Autistic adults have poorer quality healthcare and worse health based on self-report data

The autism and healthcare experiences survey and participant cohort

We administered an anonymous, online survey via Qualtrics on quality of healthcare, including questions regarding demographic information, a short version of the Autism Spectrum Quotient (a measure of autistic traits, AQ-10) [31, 32], current healthcare behavior, sensory experience, communication, anxiety, access and advocacy, system problems, shutdowns, meltdowns, autism-specific experiences, and most recent healthcare experience. The questionnaire included 63 questions on health and healthcare, including questions that were multiple-choice, used a 4–point Likert Scale (with the options Strongly Agree, Somewhat Agree, Somewhat Disagree, Strongly Disagree), and requested open-ended free text responses. Further information on the content of each of these sections can be found in the additional file. Specifically, Additional file 1: Figures S1-S15 provide images of each of the survey questions exactly as they were presented to participants. We used publicly available materials from the National Health Service (NHS), National Institute for Health and Care Excellence (NICE), National Institutes of Health (NIH), and the World Health Organization (WHO) to develop the survey questions. After developing a draft of the survey, we also conducted in-depth interviews (lasting several hours each) with two middle-aged autistic adults about their experiences and asked them to provide feedback on the survey study, and we revised the survey accordingly.

We employed a cross-sectional, convenience sampling design and recruited participants via the Cambridge Autism Research Database (CARD), Autistica’s Discover Network, autism charities (including the Autism Research Trust), and social media (specifically Twitter, Facebook, and Reddit). Survey collection took place from July 2019 to January 2021. Our non-autistic sample may be biased toward individuals with an interest in autism, as the study advertisements and consent form indicated that the study compares the experiences of autistic and non-autistic adults. However, we attempted to mitigate this bias by advertising our study to the general population via paid advertisements on Facebook and Reddit. In addition, we excluded all individuals who reported that they suspected they were autistic, or who were awaiting assessment, or who self-diagnosed as autistic. At all stages of recruitment, both autistic and non-autistic individuals were invited to participate. Our use of social media advertisements enabled us to attempt to recruit a diverse, international sample, including people from 79 countries.

N = 4158 individuals accessed the survey. Any individual aged 16 years or older who consented to participate was eligible. We excluded n = 1371 individuals due to incomplete response, failure to consent, or unconfirmed age (for ethical consent reasons). Although nearly all questions on the survey were optional, an incomplete response was defined as a participant who failed to complete any questions across all of the following sections: sensory experience, communication, anxiety, access and advocacy, system problems, meltdowns, shutdowns, or autism-specific experiences. In addition, a further n = 112 participants were excluded due to suspected duplicate response. We were able to use an existing survey setting in the Qualtrics system to prevent individuals from responding to the survey multiple times from the same IP address. However, as the survey participants were all anonymous, there was no direct way to exclude individuals who had responded to the survey multiple times. Thus, we used an algorithm to identify potential duplicate responses, excluding all participant records that matched any previous participant record across 12 criteria (autism diagnosis (yes/no), specific autism diagnosis, type of diagnosing practitioner, year of autism diagnosis, autistic family members (yes/no), age, country of residence, sex assigned at birth, current gender identity, education level, ethnicity, and AQ-10 score). Finally, as the study was anonymized, all autism diagnoses were self-reported; however, to confirm a clinical diagnosis, we asked participants to provide further information including the type of practitioner who diagnosed them (e.g., psychiatrist, clinical psychologist, pediatrician, etc.), their year of diagnosis, and their specific diagnosis (e.g., autism spectrum disorder, Asperger’s, etc.). N = 26 further individuals were excluded from the study (both the autistic and non-autistic cohorts), as we were unable to confirm their autism status (e.g. awaiting assessment, suspected autism, or self-diagnosed as autistic). The final sample included n = 2649 individuals (n = 1285 autistic individuals).

Statistical analysis

We used R Version 3.6.2 to employ unadjusted and adjusted models to assess the lifetime prevalence of mental and physical health conditions and healthcare quality across a wide variety of topics. We used Fisher’s exact tests (‘CrossTable’ function of the ‘gmodels’ package) to provide our unadjusted estimate. All adjusted analyses controlled for age, ethnicity, education level, and country of residence and employed the ‘glm’ function from the ‘stats’ package. We used a significance threshold of p < 0.001 for all analyses to correct for type II errors due to multiple comparisons. For adjusted analyses only, we utilized five iterations of multiple imputation for chained equations (MICE package) to address missingness across the covariates of ethnicity (5.29% missing), education level (1.13% missing), and country of residence (1.89% missing) [33]. Although nearly all questions in the survey were optional, there was very little missing data regarding outcomes (< 10% per question). Information on missing data per question has been provided in Additional file 1: Table S1. As a note, we did not impute any data for outcomes—only for covariates as specified above.

