Operational Definitions

Overview

An operational definition states exactly how a study will recognise, measure, derive or classify an element of interest. It converts phrases such as postpartum follow-up, smoking exposure, treatment adherence, incident infection or clinical improvement into rules that another trained person could apply to the same source data.

A useful definition identifies the entity, observable evidence, data source or instrument, timing, calculation or classification rule, and handling of ambiguity. Operationalisation is therefore more than assigning a column name. It connects conceptual meaning to data collection, quality control and analysis [1,2].

Why this matters

Two studies can use the same label while measuring different quantities. “Readmission” may mean any hospital admission within 30 days, an unplanned admission to the same hospital, or a condition-specific admission identified across linked hospitals. “Postpartum screening completed” may include fasting glucose, an oral glucose-tolerance test or glycated haemoglobin measured within different windows.

These differences affect prevalence, group comparisons, eligibility, missingness and interpretation. A vague definition also permits undocumented decisions after researchers have seen the data, increasing the risk of inconsistent classification and selective analysis.

Anatomy of an operational definition

A complete definition usually contains the following elements.

Intended concept or event

State what is being represented and why it matters to the research question.

Unit

Specify whether the definition applies to a person, pregnancy, admission, clinic, specimen, visit, episode or another unit.

Observable evidence

Describe the finding that establishes presence, absence or value: a laboratory result, coded diagnosis, questionnaire response, observation, device reading or combination.

Source or instrument

Identify the record system, instrument version, assessor, laboratory method or data linkage used.

Timing and reference point

Define the time origin, allowed window, frequency and handling of repeated observations. “Within 12 weeks” is incomplete unless the starting event is clear.

Derivation or classification rule

Specify formulae, thresholds, hierarchy when sources disagree, and whether the variable is retained continuously or categorised.

Ambiguity and missingness rule

State how indeterminate, conflicting, implausible, unavailable and not-applicable values will be handled. Do not convert “not recorded” automatically into “no”.

Operational-definition template

illustrative

Table 1. Fields for an auditable operational definition

Field	Question to answer
Name and concept	What quantity, event or category is intended?
Unit	To whom or what does the value apply?
Evidence	What observation establishes the value?
Source	Where and how is the evidence obtained?
Time	Relative to which origin and within what window?
Rule	How are values calculated, combined or classified?
Exceptions	How are conflicts, repeats, missingness and uncertainty handled?
Version	Which protocol, instrument, code list or algorithm version applies?

Defining the population operationally

Population descriptions require executable eligibility rules.

“Women with gestational diabetes” could mean:

• a diagnosis recorded in the maternity system;
• glucose values meeting specified diagnostic criteria;
• treatment for gestational diabetes;
• inclusion in a specialist clinic register; or
• any of these through a prespecified algorithm.

The target population may be broad, but the study population must be identified through available evidence. The definition should specify pregnancy dates, diagnostic period, age range, residency or service area, multiple pregnancies, previous diabetes, pregnancy outcome and the handling of transfers.

Eligibility criteria should not include unnecessary exclusions merely to create a convenient homogeneous sample. Each exclusion should protect safety, feasibility or interpretability.

Defining exposures and interventions

An exposure definition should specify intensity, duration, timing and reference category where relevant. “Smoker” is ambiguous without distinguishing current, former and never smoking, defining recency and deciding how occasional use is treated.

An intervention definition should identify:

• content and intended mechanism;
• provider and delivery mode;
• dose, frequency and duration;
• start and allowable deviations;
• co-interventions;
• fidelity or receipt measures; and
• the comparison condition.

Assignment, receipt and adherence are different variables. A participant can be assigned an intervention but not receive it, or receive it incompletely. In trials, the treatment effect of interest must be defined in relation to events such as discontinuation, switching and rescue treatment. The ICH E9(R1) estimand framework emphasises aligning the clinical question, treatment conditions, population, variable, intercurrent-event strategy and population-level summary with the analysis [3].

Defining outcomes and events

A primary outcome definition should specify:

• domain or event;
• exact measure;
• metric, such as final value, change, time to event or proportion;
• method of aggregation;
• assessment time point or window; and
• ascertainment and adjudication process.

For events such as infection, relapse or readmission, specify whether the first event or all events count, how closely spaced episodes are separated, and how competing events are handled.

Defining time

Time definitions require a clear origin. Examples include date of randomisation, treatment initiation, hospital discharge, delivery, diagnosis or first eligible visit.

