Lakhasly

Sample Data come from the Trajectories and Origins (TeO)

gaps in foregone care between groups.

study [23], a large-scale, nationally representative cross-

Rvenbark and ichou BMC Public Health (2020) 20:31

Page 3 of 10

sectional survey of France. The survey was conducted from 2008 to 2009 with in-person home interviews across France. The sample consisted of 21,761 individ- uals aged 18 to 59, with oversamples of immigrants and individuals bom to at least one immigrant (> 8000 of each group).

Theoretical framework

Models were conceptualized in line with the adapted Behavioral Model for Vulnerable Populations described by Gelberg and colleagues [28], in which the use of healthcare services represents a health behavior that is influenced by upstream population characteristics. The main population characteristics of interest in this study include demographic characteristics "predisposing" factors) of gender, ethnicity, immigrant generation, and religion. Other factors that we attempt to account for given the available data include the "predisposing" factors of age, marital status, education, and employment; the "enabling" factor of family income, and the "need factor of perceived and evaluated health status.

Measures Healthcare experiences

Discrimination in healthcare was measured with a single yes/no question: "Has a doctor or other medical care worker ever treated you less well or received you less well than other patients?" Likewise, foregone healthcare was also assessed with a yes/no question: "During the past 12 months, have you foregone health care for your self?". Each measure was coded dichotomously.

Demographic characteristics

As this study was explicitly interested in group disparities in healthcare experiences, we conducted analyses across a series of demographic measures, all of which were self- reported in the survey. Characteristics of interest include gender, immigrant generation French-born", which refers to French-born individuals to French-bom parentse first generation immigrant; or second generation immigrant), country of origin (for either the individual or parent, de pending on the relevant immigrant generation, grouped into geographic categories), and religion.

Covariates Additional survey items were included as control vari ables in this study, including age age (weighted M 39.1, SD12.4), marital status (married 46.7%, weighted) socioeconomic status, and health status. Socioeconomic status was measured with three variables for self-reported monthly income (weighted M=16816, SD=9546), educa- tional attainment (weighted: less than middle school equivalent 11.3%, middle school equivalent 13.3%, voca tional training 26.9%, high school equivalent or higher 48.6%), and employment status (weighted: employed

by various groups as the predicted probabilities of experien- cing discrimination based on demographic characteristics. We calculated these predicted probabilities from logistic regression models of healthcare discrimination, and we contrasted coefficient estimates against a reference group for statistical comparison. For each demographic factor of interest (gender, migrant generation, origin, and religion), we constructed three nested models. The first model in cluded the demographic predictor, with age and gender (if gender was not the factor investigated) as covariates the second model added covariates for socioeconomic status; the third model added covariates for health status.

Second, we reported the predicted probabilities of foregoing healthcare across the demographic groups of interest, and then calculated the average marginal effects (AMEs) of the demographic characteristics of interest on those predicted probabilities. We did this by modeling. reports of foregone healthcare across three nested logis- regression models: the first included only the demo graphic factor of interest; the second added discrimination; and the third added all other demo graphic characteristics, socioeconomic status, and health status. We present our findings as AMEs for two main

tic reasons. First, AMEs are less affected by bias arising from unobserved heterogeneity across nested logistic models than odds ratios or raw logistic regression coeffi cients (29-31]. Second, we believe that AMEs provide a more intuitive description of effect size than odds ratios or logistic regression coefficients, as AMEs can be read as percentage-point increases in predicted probability. Finally, we determined how much of the disparities int

foregoing healthcare across various groups is potentially

explained by experiences of discrimination in healthcare. We did this by calculating the percentage of the Model 1 AME (that is, the AME of a group demographic char acteristic) explained by the addition of discrimination as a covariate in Model 2, so that: % explained = 1- (AME- Mulel 2 AMEtet. Statistical significance of the "per- ant explained" was tested by contrasting a demographic. characteristic's AME in Model 2 against the same AME in Model 1. Put another way, we tested the null hypoth esis that the addition of discrimination in the model re- sulted in no change in the estimated AME for a demographic characteristic.

Page 4 of 10

Avenbark and knou BMC Public Health (2020) 20:31

Results

Descriptive statistics of the sample are shown in Table 1. Overall, the survey-weighted prevalence of reporting discrimination in healthcare settings was 3.9%, with a range of 26 to 9.3% across the various demographic groups examined. In bivariate comparisons, significantly higher rates of discrimination were observed for: women compared to men; 1st generation immigrants compared to French-bom; those with origins in Overseas France, Africa, and Turkey compared to those from Mainland France, and Muslims and those with no religion com- pared to Christians.

Also seen in Table 1, the survey-weighted rate of fore gone healthcare was 10.9% overall, ranging from 6.2 to 22.0% across demographic groups. Bivariate comparison tests are displayed in the table, and represented

73.1%, unemployed 8.8%, student 5.4%, inactive 12.7%). Health status was also measured with three vari- ables, consisting of self-rated health (weighted M=1.83, SD=79), history of chronic illnesses (yes 27.1%, weighted), and number of healthcare visits in the last year (weighted: none 8.2%, once 24.4%, several 67.5%).

Analyses

Analyses proceeded in three main steps. First, we described