03 · What You Need to Know
How to Construct a Scientifically Defensible Case Definition
What Is a Case Definition in Analytical Research?
A case definition is a prespecified set of criteria used to determine whether an individual qualifies as having the outcome under investigation.
It translates an abstract concept such as disease, injury, academic failure, or institutional closure into an observable classification that researchers can apply to eligible individuals.
For a clinical outcome, the definition might involve symptoms, diagnostic tests, physician assessments, or standardized classification criteria. For a nonclinical outcome, it could involve an administrative event, a validated assessment threshold, or another explicitly defined occurrence.
A case definition has two related but distinct functions. It identifies the outcome and establishes which people with that outcome belong to the study's eligible case population.
For example, a diagnosis of a neurological condition may establish outcome status, while residence within a specified region and diagnosis during a defined period determine eligibility for a particular investigation.
Both decisions need justification.
What Elements Should a Case Definition Specify?
A useful case definition generally addresses the following dimensions, although their precise importance varies by outcome and study context.
| Element |
Question to resolve |
Illustrative specification |
| Outcome criteria |
What qualifies as the event or condition? |
Diagnosis meeting specified clinical criteria |
| Diagnostic evidence |
What evidence is sufficient for classification? |
Documented specialist assessment or validated test result |
| Person eligibility |
Who belongs to the population under investigation? |
Adults within a defined age range |
| Geographical boundary |
Where must eligible individuals belong? |
Residents of a specified catchment area |
| Time boundary |
When must the outcome have occurred? |
First qualifying diagnosis during a defined calendar period |
| Case type |
Are newly occurring or existing cases eligible? |
Incident cases only |
| Exclusions |
What conditions make a person ineligible? |
Documented alternative diagnosis meeting predefined exclusion criteria |
The objective is not to make the definition unnecessarily restrictive. It is to ensure that the same eligibility rules can be applied across participants, recruitment sites, and assessors.
How Specific Should the Diagnostic Criteria Be?
The required specificity depends on the outcome, available measurement methods, and consequences of misclassification.
For a disease with established diagnostic criteria, researchers should normally use recognized criteria appropriate to the study period and population. A vague label such as "patients with neurological problems" would be insufficient if the research question concerns one specific neurological disorder.
When standardized criteria do not exist, researchers may need to construct a transparent operational definition based on defensible indicators.
For example, a study of severe academic disengagement might define cases using a validated measure and a prespecified threshold, rather than relying on a teacher's informal judgment that a student appears disengaged.
Such thresholds require justification because different cutoffs can identify substantially different groups.
Researchers should also distinguish a diagnostic classification from the underlying biological or social phenomenon. A recorded diagnosis is an imperfect measurement of disease status, just as an assessment score is an imperfect measurement of a psychological construct.
Should You Include Incident Cases or Prevalent Cases?
This decision can materially affect the meaning of a case-control study.
Incident cases
Individuals whose outcome newly occurs during a defined observation period, according to the study's operational definition of onset or diagnosis.
Prevalent cases
Individuals who already have the outcome at a specified point or during a specified period, regardless of when it first developed.
Incident cases are often preferable when the research question concerns factors associated with disease occurrence. They may reduce the influence of survival or disease duration on who becomes available for recruitment.
Prevalent cases can be more accessible, particularly for rare or chronic conditions. However, they may disproportionately represent people who survive longer, remain in care, or have persistent disease.
Suppose an exposure increases the probability of developing a disease but also reduces survival after diagnosis. A study recruiting only living prevalent cases could underrepresent exposed individuals who died before recruitment. The resulting association might differ from the association with disease occurrence.
This is a form of prevalence-incidence bias, often discussed in relation to Neyman bias.
Incident case recruitment does not eliminate all selection problems. For example, newly diagnosed cases may still exclude people who die before diagnosis or never access healthcare.
Is Date of Diagnosis the Same as Date of Disease Onset?
Not necessarily. A disease may begin before it is recognized, tested, or formally diagnosed.
This distinction matters when establishing exposure windows and temporal ordering. An exposure recorded shortly before diagnosis may actually have occurred after the disease process began.
Researchers should determine whether the relevant date is symptom onset, first clinical presentation, diagnostic confirmation, or another event.
For some outcomes, the exact biological onset cannot be observed. In those circumstances, the study should identify the date used operationally and acknowledge the resulting uncertainty.
Consistency is essential, but consistency alone does not guarantee validity. A uniformly recorded diagnosis date can still be a poor proxy for the beginning of a disease process.
Should You Use Confirmed, Probable, or Suspected Cases?
Some conditions have established surveillance classifications that distinguish confirmed, probable, and suspected cases. These classifications may reflect different combinations of clinical, laboratory, epidemiological, or other evidence.
Researchers should not automatically combine them.
A highly restrictive confirmed-case definition may improve diagnostic specificity but exclude genuine cases who lacked access to confirmatory testing. A broader definition may improve case capture while increasing the possibility of including individuals without the target outcome.
Potential Advantages of a Narrower Definition
- Greater diagnostic consistency when confirmatory criteria are valid.
- Reduced inclusion of individuals with unrelated conditions.
- Clearer interpretation for a precisely defined outcome.
Potential Limitations of a Narrower Definition
- Fewer eligible cases and potentially lower statistical precision.
- Possible exclusion of cases without access to testing.
- Selection of a subgroup whose characteristics differ from the wider affected population.
The best approach depends on the research objective and ascertainment process. Researchers might restrict the primary analysis to confirmed cases and examine a broader classification in a prespecified sensitivity analysis, provided the classifications are defensible and the data support them.
How Can Case Ascertainment Affect the Study?
Case definition and case ascertainment are related but not identical.
The case definition specifies who qualifies. Case ascertainment describes how researchers find and verify individuals who meet those criteria.