Specify:

• whether boundaries are inclusive;
• calendar versus elapsed time;
• permitted visit windows;
• rule when several measurements fall in the window;
• rule for measurements outside the window;
• censoring or end of observation; and
• whether time-dependent status can change.

“Three-month outcome” might mean the closest measurement from day 75 to day 105, exactly 90 days, or any assessment during the third calendar month. The choice can change who has observed data.

Thresholds and categorisation

A threshold may come from clinical guidance, a validated instrument, a biological limit, policy, or a prespecified decision requirement. It should not be selected after examining which cut-off gives the most favourable result.

Whenever possible, preserve the original continuous value even when a category is needed. Categorisation discards information and may create artificial discontinuities [1]. Record both the measurement and the reasoned category where feasible.

Distinguish:

• a diagnostic threshold used for individual clinical decisions;
• a screening threshold prioritising sensitivity or another purpose;
• a research category used for description;
• a model-derived threshold requiring external validation; and
• an arbitrary cut-point with no established interpretation.

Composite variables and algorithms

A composite combines several observations into one value. Examples include multimorbidity indices, composite cardiovascular outcomes and algorithms identifying disease from coded records.

Specify:

• included components;
• whether any component is sufficient;
• weighting;
• required number of non-missing components;
• hierarchy when components conflict;
• time relationship among components;
• version of code lists; and
• whether component results will also be reported.

A composite may increase event frequency but can obscure differences among components with unequal clinical importance. Its interpretation should match what the components jointly represent.

Operational definition for postpartum testing completion

illustrative

Table 2. Example primary-outcome specification

Element	Prespecified rule
Unit	One eligible pregnancy ending in a live birth
Time origin	Date of delivery
Window	Day 28 through day 112 after delivery, inclusive
Accepted evidence	Laboratory-recorded 75-g oral glucose-tolerance test with fasting and 2-hour values
Sources	Linked hospital and participating primary-care laboratory systems
Repeated tests	Earliest complete eligible test in the window
Self-report only	Recorded separately; not classified as laboratory-confirmed completion
No linked record	Missing ascertainment unless the linkage process confirms complete coverage

Note: This definition is illustrative and is not presented as a clinical recommendation.

Repeated observations and source precedence

Operational definitions become more difficult when several values exist for the same unit. The protocol should distinguish a genuine change over time from duplicate recording, technical repetition and conflicting sources.

For repeated laboratory measurements, specify whether the study uses the first eligible result, the closest result to a target time, the worst value, an average, or all values in a longitudinal structure. Each rule answers a different question. Selecting the worst value may be reasonable for safety surveillance but inappropriate for estimating a typical level. Averaging repeated values may improve precision but can obscure clinically important peaks.

When sources conflict, use a prespecified hierarchy. For example, a time-stamped laboratory result may take precedence over a clinic note stating that a test was completed. A discharge summary may be preferred over a billing code for treatment assignment. The hierarchy should reflect validity and proximity to the event, not whichever source produces the desired classification.

A complete rule should also distinguish:

• duplicate copies of one event from separate events;
• corrected results from original results;
• preliminary from final laboratory reports;
• patient report from verified documentation;
• values recorded in different units; and
• data imported after the analysis cut-off.

Documenting these decisions in a data dictionary allows programmers, abstractors and investigators to reproduce the same derived variable. For manually abstracted outcomes, training cases and duplicate review of a sample can identify ambiguity before the main extraction is complete.

Clinical labels, case definitions and research definitions

A clinical diagnosis supports care for an individual. A surveillance case definition standardises population monitoring. A research operational definition serves a specified study question. These may overlap but have different purposes and acceptable trade-offs.

A research team should not silently redefine a standard clinical term. Where a study-specific rule is necessary, state it explicitly and explain how it differs from clinical or surveillance use.

Developing and testing definitions

Operational definitions should be drafted during protocol development and tested against realistic cases.

PROTOCOL CHECKLIST

Test each major operational definition

• Apply the rule to clear positive and clear negative examples.
• Include borderline, conflicting and missing cases.
• Check whether independent reviewers reach the same classification.
• Verify that required data are actually available at the stated time and source.
• Examine whether the rule disadvantages particular settings or participant groups.
• Record changes through version control before the main analysis.
• Preserve an audit trail from source data to derived variables.

Pilot testing may reveal that a source is incomplete or a time window unrealistic. A protocol amendment can then revise the rule transparently. After outcome data are examined, changes require particular caution and should be reported as post hoc where applicable.

