2.1 | Community involvement statement
The research team recognizes that this study could not be possible without the guidance from one team member (ZJW) who is an autistic self-advocate and autism researcher. Together, our team has expertise in genetics, genetic counseling, autism research, clinical care of autistic individuals and their families, and advocacy.
2.2 | Survey instrument
Participants completed an ad-hoc survey that assessed demographic characteristics; attitudes and beliefs about genetic testing, genetics research, and genetic counseling; and basic knowledge of autism genetics. The present investigation focuses specifically on the questions relevant to attitudes and beliefs toward genetic testing, as well as their demographic and clinical correlates. Our survey was guided by constructs found in the Theory of Planned Behavior (TPB; Azjen, 1991) conceptual framework (Fig. 1). The TPB has been utilized to study an individual’s beliefs, attitudes, and intention to perform a variety of behaviors. Intention, and ultimately behavior, is impacted by a person’s attitude toward the behavior and their perceived benefits (beliefs) of the behavior (Ajzen, 1991). This model has been applied in studies looking at factors that affect intention to pursue genetic testing (Wolff et al., 2011) and attitudes towards preconception carrier screening (Lakeman et al., 2009; Voorwinden et al., 2017). Zhao et al. (2021) also applied the TPB framework in a study looking at parents’ intention to pursue genetic testing for their autistic children. Our survey followed Ajzen’s (2019) TPB questionnaire outline by asking closed-ended questions on a Likert scale to assess the beliefs and attitudes of our participants.
TPB constructs specific to our study include ‘perceived benefits,’ ‘attitudes,’ and ‘intention.’ The first construct, ‘perceived benefits’ refers to the beliefs our participants have regarding the utility of genetic testing. ‘Attitudes’ includes the values that our survey respondents have towards genetic testing, and ‘intention’ refers to the likelihood that participants pursue genetic testing. Specific copies of the survey questions representing the TPB constructs can be found in Supplemental Table S2.
In addition to collecting demographic information (sex assigned at birth, gender identity, current age, person transcribing responses, country of residence, race/ethnicity, level of education, and information on autism diagnosis), our survey consisted of multiple-choice questions, 36 Likert-scale questions, and free-text responses specific to genetic testing. We also included some multiple choice and Likert-scale questions about personal experiences with genetic counselors, if any, and perceptions of genetic counseling, in general. To assess beliefs, attitudes, and intentions, respondents were asked to report ‘level of agreement’ and to report ‘likelihood’ to pursue genetic testing. The survey was self-directed and included a few open-end questions used for descriptive purposes only. Study data were collected and managed using REDCap electronic data capture tools hosted at Vanderbilt University (Harris et al., 2009; Harris et al., 2019).
2.3 | Participants and procedures
The Institutional Review Board of Vanderbilt University Medical Center approved our research protocol (IRB #221551). Our target sample population were autistic adults who were 18 years or older with a clinical or self-diagnosis of autism. Participants had to be able to complete our internet survey in English themselves or by having someone transcribe their responses. Participants were recruited from multiple sources, including social media (Facebook, Twitter) and direct contact with local and national autism advocacy groups. Participation was incentivized by offering 30 randomly-selected participants a $50 USD Amazon gift card or $50 USD via PayPal.
From an initial 444 survey responses, we removed all individuals not identifying as autistic (n = 62), as well as individuals who did not complete the relevant ‘genetic testing’ questions within the larger survey (n = 32). Additionally, the survey contained seven “instructed response” attention check questions (e.g., “To show that you’re paying attention, please select strongly agree.”) to flag participants for careless/low-effort responding, and individuals providing incorrect answers on more than one of these items were also removed from the dataset (n = 150). An additional seven participants were removed due to contradictory survey responses (i.e., impossible combinations of autism diagnostic status, autism family history, and receiving a genetic test specifically for autism). Lastly, free-text responses were examined for responses that appeared invalid (e.g., verbatim replications of other participants’ responses or the same free-text response across multiple text fields, text that did not answer the prompt, nonsense answers), which removed 20 potentially invalid responses. After exclusion of participants, the remaining dataset contained responses from 173 autistic adults (Fig. 2).
