Addi van Bergen

Social Exclusion Index-f or Health Surveys (SEI-HS) 73 4 other, local, questions [41, 42]. In our effort to construct a national measure for SE, we aimed at making maximum use of the available PHM standard questions, and using supplementary items from the SCP index only where the PHM fell short. In this paper we describe the construction and validation of this embedded measure for SE, the Social Exclusion Index for Health Surveys (SEI-HS). METHODS Data source and participants This survey study was conducted fall 2012 by 19 of the 28 Dutch Community Health Services who were involved in the implementation of the PHM. These 19 Community Health Service regions cover 71% of the Dutch population. In each Community Health Service region a sample was drawn from the non-institutionalised population aged 19 years and older (as of September 1, 2012), stratified by municipality, neighbourhood and age category (19-64 years and 65 years and older). In total, the 19 samples contained 566,521 persons. Selected persons received an announcement letter by mail, followed one week later by a questionnaire. The questionnaires could be filled out in writing or online. Non- responders received at least one written reminder. The four largest cities, having a higher proportion of hard to reach groups, made additional efforts such as home visits after the second written reminder, providing translated questionnaires (Turkish, English and Arabic) and offering personal assistance in completing the questionnaire if needed. Questionnaires were excluded if two third or more of the SE questions were not answered or in the case of lacking information on at least two thirds of the core questions. According to the national protocol, core questions include a.o. educational level, employment status, body weight and smoking. The net response rate was 45.7% (258,928 respondents). Weighting was used to correct for selective non-response and unequal selection probabilities caused by the stratified sampling design. Adjustment weights were calculated for the national sample, based on a linear model with auxiliary variables Community Health Service region (28 categories), gender (2), age (13), marital status (4), degree of urbanisation (5), household size (5), ethnicity (3), income (5) and municipality (391), and their interaction terms [43]. We adjusted these weights in accordance with the sample composition of our study. Item selection In our previous research [32] we identified with nonlinear canonical correlation analysis 16 PHM items from a pool of 62 potential items, measuring various aspects.