Interlenghi Gabriela S, Reichenheim Michael E, Segall-Corrêa Ana M, Pérez-Escamilla Rafael, Moraes Claudia L, Salles-Costa Rosana
Department of Epidemiology, Institute of Social Medicine, Rio de Janeiro State University, Rio de Janeiro, Brazil;
Department of Epidemiology, Institute of Social Medicine, Rio de Janeiro State University, Rio de Janeiro, Brazil.
J Nutr. 2017 Jul;147(7):1356-1365. doi: 10.3945/jn.117.249581. Epub 2017 May 31.
This is the second part of a model-based approach to examine the suitability of the current cutoffs applied to the raw score of the Brazilian Household Food Insecurity Measurement Scale [Escala Brasileira de Insegurança Alimentar (EBIA)]. The approach allows identification of homogeneous groups who correspond to severity levels of food insecurity (FI) and, by extension, discriminant cutoffs able to accurately distinguish these groups. This study aims to examine whether the model-based approach for identifying optimal cutoffs first implemented in a local sample is replicated in a countrywide representative sample. Data were derived from the Brazilian National Household Sample Survey of 2013 ( = 116,543 households). Latent class factor analysis (LCFA) models from 2 to 5 classes were applied to the scale's items to identify the number of underlying FI latent classes. Next, identification of optimal cutoffs on the overall raw score was ascertained from these identified classes. Analyses were conducted in the aggregate data and by macroregions. Finally, model-based classifications (latent classes and groupings identified thereafter) were contrasted to the traditionally used classification. LCFA identified 4 homogeneous groups with a very high degree of class separation (entropy = 0.934-0.975). The following cutoffs were identified in the aggregate data: between 1 and 2 (1/2), 5 and 6 (5/6), and 10 and 11 (10/11) in households with children and/or adolescents <18 y of age (score range: 0-14), and 1/2, between 4 and 5 (4/5), and between 6 and 7 (6/7) in adult-only households (range: 0-8). With minor variations, the same cutoffs were also identified in the macroregions. Although our findings confirm, in general, the classification currently used, the limit of 1/2 (compared with 0/1) for separating the milder from the baseline category emerged consistently in all analyses. Nationwide findings corroborate previous local evidence that households with an overall score of 1 are more akin to those scoring negative on all items. These results may contribute to guide experts' and policymakers' decisions on the most appropriate EBIA cutoffs.
这是基于模型的方法的第二部分，用于检验当前应用于巴西家庭粮食不安全测量量表[巴西粮食不安全量表（EBIA）]原始分数的临界值的适用性。该方法能够识别出与粮食不安全（FI）严重程度相对应的同质群体，进而确定能够准确区分这些群体的判别临界值。本研究旨在检验在当地样本中首次实施的基于模型的确定最优临界值的方法，在全国代表性样本中是否能够得到重复验证。数据来源于2013年巴西全国家庭抽样调查（n = 116,543户）。将2至5类的潜在类别因子分析（LCFA）模型应用于该量表的项目，以确定潜在的FI类别数量。接下来，从这些已识别的类别中确定总体原始分数的最优临界值。分析在汇总数据和按大区进行。最后，将基于模型的分类（潜在类别及此后确定的分组）与传统使用的分类进行对比。LCFA识别出4个具有高度类别分离度的同质群体（熵 = 0.934 - 0.975）。在汇总数据中确定了以下临界值：在有18岁及以下儿童和/或青少年的家庭中（分数范围：0 - 14）为1至2（1/2）、5至6（5/6）和10至11（10/11），在仅为成年人的家庭中（范围：0 - 8）为1/2、4至5（4/5）和6至7（6/7）。在各区域中也确定了相同的临界值，仅有微小差异。虽然我们的研究结果总体上证实了目前使用的分类，但在所有分析中，将较轻类别与基线类别区分开来时，1/2（与0/1相比）的临界值始终出现。全国范围的研究结果证实了先前当地的证据，即总体得分1的家庭与所有项目得分均为负的家庭更为相似。这些结果可能有助于指导专家和政策制定者关于最合适的EBIA临界值的决策。
