Translating a patient-reported outcome measure

Translating a patient-reported outcome measure requires culturally appropriate language that preserves the concepts being measured across populations. Linguistic validation combines translation, review and testing with the intended audience.

Conceptual equivalence and cultural relevance

Patient-reported outcome (PRO) measures must use non-technical language understood by diverse lay populations, including people with limited education or literacy, and must be interpreted consistently enough to yield meaningful data. ¹ Multilingual, multinational trials add the requirement that often subjective concepts be understandable within each diverse local population and consistent across languages. ^{1, 2}

Translators must preserve the content and register of everyday expressions through potentially very different everyday expressions in another language. ³ The balance between cultural adaptation and conceptual equivalence makes PRO translation a distinct medical-translation specialism, whose quality depends on both the validation process and translator competencies. ^{1, 3}

Linguistic validation translates a PRO measure for a new language or culture while preserving conceptual, item, semantic and operational equivalence between versions. ⁴ Its aims include validity and integrity of data gathered across languages and cultures, ethical considerations, regulatory compliance and improved patient-centred outcomes. ⁴

Regulatory and industry guidance

In 2018, European Medicines Agency (EMA) and US Food and Drug Administration (FDA) guidance emphasised evaluating the translation process to assess whether content validity was maintained across languages, because translation and cultural adaptation affect primary- and secondary-endpoint data collection; neither agency specified a detailed translation process. ² EMA guidance from 2005 and FDA guidance from 2009 requested process details, descriptions of patient testing, language- or culture-specific concerns, reasons for new versions, translated or adapted versions, and evidence of comparable content validity and measurement properties. ⁵

The International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Taskforce for Translation and Cultural Adaptation published best practices in 2005, followed by recommended minimum standards from the International Society for Quality of Life Research (ISOQOL) in 2013. ^{2, 5} Linguistic-validation workflows reflect these best practices, with variations among questionnaire developers and translation companies. ^{2, 6} A 2024 practitioner assessment judged that these guidelines still consistently yielded high-quality results when correctly implemented. ⁵

Both sets of recommendations emphasise qualitative assessment, usually cognitive interviewing or debriefing, to check understanding, accurate interpretation and cultural relevance. ² A 2018 description allowed testing with patients from the target population or alternatively laypeople, whereas a 2024 description stated that cognitive debriefing with target-population patients was required. ^{2, 5} Maintaining an audit trail of every linguistic step, edit and decision is important throughout the project. ⁵

A seven-stage validation model

A seven-stage linguistic-validation model described in 2018 includes the following safeguards. ⁶

1. Explanation of concepts. The questionnaire developer or translation company prepares a document during or after measure development to clarify the intended meaning of key source concepts, providing a reference throughout validation.
2. Forward translation and reconciliation. Two translators independently translate the source, then collaborate on one reconciled version, checking their interpretations and selecting accurate, appropriate wording.
3. Back translation and review. A third independent translator translates the reconciled text back into English or the other source language. Comparison with the source identifies conceptual discrepancies, which are resolved through team discussion, forward-translation revision and back-translation of revised wording.
4. International harmonisation. Back-translations from all project languages are reviewed together for conceptual discrepancies and a coherent approach, and developer guidance may be sought for problematic concepts.
5. Expert reviews. Target-country experts, such as clinicians or client or developer subsidiaries, may review drafts for alignment with local usage.
6. Cognitive debriefing. Target-country interviewers test drafts with diverse laypeople or patients for misunderstood or culturally irrelevant concepts. Highly problematic concepts may be revised through translator discussion and a process like step 3, considering problems found in other languages.
7. Proofreading. A translator proofreads the final product for errors.

^{6, 7}

A six-stage validation model

A 2024 workflow groups the process into six stages. ^{5, 8}

1. Concept elaboration. Explain the PRO measure's health-related concepts as a roadmap for linguists.
2. Forward translation or adaptation. Translate from source to target language, or adapt for the same language in a different country.
3. Reconciliation. Combine the best of the two forward versions through discussion between project managers and linguists.
4. Back translation and review. Use single or dual back translation, reviewed by linguistic-validation project managers, with developer, author or clinician review where required.
5. Cognitive debriefing. Conduct one-to-one interviews with target-population patients and report the results.
6. Quality assurance. Proofread and check formatting and quality, perform a final check and issue a translation certificate.

