Principles of Measurement in Health Research

Overview

Measurement is the disciplined process of assigning observations to a health characteristic, behaviour, event or construct according to stated rules. The process is broader than choosing a questionnaire or laboratory test. It begins by defining what is to be measured, continues through the instrument and procedure used to obtain the observation, and ends with a value or category that can support the intended research inference [1–3].

A measurement target is the phenomenon or attribute of interest, such as postpartum testing completion, perceived access burden, depressive symptom severity or systolic blood pressure. An indicator is observable evidence about that target. An instrument is the tool used to obtain the evidence. A measurement procedure specifies how, when, where and by whom the instrument is applied. The resulting observation may then be coded, combined or transformed into a score.

Measurement quality therefore depends on the entire chain. A well-designed instrument used with an inconsistent procedure can produce poor data. A precise laboratory analyser cannot rescue an inappropriate specimen or an incorrect time point. A lengthy questionnaire cannot compensate for an unclear construct.

Why this matters

Poor measurement can make a well-designed study incapable of answering its question. It can obscure genuine differences, create artificial associations, misclassify participants, reduce statistical precision and limit comparability with other studies. Measurement problems also have ethical consequences: participants may contribute time, discomfort or personal information to research that cannot generate interpretable evidence.

Measurement decisions should be made at protocol stage. They determine eligibility, outcomes, exposure definitions, follow-up schedules, data-collection burden, sample-size requirements and the eventual analysis. Statistical methods quantify patterns in recorded values; they cannot reconstruct a target that was never measured adequately.

The measurement chain

A defensible measurement can be represented as a sequence:

1. define the target and intended interpretation;
2. identify observable indicators;
3. select or develop an instrument;
4. standardise administration and recording;
5. derive the value or score using prespecified rules;
6. evaluate measurement properties in the intended context; and
7. interpret the result within its known limitations.

Illustrative measurement specification

illustrative

Table 1. Measurement plan for selected study elements

Target	Indicator and source	Procedure	Recorded value
Test completion	Eligible laboratory result in linked records	Search prespecified systems from 4 to 16 weeks postpartum	completed/not completed/uncertain
Access burden	Participant experiences of arranging and attending testing	Interviewer-administered items at 12 weeks postpartum	item responses and prespecified domain score
Glycaemic status	Plasma glucose concentration	Accredited laboratory method using an eligible specimen	mmol/L and test type

Note: The definitions and time windows are illustrative rather than clinical recommendations.

What can serve as a measurement instrument?

Health research uses several broad forms of measurement.

Patient-reported measures

Regulatory and professional measurement guidance emphasises that patient-reported measures should be linked to a clearly defined concept of interest and context of use [5,6]. A patient-reported outcome or experience is provided directly by the participant without interpretation by a clinician or another observer. Symptoms, perceived functioning, treatment burden and quality of life often require patient report because no external observer has direct access to the experience. The item wording, recall period, language and response options become part of the instrument.

Clinician-reported and observer-reported measures

A clinician or trained observer may classify physical signs, imaging findings, behaviour or functional performance. The result depends on operational criteria, training, blinding, access to contextual information and how disagreement is handled. A clinical judgement is not automatically more objective than a participant report; it measures a different perspective.

Performance-based measures

Performance tests ask a participant to complete a standardised task, such as walking a fixed distance, recalling information or demonstrating a procedure. Motivation, instructions, equipment, practice effects and environmental conditions may influence the result.

Laboratory and device-based measures

Biochemical assays, physiological monitors and imaging systems can appear highly objective, yet they remain vulnerable to pre-analytical, analytical and post-analytical variation. Specimen collection, transport, calibration, algorithm version, operator technique and reporting thresholds must be controlled.

Record-based measures

Administrative, clinical and registry records may offer efficient measurement of diagnoses, treatments and service use. Their validity depends on why the data were originally recorded, completeness across care settings, coding rules and changes in information systems. Absence of a record is not always absence of the event.

