The sample for this study was a subset of health care professional students and faculty previously surveyed for another purpose.33 The present sample was limited to students in the control group of the primary study who answered all eight independent variables being tested (n=48).T
Participant characteristics are shown in Table 1. The sample was primarily white (65%), female (68.8%), and heterosexual (66.7%). The majority of students were medical students in clinical years of training (52.1%). Approximately 90% of participants reported being mostly or very liberal. Overall, the sample was more spiritual than religious and represented a variety of religions.
[Insert Table 1 near here]
The George Washington University IRB determined this study to be exempt (#180645) under Department of Health and Human Services regulatory categories 2 and 4.
The online survey asked a total of 144 questions (see Supplemental Material), 72 of which were included in this study. Items included 13 demographic and experience questions, the Lesbian, Gay, Bisexual, Transgender Development of Clinical Skills Scale (LGBT-DOCSS),14 the Attitudes Toward LGBT Patients Scale (ATLPS),33 and the Gay Affirming Practice Scale (GAPS).34
The LGBT-DOCSS is an 18-item scale with three subscales that measure constructs associated with self-reported competence in caring for SGM patients across interdisciplinary health care professionals. The scale has been tested for factor structure, reliability, and validity.14 In the original instrument, respondents rated their agreement with each item on a 7-point scale from strongly disagree (1) to strongly agree (7) for a total score ranging from 18-126 for the overall scale. Subscale ranges are: Basic Knowledge (4-28), Attitudinal Awareness (7-49); Clinical Preparedness (7-49). Total scores for the full scale and each subscale are intended to be tallied and then divided by the total number of items to obtain a mean score. Higher scores reflect greater self-reported competence in each domain.
In this study, the LGBT-DOCSS was altered in four ways: First, the scale was reduced from a 7-point to a 5-point scale. Second, the visual display was reversed, but the greater values were retained for “strongly agree” and the lesser value for “strongly disagree.” Both changes were made to ensure cognitive consistency for respondents—i.e., the same Likert scale direction for each of the three instruments on the questionnaire. Third, the middle answer option was moved to the far right to distinguish it as “Not sure” rather than neutral. This method was recommended by Dillman35 to provide a more authentic non-response option while retaining reasonable estimates of respondent attitudes.36 Finally, one item in the factor analysis of the LGBT-DOCSS manuscript was different from the final instrument published.14 Therefore, both items were included. After correspondence with the scale author (M. Pratt-Chapman to M. Bidell, October 2018); however, only the confirmed item was used in this analysis. Subscales were tallied for composite scores with a range of 4-20 (Knowledge), 7-35 (Attitudes), and 7-35 (Clinical Preparedness). Higher scores reflect greater knowledge, more affirming attitudes, and greater clinical preparedness, respectively.
The ATLPS is an 11-item scale measuring practitioner attitudes--including comfort with SGM patient encounters, opinions of SGM people, and beliefs about professional role. Responses are measured on a 5-point Likert scale from strongly disagree (1) to strongly agree (5) for a total score of 11-55 with higher scores reflecting more affirming SGM attitudes. For this study, the rating scale was identical to the published instrument, but the directionality was reversed and the neutral answer option was changed to “no opinion” and shifted to the far right to provide a clear non-response option for cognitive consistency across all scales.
The ATLPS was adapted from a prior scale of the same name originally created to assess differences in medical student attitudes about gay and lesbian patients.37 Wilson et al.29 made the scale more inclusive by changing “gay and lesbian” or “homosexual” to “LGBT” for three measures, by changing the word “physician” to “healthcare professionals” in another item, and by consolidating four items to two while simplifying language to be more accessible. Sanchez et al.’s original scale adapted items from a validated survey about physician attitudes toward patients with AIDS.38 Validity of Sanchez’s ATLPS has not been reported, but Wilson et al.29 found strong internal reliability of items when used as a single factor scale. This study used Wilson’s version of the scale, but face validity of two items was determined to be highly questionable. Therefore, while this scale was chosen in order to compare outcomes with other published studies, findings should be interpreted with caution.
The GAPS is a 30-item scale designed to measure health practitioners’ beliefs and behaviors regarding care of gay and lesbian individuals. The instrument uses a 5-point Likert scale from strongly agree (5) to strongly disagree (1) for items 1-15 and from always (5) to never (1) for items 16-30. The directionality and scoring for items were retained from the original instrument with the neutral answer option shifted to the far right to allow for a genuine non-response option as with the prior two scales. The range of possible scores for each subscale is 15-75 with a higher score reflecting more affirming Beliefs or Behaviors, respectively. Crisp established construct validity and strong internal reliability for each subscale.
Data were accessed through the secure RedCap database and analyzed using SPSS 24 (Armonk, NY). Since answers to the independent variables being tested were criteria for inclusion in the study, independent variables had no missing data but resulted in a limited sample size. Data for independent variables were not imputed due to the personal nature of sex-assigned-at-birth, sexual orientation, religiosity, spirituality, and political affiliation—characteristics that are inherent to the nature of the respondent. Missing data for dependent variables was less than 5%. Based on Cheema,39 this low amount of missing data can be dealt with in numerous ways, including multiple imputation techniques or leaving data as missing. For this study, data were left as missing.
G*Power 18.104.22.168 (Faul, Erdfelder, Buchner, & Lang, 2009) was used to conduct posthoc power analyses for all models, individual predictor variables within models, and model comparisons. F tests were used to test Models using the “Fixed model: R2 deviation from zero” option in with effect size set to medium (f2=.15) and a=.05. To test individual predictor variables, two-tailed t-tests were conducted for multiple regression assuming a fixed model. Based on the posthoc power analyses, the secondary sample was underpowered (1-b<.80) for most models to explain a medium effect (f2=.13) for a=.05 and for most individual predictors to detect a small effect (f2=.02) for a=.05.40 For Reduced Models, power ranged from (1-b)=.36-.75 with all Reduced Models powered at (1-b)≥.50. Because the sample was underpowered, variance in the criterion variable explained by individual predictors and for each model were examined rather than statistical significance.40
Multiple linear regression was used to test the value of an eight-variable model (Full Model) for each criterion variable. The eight independent variables were: sexual orientation, sex-assigned-at-birth, political affiliation, religiosity, spirituality, SGM affiliation (identifying as or having a friend or family member who identifies as LGBTQ), number of SGM-specific training hours, and number of SGM patient interactions in the last six months. Statistical significance of independent variables within each model as well as percent of variance explained was examined. Using Cohen’s40 benchmark’s for a small proportion of variance explained, any variable explaining >2% unique variance was included in the Reduced Model. Reduced Models were examined for statistical significance and proportion of variance explained based on Cohen’s40 benchmarks: small (R2 =.02), medium (R2=.13), and large (R2=.26). For all Reduced Models, interaction effects were examined by creating cross-product terms.42 Selection of final variables was based on model comparisons.41,42 Tolerance and VIF were checked for all Reduced Models to ensure that collinearity did not apply. Correlations of all independent variables and each criterion variable were also examined.
Descriptive and inferential statistics were reported. Ordinary Least Squares was used to test individual predictor variables. Multiple R was reported for correlation between the criterion variable and all predictors in each model. Multiple R2 was reported for percent variance in each criterion variable explained by all predictors in each model. Reduced Models were considered meaningful and parsimonious if there was no more than a 10% drop in total variance from the Full to the Reduced Model.