Haggerty Jeannie L, Bouharaoui Fatima, Santor Darcy A
Department of Family Medicine, McGill University, Montreal, QC.
Healthc Policy. 2011 Dec;7(Spec Issue):47-65.
Evaluating the extent to which groups or subgroups of individuals differ with respect to primary healthcare experience depends on first ruling out the possibility of bias.
To determine whether item or subscale performance differs systematically between French/English, high/low education subgroups and urban/rural residency.
A sample of 645 adult users balanced by French/English language (in Quebec and Nova Scotia, respectively), high/low education and urban/rural residency responded to six validated instruments: the Primary Care Assessment Survey (PCAS); the Primary Care Assessment Tool - Short Form (PCAT-S); the Components of Primary Care Index (CPCI); the first version of the EUROPEP (EUROPEP-I); the Interpersonal Processes of Care Survey, version II (IPC-II); and part of the Veterans Affairs National Outpatient Customer Satisfaction Survey (VANOCSS). We normalized subscale scores to a 0-to-10 scale and tested for between-group differences using ANOVA tests. We used a parametric item response model to test for differences between subgroups in item discriminability and item difficulty. We re-examined group differences after removing items with differential item functioning.
Experience of care was assessed more positively in the English-speaking (Nova Scotia) than in the French-speaking (Quebec) respondents. We found differential English/French item functioning in 48% of the 153 items: discriminability in 20% and differential difficulty in 28%. English items were more discriminating generally than the French. Removing problematic items did not change the differences in French/English assessments. Differential item functioning by high/low education status affected 27% of items, with items being generally more discriminating in high-education groups. Between-group comparisons were unchanged. In contrast, only 9% of items showed differential item functioning by geography, affecting principally the accessibility attribute. Removing problematic items reversed a previously non-significant finding, revealing poorer first-contact access in rural than in urban areas.
Differential item functioning does not bias or invalidate French/English comparisons on subscales, but additional development is required to make French and English items equivalent. These instruments are relatively robust by educational status and geography, but results suggest potential differences in the underlying construct in low-education and rural respondents.
评估不同群体或亚群体在初级医疗保健体验方面的差异程度，首先取决于排除偏差的可能性。
确定项目或分量表的表现是否在法语/英语、高/低教育程度亚群体以及城市/农村居住情况之间存在系统性差异。
选取645名成年使用者作为样本，根据法语/英语语言（分别来自魁北克和新斯科舍）、高/低教育程度以及城市/农村居住情况进行均衡分组，他们对六种经过验证的工具进行了回答：初级保健评估调查（PCAS）；初级保健评估工具简表（PCAT-S）；初级保健指数组成部分（CPCI）；欧洲初级保健患者满意度调查问卷第一版（EUROPEP-I）；人际关怀过程调查第二版（IPC-II）；以及退伍军人事务部全国门诊患者满意度调查的一部分（VANOCSS）。我们将分量表分数归一化为0至10的量表，并使用方差分析测试组间差异。我们使用参数化项目反应模型来测试亚群体在项目区分度和项目难度方面的差异。在去除具有差异项目功能的项目后，我们重新检查了组间差异。
在说英语（新斯科舍）的受访者中，对护理体验的评价比对说法语（魁北克）的受访者更为积极。我们发现，在153个项目中有48%存在英语/法语项目功能差异：20%存在区分度差异，28%存在难度差异。总体而言，英语项目的区分度比法语项目更高。去除有问题的项目并没有改变法语/英语评估中的差异。高/低教育程度导致的项目功能差异影响了27%的项目，在高教育程度组中项目的区分度通常更高。组间比较没有变化。相比之下，只有9%的项目显示出地域导致的项目功能差异，主要影响可及性属性。去除有问题的项目扭转了之前一个不显著的发现，揭示出农村地区首次接触医疗服务的可及性比城市地区更差。
项目功能差异不会使分量表上的法语/英语比较产生偏差或无效，但需要进一步改进以使法语和英语项目等效。这些工具在教育程度和地域方面相对稳健，但结果表明低教育程度和农村受访者在潜在结构上可能存在差异。
