03 · What You Need to Know
How to Build a Defensible Historical Exposure Assessment Plan
What Counts as an Exposure in a Case-Control Study?
An exposure is a characteristic, experience, condition, or event whose relationship with an outcome is being investigated. It need not involve a physical substance.
Exposures may include occupational chemicals, environmental conditions, medication use, infectious agents, behaviors, institutional practices, or social circumstances.
For example, a study of university withdrawal might investigate previous financial difficulties, academic workload, or changes in employment. A clinical study might examine prior medication use or occupational hazards.
The term describes the variable's role in the research question, not a conclusion that the variable causes the outcome.
Researchers should distinguish the primary exposure of interest from potential confounders, mediators, effect modifiers, and variables collected for descriptive purposes.
Why Should Exposure Selection Begin With a Causal or Theoretical Rationale?
Historical exposure assessment should begin with what is already known or plausibly hypothesized about the outcome.
Relevant sources include original epidemiological studies, systematic reviews, established mechanisms, occupational or environmental evidence, and discipline-specific theory.
The objective is not to restrict research to relationships already demonstrated. Exploratory studies may investigate less-established exposures, provided the rationale and exploratory status are transparent.
However, a variable should not be included solely because it appears in an accessible database or because the researcher hopes that enough statistical tests will produce significant findings.
A prespecified rationale helps distinguish scientifically motivated exposure selection from post hoc searching.
It also clarifies what information must be measured. If the hypothesis concerns repeated exposure to a chemical, a simple yes-or-no question about ever having worked in an industry may be an inadequate operational measure.
How Do You Establish Whether Exposure Preceded the Outcome?
Temporal ordering is fundamental when investigating whether an exposure may contribute to an outcome.
Researchers should identify the relevant event date and determine which exposures occurred before it.
For incident cases, the reference point may be symptom onset, diagnosis, or another justified event date. These dates are not always equivalent.
Controls should receive appropriate reference dates consistent with the control-sampling strategy so that exposure histories represent comparable opportunities for exposure.
For example, if cases are asked about medication use during the five years preceding diagnosis, controls should not automatically be asked about the five years preceding their interview. Different reference periods could produce unequal exposure opportunities.
Exposure measurement should also consider whether early manifestations of the outcome could have changed the exposure before formal diagnosis.
A behavior recorded before diagnosis may still be a consequence of an undiagnosed disease process.
What Is an Exposure Window?
An exposure window is the period during which an exposure is assessed for its potential relationship with the outcome.
The relevant window depends on the hypothesized causal mechanism, latency, induction period, and available information.
For some acute outcomes, exposure shortly before the event may be most relevant. For diseases that develop over many years, cumulative or distant historical exposure may be more informative.
Researchers should distinguish the period during which exposure may contribute to disease development from the period between exposure and observable disease occurrence.
For example, a hypothetical occupational study might evaluate exposure occurring between 5 and 20 years before diagnosis rather than exposure during the immediately preceding year.
That window requires scientific justification. It should not be selected merely because it produces the strongest observed association.
Should You Measure Recent, Distant, or Lifetime Exposure?
The answer depends on the outcome and exposure mechanism.
| Exposure window |
Potential use |
Important limitation |
| Recent exposure |
Investigating acute or short-latency effects |
May be affected by symptoms or behavior changes before diagnosis |
| Distant historical exposure |
Investigating outcomes with long induction or latency periods |
Records may be incomplete and recall less accurate |
| Cumulative exposure |
Investigating effects associated with repeated exposure over time |
Requires assumptions about duration, intensity, and accumulation |
| Lifetime exposure |
Characterizing broad exposure history when scientifically justified |
Can obscure critical windows and produce substantial measurement burden |
| Time-specific exposure |
Investigating sensitive developmental or occupational periods |
Requires reliable timing and sufficient observations within each period |
These categories may overlap. A cumulative measure can be calculated within a specific historical window, and several windows may be examined when justified by competing causal hypotheses.
However, testing many alternative windows without a prespecified strategy increases the risk of selective interpretation.
Should Exposure Be Measured as Yes or No?
A binary classification may be appropriate when the exposure is clearly defined and the research question concerns whether it occurred.
However, many exposures vary in frequency, duration, intensity, or timing.
For example, two workers may both be classified as exposed to a solvent, although one handled it occasionally for six months and the other worked with it daily for ten years.
Combining them into a single exposed category may conceal important differences.
Depending on the hypothesis and available data, researchers may measure exposure using categories, continuous variables, cumulative indices, or time-varying histories.
The measurement scale should reflect meaningful exposure differences and the reliability of available information.
More detailed measurement is not automatically more accurate. Asking participants to estimate precise historical concentrations without records may create an appearance of precision unsupported by their memories.
How Do Duration, Frequency, and Intensity Differ?
Duration describes how long an exposure occurred. Frequency describes how often it occurred, while intensity describes its magnitude or concentration during exposure.
These dimensions may have different relationships with the outcome.
For example, a study of occupational noise might measure years of employment, hours of exposure per week, and sound intensity. None of these alone necessarily captures the complete exposure experience.
When the causal hypothesis concerns accumulated exposure, a cumulative measure may be appropriate.
The resulting index describes accumulated exposure under the chosen model. It does not establish biological dose, absorbed exposure, or a causal effect.