DECISION GUIDE

Choose the measurement perspective

1. Who has direct access to the target phenomenon?
2. Is the purpose to capture experience, clinical judgement, performance, biology or service documentation?
3. At what time point and over what recall period should the target be represented?
4. Which source is least vulnerable to missingness or systematic distortion?
5. Would combining sources resolve uncertainty or create incompatible definitions?

Standardisation and calibration

Standardisation means applying the same documented procedure across participants, observers, sites and time points unless a planned difference is part of the design. It includes instructions, equipment, environmental conditions, order of procedures, permissible prompts, scoring rules and data-entry conventions.

Calibration compares an instrument with a reference or known standard and adjusts or documents its performance. Calibration may apply to laboratory analysers, scales, blood-pressure devices and other equipment. Human observers also require standardisation through training materials, practice cases and periodic quality checks.

Standardisation does not require ignoring legitimate adaptations. A participant with visual impairment may need an accessible administration mode. The protocol should preserve the intended construct while documenting the accommodation and considering whether mode affects comparability.

Precision, reliability, validity and agreement

These concepts answer different questions.

• Reliability concerns the proportion of observed variation attributable to differences between people rather than measurement error.
• Agreement concerns how close repeated values or ratings are in the original units or categories.
• Validity concerns whether evidence supports the intended interpretation and use of the measurement.
• Precision describes the degree of random variability in repeated measurement or estimation.

A measure can be reliable but invalid. A scale that consistently overstates body weight may rank participants similarly each time while providing inaccurate values. Conversely, a measure may be valid for group-level research but too imprecise for individual clinical decisions. COSMIN terminology and practical scale guidance separate validity, reliability and responsiveness because none can substitute for the others [2–4].

Responsiveness and interpretability

Responsiveness is the ability of a measurement to detect change over time in the construct intended to be measured [2]. It matters when the research question concerns improvement, deterioration or treatment response. A measure designed to distinguish people at one time point may contain too much random error or too few response levels to detect meaningful within-person change.

Interpretability is the degree to which qualitative meaning can be assigned to scores or changes. Researchers should know the possible range, direction, reference values, floor and ceiling effects, and what magnitude of difference is meaningful for the intended context. Interpretability is important but is not itself proof of validity.

Reflective indicators and causal indicatorsMMed extension

When items are treated as reflective indicators, the construct is assumed to cause the item responses; related items should therefore covary. When indicators jointly form a construct, as travel time, cost and appointment inflexibility may form access burden, high internal correlations are not necessarily expected. The measurement model should follow the construct definition rather than being selected from a convenient statistical rule.

Designing a measurement specification

A protocol should specify more than the instrument name.

PROTOCOL CHECKLIST

Specify each important measurement

• construct or variable and its role in the research question;
• operational definition and unit of observation;
• instrument, source and version;
• respondent or observer;
• language and administration mode;
• time point and recall period;
• equipment, preparation and environmental conditions;
• scoring, coding and derivation rules;
• handling of uncertain, missing or out-of-range values;
• training, calibration and quality-control procedures; and
• evidence supporting use in the target population and purpose.

Ethical and participant-centred measurement

Measurement burden should be proportionate to its contribution. Duplicate questionnaires, repeated blood sampling and long assessments can increase fatigue, non-response and withdrawal. Sensitive questions require a clear rationale, privacy safeguards and an appropriate response pathway when answers reveal risk or unmet need.

Interface with statistical analysis

The analysis plan needs the measurement scale, unit, timing, direction and derivation rules. It must also recognise repeated measurements, clustering, lower or upper detection limits, and whether a score is directly observed or constructed from items. Categorising a continuous measurement can discard information, while treating an ordinal category as if equal differences were guaranteed may impose an unsupported interpretation.

Detailed statistical analysis belongs in the Biostatistics series. The methodological responsibility is to ensure that the recorded variable represents the intended target and that its error structure is understood before modelling.