Maintaining a version-controlled data dictionary

Operational definitions should be stored in a data dictionary rather than scattered across the protocol, questionnaire and analysis code. The dictionary should record the variable name, label, type, permitted values, units, source, derivation, missing-value codes, validation checks and current version. A change log should state what changed, why, when and who approved it.

The final manuscript may summarise definitions, but the complete dictionary supports reproducibility and future reuse. It also supports independent verification of derived outcomes and consistent implementation when several analysts or sites contribute data. When routine-data codes or external classifications change over time, preserve the version that applied to each observation and assess whether harmonisation is required.

Ethical and governance considerations

Operational definitions often determine who is labelled as diseased, adherent, high risk or eligible. Misclassification can stigmatise participants or affect clinical follow-up if research data enter care pathways. Researchers should minimise unnecessary sensitive classification, restrict access to identifiable source data and specify whether research classifications will be communicated clinically.

Interface with statistical analysis

The operational definition determines the analysis variable and estimand. It affects denominator, event count, missingness, timing and comparability.

The analysis plan should be able to answer:

• What exactly is estimated?
• Who is included in the denominator?
• Which observations count?
• How are repeated or competing events handled?
• What does a missing value represent?
• Are alternative plausible definitions examined as sensitivity analyses?

Sensitivity analysis should vary a meaningful assumption while targeting the same question, rather than presenting unrelated alternative outcomes as if they were equivalent [3].

Misclassification and estimand driftMMed extension

A change in operational definition can change more than measurement accuracy. It may change the target quantity itself. For example, expanding an outcome from laboratory-confirmed testing to any self-reported testing changes both ascertainment and interpretation. Sensitivity analyses should distinguish uncertainty about measurement from a deliberate change in the estimand.

Worked example

WORKED EXAMPLE

Revising a vague outcome definition

Vague version: Postpartum screening completed within three months.

Problems: The time origin, exact tests, boundary dates, data sources, repeated tests and unverified self-report are unspecified.

Revised version: For each eligible pregnancy, laboratory-confirmed completion is defined as a complete 75-g oral glucose-tolerance test recorded in linked participating laboratory systems from day 28 through day 112 after delivery, inclusive. If more than one eligible test is recorded, the earliest complete test is used. Self-reported testing without a retrievable laboratory result is described separately and is not classified as laboratory-confirmed completion.

Remaining limitation: Testing outside participating systems may be missed. The study must assess coverage and avoid interpreting an absent record as definite non-completion where linkage is incomplete.

Suggested protocol or reporting wording

This record is especially important when coding is performed in batches or repeated after database updates.

Common mistakes

Knowledge check

Why is 'hypertension was present' not a complete operational definition?

Answer and reasoning

It does not state the unit, measurement procedure, number and timing of readings, device requirements, threshold, medication rule or data source. Different assessors could classify the same participant differently.

A routine dataset has no smoking code for 40% of patients. Can these patients be classified as non-smokers?

Answer and reasoning

Not without evidence that absence of a code reliably means never smoking. The value may be unrecorded, unknown or not assessed. The operational definition must distinguish these states.

What is the problem with choosing the outcome cut-off that produces the smallest p-value?

Answer and reasoning

The cut-off becomes outcome-driven and can exaggerate evidence. Thresholds should be justified and prespecified; exploratory alternatives should be labelled and interpreted cautiously.

Why retain component outcomes when using a composite outcome?

Answer and reasoning

Components may differ in frequency, severity and intervention effect. Reporting components helps readers understand what drove the composite and whether its overall interpretation is clinically coherent.

Key take-home messages

SUMMARY

Key take-home messages

• An operational definition is an executable rule linking a concept to observable evidence.
• Specify the unit, source, timing, derivation, thresholds and handling of ambiguity.
• The same label can represent different quantities across studies.
• Preserve original measurements and audit trails whenever feasible.
• Definition changes can alter both measurement and the target question.

References

1. Andrade C. A student’s guide to the classification and operationalization of variables in the conceptualization and design of a clinical study: part 2. Indian J Psychol Med. 2021;43(3):265–268. https://doi.org/10.1177/0253717621996151
2. Vandenbroucke JP, von Elm E, Altman DG, Gøtzsche PC, Mulrow CD, Pocock SJ, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007;4(10):e297. https://doi.org/10.1371/journal.pmed.0040297
3. International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. ICH E9(R1): addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials. Geneva: ICH; 2019. https://database.ich.org/sites/default/files/E9-R1_Step4_Guideline_2019_1203.pdf