For some exposures, peaks, recovery periods, thresholds, or nonlinear relationships may matter more than a simple cumulative sum.
How Should Researchers Identify Potential Confounders?
Exposure assessment should include information needed to evaluate plausible confounding, not only the primary exposure.
A confounder is generally a variable whose causal relationship with exposure and outcome can distort the association of interest if it is not appropriately addressed.
Researchers should use substantive knowledge and causal reasoning to identify candidate confounders before analysis.
For example, age may influence both historical occupational exposure and disease occurrence. Other variables may represent shared causes of exposure and outcome.
However, not every predictor of the outcome is a confounder. Variables on the causal pathway or acting as colliders may be inappropriate to control when estimating a total effect.
A directed acyclic graph can help organize assumptions about these relationships.
The same reasoning is relevant when deciding which variables should not be used for matching.
How Many Exposures Can One Case-Control Study Examine?
A case-control study can investigate multiple exposures, particularly when several plausible factors may be associated with the same outcome.
However, the number of variables should be consistent with the available sample size, measurement quality, and analytical objectives.
Each additional exposure may require separate operational definitions, historical windows, validation procedures, and decisions about confounding.
Researchers should distinguish a primary hypothesis from secondary and exploratory analyses.
When numerous exposures are tested, multiple comparisons and selective reporting become important concerns.
The design considerations for examining multiple exposures within one case-control study therefore extend beyond simply adding more questionnaire items.
What If an Exposure Is Rare?
A rare exposure may produce very few exposed cases or controls, limiting precision and the feasibility of adjusted analyses.
Suppose an exposure occurs in only 2% of the source population. Even a moderately sized control sample may contain few exposed individuals.
Before selecting such an exposure as a primary variable, researchers should estimate its likely frequency and assess whether the study can provide useful information about the intended association.
A rare exposure may still be scientifically important, especially when its potential consequences are substantial. However, the study may require a larger sample, specialized sampling, pooling across sites, or a different design.
Exposure selection should therefore account for both scientific relevance and the availability of informative observations.
How Do You Evaluate Historical Measurement Quality?
The best exposure variable is not necessarily the one with the most detailed conceptual definition. It must also be measurable with sufficient credibility.
Historical information may come from interviews, employment records, environmental monitoring, medication databases, archived questionnaires, or stored biological specimens.
Each source measures a particular aspect of exposure and has limitations.
Employment records may establish job duration but not actual chemical concentration. Prescription records may document dispensing but not adherence. Biological specimens may capture recent exposure but fail to represent an earlier critical window.
Researchers should assess whether the available measure corresponds to the intended exposure construct and whether its accuracy could differ between cases and controls.
When retrospective self-report is unavoidable, procedures for reducing differential recall bias become especially important.
Can Exposure Measurement Be Influenced by the Outcome?
Yes. This can occur through recall, clinical testing, changes in behavior, or differences in record availability.
For example, people diagnosed with a disease may undergo more extensive testing for a suspected exposure than individuals without the disease.
If researchers use those clinical test results without considering why testing occurred, apparent exposure differences may partly reflect outcome-dependent measurement.
Similarly, individuals may stop taking a medication after early symptoms appear. Medication use shortly before diagnosis may therefore be influenced by the developing outcome.
Researchers should distinguish exposures that plausibly preceded disease development from measurements obtained after the disease process could have altered behavior or ascertainment.
Should You Include Exposures With Weak Existing Evidence?
Potentially, particularly when the study has an exploratory purpose and a plausible scientific rationale.
Weak evidence does not mean an exposure is irrelevant. Some relationships have received little investigation because measurement is difficult or the exposed population is small.
However, researchers should distinguish an exploratory hypothesis from an established risk factor.
They should also consider whether the study is capable of producing interpretable evidence about the exposure.
A poorly measured variable included only because it is novel may contribute less than a carefully specified variable grounded in a plausible mechanism.
Exploratory analyses should be identified transparently and interpreted with appropriate uncertainty.
How Should Exposures Be Prioritized When Resources Are Limited?
A practical prioritization process can evaluate each candidate exposure according to scientific relevance, temporal plausibility, measurement validity, expected variation, and resource requirements.
These criteria should not be reduced mechanically to a single numerical score unless the scoring framework itself has a defensible basis.
For example, an exposure with strong theoretical relevance but poor historical measurement may require additional validation before becoming the primary variable.
An exposure with excellent administrative records but no plausible relationship to the outcome may be better treated as descriptive information or excluded.
Researchers should prioritize measurements that can meaningfully answer the scientific question rather than maximize the number of variables collected.
When Should Exposure Definitions Be Finalized?
Primary exposure definitions, reference periods, categories, and analytical plans should generally be specified before examining the exposure-outcome results.
This reduces opportunities to select whichever exposure definition produces the most favorable estimate.
For example, testing exposure windows of one, three, five, seven, and ten years and reporting only the strongest association can create selective-analysis concerns.
Alternative windows may be scientifically justified, but their rationale and exploratory status should be documented.
Changes prompted by new methodological information can be legitimate. Researchers should distinguish such amendments from decisions made after reviewing the observed associations.
Transparent prespecification does not guarantee correctness, but it helps readers understand which findings were anticipated and which emerged during analysis.
