研究目的 

本研究旨在利用项目反应理论简化《中国居民健康素养量表》，评估简化版量表的信度和效度，使之适用于社区老年人群健康素养的测量。同时，利用简化版量表，探讨服药依从性对社区患慢性病老年人群健康素养与健康结局（失能）关联的中介效应，以及认知对其关联的调节作用。

研究方法 

本研究利用2020年12月开展的一项针对北京市社区60岁及以上老年人群健康调查数据，该数据涵盖基本人口学信息、健康素养测量、服药依从性、失能状况评估和自我认知评估等方面内容。研究中，采用完全随机的方法，将由5829名调查对象的总样本拆分为两个子样本（简称“样本A”和“样本B”），样本A用于《中国居民健康素养量表》的简化，样本B用于简化版量表的信度和效度验证。量表简化阶段，基于项目反应理论，根据项目难度、区分度、信息量水平及项目功能差异删减不符合标准以及在教育水平和认知状况上存在不公平测验现象的项目。简化版量表验证阶段，在样本B中，采用Cronbach’s alpha和McDonald’ s omega评价量表及其子维度的信度；利用验证性因子分析检验评估量表的效度，使用多群组验证性因子分析检验不同性别、不同年龄、不同教育水平和不同认知状况下检验简化版量表效度的测量不变性。同时，采用网络分析的方法将简化版量表的内部项目进行了可视化，评估不同项目之间的相互关系以及每个项目在网络中的重要性程度。

同时，本研究在患有慢性病的老年人群中，以“失能”作为主要的健康结局指标，以简化版量表测量的“健康素养”作为暴露指标，以服药依从性作为中介变量，采用结构方程模型，通过Bootstrapping检验中介的方法，探讨了服药依从性在健康素养和失能的关联关系的中介效应，以及认知功能对其产生的调节作用。

研究结果 

本研究通过项目反应理论分析，在简化阶段共删除了23个不符合标准的项目，保留了27个项目形成最终的简化版量表。删除的项目中，3个项目区分度过低（α<0.5），1个项目的难度过低（b<-3.0），20个项目所提供的信息量水平过低（I(θ)≤2.0），3个项目在认知状况的不同分组上存在项目功能差异（p<0.05），1个项目在教育水平的不同分组上存在项目功能差异（p<0.05）。

在简化版量表的验证阶段，简化版量表的Cronbach’ s alpha和McDonald’ s omega均为0.87，三个子维度的信度也均在0.60及以上。模型拟合指标中，CFI、TLI均大于0.8，GFI大于0.9，RMSEA小于0.08，在拟合度良好的范围之内。多群组验证性因子分析的结果显示，比较不同性别、不同年龄分组、不同教育水平和不同认知状况人群，Dc2不存在显著性差异，且单位等值、尺度等值、残差等值模型与构形不变性模型（参照）相比，在CFI、TLI、GFI、RMSEA等主要模型拟合指标的差距都小于0.02，这些结果表明简化版量表在上述四个不同特征人群分组上均具备测量的不变性。敏感性分析中采用重抽样的方法，模拟教育水平分布、城乡分布和年龄分布不同的三类虚拟样本人群，分析评估不同简化版量表的信度、效度和不同人群中的测量不变形，均得到与样本B相似的结果，这在一定程度上反映了简化版量表在老年人群中可能具有一定的普适性。

利用老年人群健康调查数据，将简化版量表中各个项目采用网络分析法构建网络图，结果显示网络中共包含27个节点（即项目），231/251条有效边（均为正向相关），平均边权重为0.032；根据强度中心性指标，量表中三个项目（即：C01多选题“关于促进心理健康的方法，以下说法正确的是”，C04多选题“孩子出现发热、皮疹等症状，家长应该”，以及C08多选题“遇到呼吸、心跳骤停的伤病员，应采取哪些措施？”）在网络中的重要性程度最为突出。通过样本下降自助法计算强度中心性的相关稳定性系数为0.750，说明强度中心性指标对节点中心性的评估稳定性良好。

简化版量表的应用研究中，本研究结果表明在社区患有慢性病的老年人群中，健康素养越高，人群失能风险越低；控制混杂因素后，无论是在认知正常或是认知异常的老年人群中，服药依从性在健康素养与失能之间都起到了中介效应。

研究结论

本研究基于项目反应理论形成了适用于老年人群的《中国居民健康素养量表》简化版，该版本具有较好的信度和效度，在不同特征人群比较时具有一定的结构不变形，敏感性分析显示该量表可能具有一定人群普适性。简化版量表的应用研究中，对于患有慢性病的老年人群，无论是认知正常或是认知障碍，服药依从性在健康素养与失能之间都起到中介变量的作用。可见，简化版量表可以作为一个相对简洁而有效的工具，适用于老年人群健康素养水平的测量和健康结局的影响评价。

Objective

This study aims to apply item response theory to simplify the Chinese Resident Health Literacy Scale, then evaluate the reliability and validity of the simplified version for measuring health literacy among the elderly individuals. Additionally, using the simplified version of the scale, we will explore the mediating effect of medication adherence on the association between health literacy and health outcomes (disability) in the community-dwelling elderly population with chronic diseases, as well as the moderating role of cognition in this association.