⁸

Providers differ in how they execute particular steps, including single or dual back translation and virtual or face-to-face cognitive debriefing. ⁵

Cultural adaptation within the measured concept

Forward translators first decide which concepts can be translated literally and which need cultural adaptation while preparing their independent and reconciled versions. ⁹ Adaptations range from different structures or idioms with similar meaning to alternative examples of sports or food, but must maintain the integrity of the concept being measured. ⁹ Cognitive debriefing allows target-country participants to judge cultural relevance and suggest better adaptations. ⁹

Cultural adaptation must not compromise conceptual validity across trial languages through misinterpretation or over-adaptation. ⁷ A discrepancy found through back-translation review may be resolved readily when it reflects a misinterpretation, or may require discussion to find wording that preserves the concept without unduly sacrificing cultural relevance. ⁷ Cognitive debriefing can evaluate that balance and helps identify necessary revisions. ⁷

Translatability during measure development

Clear, discrete concepts in the source questionnaire directly affect how closely conceptual equivalence can be maintained across languages. ¹⁰ Avoiding compound and overlapping concepts can improve translatability. ^{10, 11, 12}

Compound concepts can describe several symptoms or experiences at once, allowing source-language respondents to emphasise different elements. ¹¹ Other languages may group those elements differently, making it hard to preserve the complexity without introducing or removing concepts or shifting emphasis; fatigue, bother, frustration and distress should therefore be avoided where possible. ¹¹

Response sets are integral to data collection, but languages distinguish gradations of amount or degree differently. ¹² Distinct response concepts are preferable where possible: “None of the time / A little of the time / Some of the time / A lot of the time / Most of the time / All of the time” should be replaced by “Not at all / Rarely / Sometimes / Often / Always”. ¹² Where distinct concepts are unavailable, a numeric rating scale can replace overlapping gradations with clearly differentiated numerical responses. ¹²

Measures should undergo face validation, translatability assessment or both during development: an expert checks that concepts are clear, discrete and unambiguous in face validation, while translators identify translation difficulties in translatability assessment. ¹³ Problematic items can then be referred to the developer for revision. ¹³ Earlier attention produces less ambiguous, less culturally specific source text, easing consistent translation and cultural adaptation and leading to fewer translation delays and higher-quality trial data. ¹³

Technology and human oversight

In 2024, linguistic validation was grounded in human translation and review, making artificial intelligence (AI) integration more challenging than in other localisation fields. ¹⁴ Every stage of the six-stage model involved people, including ISO-qualified linguists experienced in medical translation, trained validation project managers, clinicians and patients in the therapeutic area, and instrument developers with advanced degrees in psychology, outcomes research or related fields. ^{5, 8} Replacing their collective expertise with technology was considered inadvisable given the complexity of cultural and conceptual equivalence. ⁸

In 2024, existing guidelines did not address translation technology, and no linguistic-validation-specific guidance advised replacing human translation, review or testing with fully automated translation. ^{8, 15, 16} Potential alternatives included humans correcting machine output or comparing machine and human versions. ¹⁶ Measure-specific memory banks drawn from ongoing and historical translations could inform AI systems, but would require substantial setup, maintenance and potentially lengthy approvals. ¹⁶

PRO instruments are short, with a 50-word measure illustrating the contrast with the vast datasets AI models need, and instrument ownership can be complex. ¹⁶ PRO data can be proprietary and sensitive, requiring respect for ownership and ethical standards when training models. ¹⁶ The 2024 advice requires disabling public term-bank features in computer-assisted translation (CAT) tools for linguistic validation because the origins of their translations may be uncertain. ¹⁶

Health concepts, idioms and cultural subtleties require human review, translation and testing; a 2024 practitioner assessment considered AI weaker than humans in contextual and cultural understanding. ¹⁶ Confusing, isolating, offensive or discriminatory language can prevent patient participation, while poor translations that damage measurement properties can prevent evidence informing clinical or health-policy decisions. ^{16, 17} Other barriers include regulatory compliance, accuracy and bias concerns, organisational risk aversion, data security, intellectual property, privacy and silos between linguistic-validation, scientific and technology teams. ¹⁶

Responsible AI innovation should be led collaboratively by qualified experts, consider lessons from other localisation fields and combine rigorous testing with human review and validation rather than focus solely on time and cost. ¹⁷ Bespoke algorithms and early human review of PRO content could provide high-quality training data, and translatability assurance during new-measure development could extend to assessing AI compatibility. ¹⁸ Type-ahead, readability highlighting, offensive-term detection, autocorrect and compliance with glossaries and style guides offer potential benefits, but context, cultural subtlety, domain knowledge and bias remain concerns, requiring human oversight and stringent quality-control frameworks for AI training. ¹⁸