exposure variable. For a higher value \(x_H\) and a lower value \(x_L\), the population contrast can be written:

Here, \(E[\cdot]\) denotes a population mean. The sign of \(\Delta(t)\) depends on the order of subtraction: it is the outcome at the higher specified exposure minus the outcome at the lower specified exposure. This expression describes a difference in outcomes associated with the two exposure values; by itself, it does not establish that changing exposure would cause the outcome to change.

Why Each Specification Matters

Suppose an outcome is measured repeatedly after treatment. The outcome at one hour and the outcome at six hours are different answers to different questions. Likewise, an average difference among newly treated patients need not match the difference among patients treated for several weeks. A reported contrast has a meaningful interpretation only alongside the population, conditions, outcome, time, and comparison that define it.

The following examples use entirely hypothetical populations and data. Their purpose is to show how the question can be formulated and how a stated sample contrast is calculated. They do not establish a causal treatment effect.

Justifying the Formulation and Checking Feasibility

The need to state the components can be seen directly from the contrast. If the population, dosing conditions, outcome, time, or exposure values are omitted, the two terms in \(\Delta(t)\) are not fully defined. For example, a contrast at one visit is \(E[Y(t_1)\mid X=x_H]-E[Y(t_1)\mid X=x_L]\), whereas one at another visit is \(E[Y(t_2)\mid X=x_H]-E[Y(t_2)\mid X=x_L]\). Unless the outcomes are known to be identical across times—which should not be presumed—these are distinct quantities. The question must therefore fix the time as well as the comparison.

Before treating a formulated question as answerable, check that the planned data could support it. Both exposure values or groups should occur in the relevant population under the stated conditions. The outcome should be recorded in a consistent way at the stated time, and the exposure variable should be available in a form that permits the intended comparison. If one group has no observations, or if measurements are taken at inconsistent times, the proposed contrast cannot be estimated as stated without additional assumptions or a change in the question.

Also decide whether the aim is descriptive or causal. A comparison of outcomes across observed exposure values can describe an association. It does not alone show that changing exposure would produce the observed difference. A causal question requires a design and assumptions that address why individuals have different exposure values, including other factors that may influence both exposure and outcome. Use causal language only when the study supports that interpretation.

Alternative Formulations and Their Limits

A question can compare prespecified groups or compare two explicit values of a continuous exposure variable. Group comparisons may be easier to communicate, but group boundaries must be stated and the results apply to those groups. A comparison at two explicit values can preserve more detail about the intended contrast, but the data must support those values. Neither wording is automatically better: the choice should match the scientific question and available observations.

A broad exploratory question can be useful when the goal is to examine patterns and generate hypotheses. It is not a substitute for a focused primary question when a particular outcome, time, and exposure contrast are intended. Similarly, collecting outcomes at many times does not make “the effect over time” a single unambiguous question. Each time-specific comparison should be identified, or the question should state clearly how the time course will be summarized.

A precise question also cannot fix poor measurement, an unrepresentative sample, or a design that does not support the intended interpretation. Formulation makes those limitations visible; it does not remove them.

Common Mistakes and Diagnostics

• Using an undefined population. “Patients with the condition” may include groups outside the intended study. State the eligibility features that delimit the target population.
• Leaving the outcome as “response.” Name the recorded outcome, its scale or units, and when it is observed.
• Writing “higher versus lower” without specifying the comparison. State the exposure variable and define the values, ranges, or groups intended by those labels.
• Using an unspecified time. “After treatment” may refer to multiple visits. Give the scheduled observation time relevant to the question.
• Interpreting an association as a causal effect. An observed group difference is not, by itself, evidence that changing exposure caused the outcome difference.
• Choosing a contrast unsupported by the data. Check that the planned population contains observations for both comparison values or groups and that the required outcomes are recorded under the stated conditions.