Methods

This study used data from a health survey conducted in December 2020 among community-dwelling individuals aged 60 and above in Beijing, China. The dataset included basic demographic information, health literacy measurements, medication adherence, disability assessment, and self-perception evaluations. Employing a completely random approach, the total of 5,829 participants was split into two subsets (marked as "Sample A" and "Sample B"). Sample A was used for the simplification of the Chinese Resident Health Literacy Scale, while Sample B was tested for the reliability and validity verification of the simplified scale.

During the scale simplification phase, item selection was based on item response theory, considering criteria such as item difficulty, discrimination, information value, and item function differences. Items that did not meet the criteria and those exhibiting unfair testing characteristics concerning educational level and cognitive status were eliminated. In the scale validation phase, within Sample B, Cronbach's alpha and McDonald's omega were employed to assess the reliability of the scale and its sub-dimensions. Confirmatory factor analysis was conducted to evaluate the validity of the scale, and multi-group confirmatory factor analysis was utilized to examine measurement invariance concerning different genders, ages, educational levels, and cognitive statuses. Furthermore, network analysis was used to visualize the internal structure of the simplified scale, and assess the interrelationships among the items and the importance of each item within the network.

In addition, this study focuses on the elderly population with chronic diseases, using "disability" as the primary health outcome measure, the simplified version of the scale to measure "health literacy" as the exposure variable, and medication adherence as the mediating variable. Structural equation modeling is employed, and the mediating effect of medication adherence on the association between health literacy and disability is examined using the Bootstrapping method. Furthermore, the moderating role of cognitive function in this relationship is investigated.

Results

In the phase of simplifying the scale, a total of 23 items that did not meet the criteria were removed based on item response theory analysis. This resulted in a final simplified version of the scale comprising 27 items. Among the removed items, three had low discrimination (α<0.5), one had low difficulty (b<-3.0), 20 had low information value (I(θ)≤2.0), three exhibited item function differences across different cognitive status groups (p<0.05), and one showed item function differences across different educational level groups (p<0.05).

In the validation phase of the simplified scale, both Cronbach's alpha and McDonald's omega were found to be 0.87, indicating the good reliability in these simplified scales. The reliability of the three sub-dimensions was also 0.60 or higher. In terms of model fit indices, CFI and TLI were above 0.8, GFI was above 0.9, and RMSEA was below 0.08, all of which were within an acceptable range of good fit. The results of multi-group confirmatory factor analysis showed no significant differences in Δχ² when comparing different gender, age, educational level, and cognitive status groups. The measurement invariance models, including configural, metric, and scalar invariance, had differences in major model fit indices (CFI, TLI, GFI, RMSEA) smaller than 0.02 compared to the reference model. These results indicate that the simplified scale exhibited measurement invariance across the four different characteristic groups mentioned above. Sensitivity analysis using resampling methods simulated three virtual sample populations with different distributions of educational level, urban-rural distribution, and age distribution. In the sensitivity analysis, the resampling method was used to simulate three virtual samples with different education level distribution, urban-rural distribution and age distribution, in order to explore the reliability, validity, measurement invariance in virtual different populations. The similar results to those of Sample B were also found. This suggests that the simplified scale may have a certain degree of universality among the elderly population. Furthermore, the network analysis method was applied to construct a network graph for the items of the simplified scale. The results showed that the network consisted of 27 nodes (i.e., items), with 231 out of 251 edges being valid and positively correlated. The average edge weight was 0.032. Based on the measure of strength centrality, three items stood out as having the highest importance in the network: item C01 (multiple-choice question: "Which of the following statements is correct about promoting mental health?"), item C04 (multiple-choice question: "When a child has symptoms such as fever and rash, what should parents do?"), and item C08 (multiple-choice question: "What measures should be taken when encountering a casualty with respiratory and cardiac arrest?"). The stability coefficient of the measure of strength centrality, calculated using the bootstrap method, was 0.750, indicating good stability in evaluating the centrality of the nodes.

In the application study of the simplified version of the scale, the results of this research indicate that in the community-dwelling elderly population with chronic diseases, higher health literacy is associated with a lower risk of disability. After controlling for confounding factors, both in cognitively normal and cognitively impaired elderly populations, medication adherence mediates the relationship between health literacy and disability.

Conclusion

Based on item response theory, this study developed a simplified version of the Chinese Health Literacy Scale for Residents that might be more suitable for the elderly population with good reliability and validity. In the application study of the simplified version of the scale, for the elderly population with chronic diseases, both in cognitively normal and cognitively impaired individuals, medication adherence serves as a mediating variable between health literacy and disability.