Worked example

WORKED EXAMPLE

Replacing a self-reported outcome with a source-integrated definition

Initial proposal: Ask each participant, “Did you complete your postpartum glucose test?”

Problems identified: Participants may confuse antenatal and postpartum tests, may not know the test name, and may have completed testing in another facility. The electronic record may also be incomplete across systems.

Revised measurement target: Completion of an eligible postpartum glucose test within the prespecified study window.

Procedure: Participants are asked about attendance, date and facility using a structured prompt. The research team searches linked laboratory and clinical systems. Results are classified as confirmed complete, confirmed incomplete or unresolved according to prespecified rules. The primary analysis uses the confirmed classification, with an uncertainty analysis planned for unresolved cases.

Interpretation: The revised procedure is more burdensome than one question, but it represents the target more defensibly and makes uncertainty visible rather than forcing every participant into a binary category.

Suggested protocol or reporting wording

Common mistakes

Knowledge check

What is the difference between a measurement target and an instrument?

Answer and reasoning

The target is the health attribute, event or construct the study intends to represent. The instrument is the tool used to obtain observable evidence about that target. A questionnaire may be an instrument for measuring access burden, but it is not the construct itself.

Why can a reliable measure still be unsuitable?

Answer and reasoning

Reliability concerns consistency or relative differentiation, not whether the measure represents the intended construct accurately. A consistently biased device or a stable questionnaire measuring the wrong domain may be reliable but invalid for the proposed use.

A clinic record contains no postpartum test result. Can the participant be classified automatically as not tested?

Answer and reasoning

Not necessarily. The test may have occurred outside the accessible record system or may have been recorded under another pathway. The protocol should define source coverage and an uncertainty category or verification process before interpreting absence of documentation as absence of testing.

Which feature becomes particularly important when the study aims to evaluate improvement over time?

Answer and reasoning

Responsiveness becomes central because the instrument must detect relevant change in the intended construct. Reliability and validity remain necessary, but a measure that cannot register change is poorly suited to a longitudinal response question.

Key take-home messages

SUMMARY

Key take-home messages

• Measurement is a chain linking a target, indicators, an instrument, a procedure and an interpretable value.
• Define what is being measured before selecting the measurement tool.
• Standardisation, calibration and quality control apply to questionnaires, observers, devices, laboratories and records.
• Reliability, agreement, validity and responsiveness answer different questions.
• Measurement burden and privacy risk should be justified by a clear research purpose.

References

1. de Vet HCW, Terwee CB, Mokkink LB, Knol DL. Measurement in medicine: a practical guide. Cambridge: Cambridge University Press; 2011. https://doi.org/10.1017/CBO9780511996214
2. Mokkink LB, Terwee CB, Patrick DL, Alonso J, Stratford PW, Knol DL, et al. The COSMIN study reached international consensus on taxonomy, terminology, and definitions of measurement properties for health-related patient-reported outcomes. J Clin Epidemiol. 2010;63(7):737–745. https://doi.org/10.1016/j.jclinepi.2010.02.006
3. Mokkink LB, Prinsen CAC, Patrick DL, Alonso J, Bouter LM, de Vet HCW, et al. COSMIN manual for systematic reviews of PROMs. Version 2.0. Amsterdam: COSMIN; 2024. https://www.cosmin.nl/wp-content/uploads/COSMIN-manual-V2_final.pdf
4. Streiner DL, Norman GR, Cairney J. Health measurement scales: a practical guide to their development and use. 5th ed. Oxford: Oxford University Press; 2015. https://doi.org/10.1093/med/9780199685219.001.0001
5. US Food and Drug Administration. Patient-focused drug development: selecting, developing, or modifying fit-for-purpose clinical outcome assessments. Silver Spring: FDA; 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/patient-focused-drug-development-selecting-developing-or-modifying-fit-purpose-clinical-outcome
6. American Educational Research Association, American Psychological Association, National Council on Measurement in Education. Standards for educational and psychological testing. Washington, DC: American Educational Research Association; 2014.