Potential ascertainment sources include disease registries, hospital records, laboratory databases, administrative systems, population surveillance, and direct screening.
Each source has a coverage mechanism. A hospital database, for instance, may capture people who seek care at that hospital but miss people treated elsewhere or never diagnosed.
If exposure affects the probability of entering the ascertainment system, the observed exposure distribution among recruited cases may not reflect that among all eligible cases in the source population.
For example, if workers with a particular occupational exposure receive more frequent medical screening, their disease may be diagnosed more often than comparable disease among unexposed workers. This can produce detection-related bias.
Researchers should therefore document the ascertainment source, its coverage, verification procedures, and any plausible relationship between exposure and case identification.
Why Must Case Eligibility Align With the Source Population?
Cases do not exist independently of the population from which they arise.
If cases are restricted to residents of a particular region during a specified period, the source population must be defined accordingly. The subsequent selection of controls from the source population must respect those boundaries and the sampling strategy.
Eligibility criteria applied to cases and controls should be coherent, but they need not be mechanically identical in every respect. Having the outcome is, of course, part of case eligibility, while control eligibility depends on the control-sampling scheme.
In risk-set sampling, for example, a person may be eligible to serve as a control at one time and subsequently become a case. The underlying principle is that the person was eligible and at risk at the time represented by that control observation.
This is why the relationship between control eligibility and the possibility of becoming a case matters when defining the study population.
Can Exclusion Criteria Introduce Bias?
Yes. Exclusion criteria should be based on scientific or operational reasons established before examining exposure-outcome associations.
Excluding people because their records are incomplete may seem reasonable, but missing information could be related to exposure, disease severity, healthcare access, or another variable affecting selection.
Similarly, excluding cases with comorbid conditions may produce a narrower population than the one addressed by the research question. Such restrictions can alter the association under investigation, particularly when the comorbidity is related to the exposure.
Researchers should distinguish genuine ineligibility from missing data or uncertain classification. These are different problems and may require different analytical responses.
Watch Out
Do not modify the case definition after examining which criteria produce a stronger exposure-outcome association. Outcome definitions chosen in response to observed results can introduce selective analysis and undermine the credibility of the findings. Any post hoc changes should be disclosed and distinguished from prespecified analyses.
How Should Researchers Handle Cases Identified at Multiple Sites?
Multisite recruitment can increase the number of eligible cases, particularly for rare outcomes, but differences in diagnostic procedures may introduce heterogeneity.
One hospital might rely on specialist assessment, another on laboratory confirmation, and a third on administrative diagnostic codes.
Before combining records, researchers should establish whether the methods identify sufficiently comparable outcomes.
Possible safeguards include a shared case-definition protocol, standardized abstraction forms, assessor training, duplicate-record checks, and independent adjudication of uncertain cases.
Where site differences cannot be eliminated, they should be documented and considered in the analysis. Site-specific sensitivity analyses may be useful when sample sizes permit.
Should Case Definitions Be Evaluated for Sensitivity and Specificity?
When a suitable reference standard exists, evaluating the sensitivity and specificity of a case-identification procedure can help researchers understand potential outcome misclassification.
Sensitivity describes the proportion of truly affected individuals correctly identified by the procedure. Specificity describes the proportion of unaffected individuals correctly classified as not meeting the outcome definition.
In practice, these properties may be unknown, especially when the outcome is defined through imperfect administrative records or when no definitive reference standard exists.
Researchers should avoid assuming that a diagnostic code, screening threshold, or self-reported condition has perfect classification accuracy.
The consequences of misclassification depend on how errors relate to exposure, outcome status, and sampling. It is not generally safe to assume that every measurement error simply weakens the observed association.
04 · A Practical Example
Developing a Case Definition for a Rare Neurological Condition
Hypothetical Example
From a Broad Hospital Search to a Verified Case Population
A research team investigates whether historical occupational exposure is associated with a specified neurological condition. The researchers obtain permission to screen records from several hospitals serving a defined geographical region.
An initial database query identifies 310 records containing relevant diagnostic terms. The researchers recognize that these records cannot automatically be treated as 310 eligible cases.
Step 1: Specify the target outcome.
The researchers adopt recognized diagnostic criteria appropriate to the condition and define the evidence required for confirmation. The precise criteria are recorded in the protocol.
Step 2: Establish person, place, and time eligibility.
Eligible cases must belong to the defined source population and have their first qualifying diagnosis during the prespecified observation period.
Step 3: Verify records systematically.
Trained reviewers assess diagnostic documentation using a standardized form. Duplicate records are identified, and uncertain classifications undergo a predefined adjudication procedure.
Step 4: Apply exclusions transparently.
Records are excluded when the individual does not meet the outcome criteria, falls outside the population or time boundaries, or meets a justified exclusion criterion. Reasons are documented.
Step 5: Finalize the eligible case population.
After verification, assume that 180 distinct individuals satisfy the complete definition. The other 130 initial records are not included, with each exclusion assigned a documented reason.
The final case population is not simply everyone whose record contained a relevant diagnostic term. It consists of individuals who satisfy a reproducible definition tied to the study question and source population.
Suppose 25 additional patients have suggestive clinical findings but insufficient evidence for confirmation. The researchers should not automatically classify them as either confirmed cases or unaffected controls. Depending on the protocol, they might be excluded from the primary confirmed-case analysis and considered separately in a prespecified sensitivity analysis.
The research report should explain how the initial 310 records were screened, why 130 did not qualify, and how the final 180 cases were established. These hypothetical counts illustrate case ascertainment, not a recommended exclusion rate.
The next methodological decision is whether the controls can appropriately represent the population that generated these 180 cases. A carefully verified case definition cannot compensate for a fundamentally unsuitable control group.
