INSERM, CIC-EC, CIE6, Service d'Épidémiologie et Évaluation Cliniques, Hôpitaux de Brabois, Allée du Morvan, 54500 Vandoeuvre-lès-Nancy, France; CHU Nancy, Epidémiologie et Evaluation Cliniques, Hôpitaux de Brabois, Allée du Morvan, 54500 Vandoeuvre-lès-Nancy, France; Université de Lorraine, Université Paris Descartes, Apemac, EA 4360, Faculté de Médecine, 9 avenue de la Forêt de Haye, 54505 Vandoeuvre-lès-Nancy, France.
Public Health Innovation, Population Health Strategic Research Centre, Deakin University, Burwood Campus, 221 Burwood Highway, Melbourne VIC 3125, Australia.
J Clin Epidemiol. 2015 Apr;68(4):360-9. doi: 10.1016/j.jclinepi.2013.07.013. Epub 2013 Sep 29.
To assess the contribution of back-translation and expert committee to the content and psychometric properties of a translated multidimensional questionnaire.
Recommendations for questionnaire translation include back-translation and expert committee, but their contribution to measurement properties is unknown. Four English to French translations of the Health Education Impact Questionnaire were generated with and without committee or back-translation. Face validity, acceptability, and structural properties were compared after random assignment to people with rheumatoid arthritis (N = 1,168), chronic renal failure (N = 2,368), and diabetes (N = 538). For face validity, 15 bilingual people compared translations quality with the original. Psychometric properties were examined using confirmatory factor analysis (metric and scalar invariance) and item response theory.
Qualitatively, there were five types of translation errors: style, intensity, frequency/time frame, breadth, and meaning. Bilingual assessors ranked best the translations with committee (P = 0.0026). All translations had good structural properties (root mean square error of approximation <0.05; comparative fit index [CFI], ≥0.899; and Tucker-Lewis index, ≥0.889). Full measurement invariance was observed between translations (ΔCFI ≤ 0.01) with metric invariance between translations and original (lowest ΔCFI = 0.022 between fully constrained models and models with free intercepts). Item characteristic curve analyses revealed no significant differences.
This is the first experimental evidence that back-translation has moderate impact, whereas expert committee helps to ensure accurate content.
评估回译和专家委员会对翻译后的多维问卷的内容和心理测量特性的贡献。
问卷翻译的建议包括回译和专家委员会，但它们对测量特性的贡献尚不清楚。使用和不使用委员会或回译生成了健康教育培训影响问卷的四个英语到法语的翻译版本。在将问卷随机分发给类风湿关节炎患者（N=1168）、慢性肾衰竭患者（N=2368）和糖尿病患者（N=538）后，比较了表面效度、可接受性和结构特性。对于表面效度，15 名双语者将翻译质量与原文进行了比较。使用验证性因子分析（度量和标度不变性）和项目反应理论检验心理测量特性。
定性分析发现有五种翻译错误类型：风格、强度、频率/时间框架、广度和含义。双语评估者对有委员会的翻译评价最高（P=0.0026）。所有翻译都具有良好的结构特性（近似均方根误差<0.05；比较拟合指数[CFI]，≥0.899；和 Tucker-Lewis 指数，≥0.889）。在翻译之间观察到完全测量不变性（ΔCFI≤0.01），并且在翻译和原始之间具有度量不变性（最低ΔCFI=0.022 在完全约束模型和具有自由截距的模型之间）。项目特征曲线分析显示没有显著差异。
这是第一个实验证据，表明回译具有中等影响，而专家委员会有助于确保内容准确。
