Abstract: Background　Diabetes is a nutrition-related chronic disease with high prevalence and disability.Therefore，the ability of patients to master and apply nutritional knowledge（i.e.，nutritional literacy） is essential for the prevention and control of diseases.However，no effective evaluation tool has been found in China.Objective　To translate the English version of the Nutrition Literacy Assessment Instrument（Nlit） into Chinese and test its reliability and validity based on CTT and Rasch model.Methods　A total of 325 diabetic inpatients from Department of Endocrinology，Tianjin Medical University General Hospital from November 2018 to May 2019 were selected by convenience sampling，and were surveyed with the revised Chinese version of Nlit（CHI-NLit）.CTT and Rasch-based approaches were used to divide the dimension of the CHI-NLit，and to evaluate its internal consistency，content validity，structural validity and validity.Results　Principal component analysis of residuals showed that the CHI-NLit was not a unidimensional test，and the six subscales met the unidimensional criteria，so the six dimensions of the original scale were retained.After the fitting degree and difficulty analysis of Rasch model，items 26，29，31，32 that were not up to standard were removed，and a total of 38 items were retained to form the CHI-NLit.The item reliability and sample reliability of CHI-NLit were 0.919 and 0.838，respectively.The item reliability and personal reliability of the subscales were 0.928/0.681，0.902/0.690，0.948/0.679，0.932/0.651，0.960/0.601，0.928/0.683，respectively.The KR-21 reliability coefficients of the scale and subscales were 0.860，0.702，0.718，0.719，0.684，0.721，respectively.The test-retest reliability of total scale and subscales were 0.936，0.895，0.915，0.906，0.931，0.923，and 0.904，respectively.In terms of validity，the content validity index（CVI） of the scale was 0.905，and the item-level CVI（I-CVI） was 0.8-1.0.The fit of each item was in line with the Rasch model，with moderate difficulty and good structural validity.The Pearson correlation coefficient of criterion validity was 0.784.There was a linear positive correlation between the score of each subscale of NLit and the score of NVS.Analysis of ROC curve of CHI-NLit to predict nutritional literacy level showed that the area under the curve（AUC） was 0.885〔95%CI（0.846，0.924）〕.When the cut-off value was 21.5 points，the Youden index was greatest（0.623）.Conclusion　CHI-NLit has good reliability and validity，which may be used as an auxiliary tool for the measurement of nutritional literacy of diabetic patients.

Key words: Diabetes mellitus, Nutrition literacy, Classica test theory, Rasch model, Evaluation, Reliability, Validity, Scale

摘要： 背景　糖尿病是一种营养相关的慢性病，患病率和致残率均较高。因此患者对营养知识的掌握和运用能力（即营养素养）对预防和控制疾病至关重要，但国内目前尚未见有效评价工具。目的　基于经典测量理论（CTT）和项目反应理论（IRT）的Rasch分析对英文版营养素养评价工具（NLit）进行汉化并验证其信效度。方法　2018年11月—2019年5月采用方便抽样的方法选择在天津医科大学总医院内分泌科住院的325例糖尿病患者作为研究对象，采用经翻译、修订形成的中文版营养素养评价工具（CHI-NLit）进行问卷调查。结合CTT和Rasch模型分析两种测量方法对问卷进行维度划分以及内部一致性、内容效度、结构效度、效标关联效度评价。结果　残差主成分分析结果显示，CHI-NLit并非单维测验，6个分量表满足单维标准，因此保留原量表的6个维度。经Rasch模型拟合度和难度分析后，删除不达标的条目26、29、31、32，共保留38个条目形成CHI-NLit。总量表的项目信度、样本信度分别为0.919和0.838，分量表的项目信度/样本信度分别为0.928/0.681、0.902/0.690、0.948/0.679、0.932/0.651、0.960/0.601、0.928/0.683；总量表的库理信度（KR-21）为0.860，分量表的KR-21分别为0.702、0.718、0.719、0.694、0.688、0.721；总量表的重测信度为0.936，分量表重测信度分别为0.895、0.915、0.906、0.931、0.923、0.904。就效度而言，总量表内容效度指数（S-CVI）为0.905，各条目的内容效度指数（I-CVI）为0.8~1.0；各条目拟合符合Rasch模型，难度适中，结构效度较好，效标效度的Pearsen相关系数为0.784，CHI-NLit各维度得分与NVS得分呈线性正相关，绘制CHI-NLit量表预测营养素养水平的ROC曲线，ROC曲线下面积（AUC）为0.885〔95%CI（0.846，0.924）〕；当截断值为21.5分时，约登指数最大为0.623。结论　CHI-NLit具有良好的信效度，可以为糖尿病患者营养素养的测量提供参考。

关键词: 糖尿病, 营养素养, 经典测量理论, Rasch模型, 评价, 信度, 效度, 量表