2.4 | Data analysis
Descriptive and exploratory analyses were performed in jamovi (version 2.3.21) (The jamovi project, 2022; R Core Team, 2021). Spearman rank correlations (rs) were used to calculate the strength of bivariate association between variables – beliefs and attitudes, attitudes intentions, and autism genetics knowledge (with confidence intervals calculated manually in R using the Caruso-Cliff [1997] method). For the purposes of our study, a positive correlation between beliefs and attitudes or between attitudes and intentions means “strongly agree” values occur together. A negative correlation indicates that a “strongly agree” value corresponds with a “strongly disagree” value. We considered bivariate correlations greater than 0.5 to demonstrate strong relationships between variables (Cohen, 1992). When comparing variables between groups who had versus had not received genetic testing, continuous variables (e.g., composite item scores) were compared with Welch (unequal-variance) t-tests (Delacre et al., 2017), and ordinal variables were compared using Brunner-Munzel tests and the probability of superiority effect size p̂ (Karch, 2021). Statistical significance for correlations and group comparisons was set to an alpha of 0.05, uncorrected.
To assess the latent structure of the belief and attitude items, we first conducted an exploratory graph analysis (Golino and Epskamp, 2017; Golino et al., 2020), examining the number and membership of Walktrap communities generated in a partial correlation network (EBICglasso estimation based on Pearson correlations). This demonstrated three separate communities of items, which we characterized as “Autism-specific beliefs” (7 items), “General beliefs unrelated to autism” (6 items, including one “attitude”), and “Attitudes about genetic testing” (10 items, including 3 “beliefs”). We reverse-scored three of the attitude items to allow positive scores on the items to represent more positive attitudes toward genetic testing in all cases. We then conducted principal components analysis (PCA) on the Pearson correlation matrix of each item set separately, extracting the first principal component from each item set to include in later correlation and regression analyses. The first principal components for the “Autism-specific beliefs,” “General beliefs,” and “Attitudes” scales explained 79%, 65%, and 59% of the variance in their respective groups of items. For the remaining between-group analyses, we used the unit-weighted mean of item scores in each community on a 1–7 scale, which could be interpreted as the degree to which a typical autistic participant agreed with the statements in that category on average. Mean item scores and principal component scores for the “Autism-specific beliefs,” “General beliefs,” and “Attitudes” constructs were generated from their respective 7-, 6-, and 10-item composites, with all items scored on a 7-point scale ranging from “Strongly Disagree” to “Strongly Agree” (and three of the attitude items reverse-scored such that higher scores indicated more positive feelings toward genetic testing).
To test whether there were associations between participants’ intention to pursue genetic testing for themselves or future children and their beliefs and attitudes towards genetic testing, we performed ordered (proportional-odds) logistic regression analyses (Bürkner & Vuorre, 2019; Harrell, 2015) using the rms R package. In each analysis, we used one of the four intention questions as the outcome variable and one of the principal component scores as a predictor, while also adjusting for age (continuous), gender (categorical: male, female, non-binary/other), and level of education (binary: 4-year college degree or more vs. did not complete college). For example one regression equation was as follows:
[Intention question] ~ [Predictor of interest] + age + gender + level of education
In these regressions, continuous predictors (such as component scores) were standardized such that they had a mean of zero and standard deviation of one in the sample. We repeated this analysis for each of the four intention questions, substituting the Autism-specific Beliefs Component with the General Beliefs Component and the Attitudes Component, respectively.
We conducted additional ordered logistic regression analyses to examine family history (presence or absence of other autistic family members) and autism diagnosis (provider- or self-diagnosed), as influencing factors on intention to pursue genetic testing. In each regression analysis, we used an intention question as the outcome variable and the family history or autism diagnosis as predictors (one focal test per model), while also adjusting for age, gender, and level of education. To account for multiple comparisons (n = 20 tests), we used a Bonferroni-Holm correction to adjust the p-values from focal hypothesis tests.
Finally, as an exploratory analysis, we also tested whether prior knowledge about autism genetics was associated with the participant’s intention to pursue genetic testing (with the standardized 0–10 knowledge composite used as a single predictor in a new regression), while adjusting for the same covariates as above. These analyses were also repeated after adjusting for both the beliefs and attitudes components that were found to significantly predict the intention variable in the initial regression analysis stage. As the analysis was exploratory, we opted not to use any statistical correction for multiple testing.