We used education level as a proxy measure of socioeconomic status. The covariate was coded as a categorical variable and was defined as the highest qualification held with the following options: 'No formal qualifications,' 'Secondary School/High School level qualifications,' 'Further vocational qualifications,' 'University Undergraduate level qualifications (BA, BSc, etc.),' and 'University Postgraduate level qualifications (MA, MSc, PhD, Certificate, etc.)'. Country of residence was also coded as a categorical variable with the following options (based on highest response frequency): 'United Kingdom,' 'United States of America,' Germany,' 'Australia,' 'Canada,' 'Netherlands' and 'Other' countries. Unfortunately, due to low response rates from individuals from all non-white ethnic backgrounds, we used a binary representation of ethnicity across all analyses. We have also included the specific details for each sub-section of analyses conducted during the study, with labeled sub-headings for ease.

Lifetime prevalence of mental and physical health conditions

Prevalence of mental and physical health conditions may vary greatly by sex assigned at birth among individuals in the general population, as well as autistic individuals specifically [4, 5, 9, 11, 15,16,17]. As such, we employed sex-stratified unadjusted and adjusted analyses for individuals assigned male and female at birth to compare self-reported diagnoses of a variety of mental and physical health conditions. We have reported all findings for which an adjusted model could be fitted for both assigned female at birth and assigned male at birth individuals, although it should be noted that the results for a few conditions have not been reported due to perfect separation of the model. Perfect separation occurs when the value of a covariate directly predicts the value of the response variable (e.g. all non-white individuals are autistic); as such, the model cannot be reliably fitted, increasing the errors (and thereby 95% confidence intervals) dramatically. N = 4 individuals identified as 'Other' for their sex assigned at birth, but all were included in the autistic sample. Thus, these individuals were excluded from the sex-stratified analyses only.

Healthcare experiences

We simplified the responses to the 4-point Likert scale into a binary form, in order to establish the total number of individuals who endorsed or rejected each item. This section of the questionnaire included 33 items across six categories: current healthcare behavior, sensory experience, communication, anxiety, access and advocacy, and system problems. At the top of each of these subsections, the following heading was provided: 'Please answer the following questions about your experiences of going to see a healthcare professional (Doctor, General Practitioner, Nurse Practitioner, Nurse, or Physician’s Assistant).' Images of the survey questions for these sections have been provided in Additional file 1: Figures S1–S6. Using both unadjusted and adjusted tests, we compared the frequency of endorsing each individual item between autistic and non-autistic adults. In addition, as we had self-reported information about clinical diagnoses of autism, and autistic adults may be more likely than others to be diagnosed with an anxiety disorder [4, 5, 7, 10], we also conducted a sensitivity analysis for the anxiety subsection of the survey using a binary covariate of anxiety diagnosis. There were no significant differences in effect and full results have been reported in Additional file 1: Table S4.

Shutdowns and meltdowns

As part of the survey, we also asked both autistic and non-autistic individuals to endorse whether a common health-related situation had ever caused a shutdown, meltdown, or neither (individuals were able to select both shutdown and meltdown for each item). These terms were defined specifically on the survey and exact phrasing can be found in Fig. 1. This section of the survey served to provide information about the severity of distress that unmet health needs or a poor-quality healthcare experience could cause autistic or non-autistic individuals. Once again, we used Fisher’s exact tests and binomial logistic regression (as described above) to determine if these experiences differed between autistic and non-autistic groups.

Fig. 1figure 1

Definitions of Shutdowns and Meltdowns provided to all participants

Figure 1 shows the question provided to all participants asking whether their previous healthcare experiences had ever resulted in shutdowns or meltdowns. All participants (both autistic and non-autistic) were also provided with the accompanying definitions of what is meant by ‘shutdown’ or ‘meltdown’ in the context of this survey.

Healthcare inequality section scores

We then combined the individual responses on the items in the sensory experience, communication, anxiety, access and advocacy, and system problems in two different ways in order to provide a measure of overall healthcare inequality. First, we collated the items for each subsection (e.g., all of the sensory experience questions together) where each self-reported negative experience was coded as '1 point' and each positive experience was coded as '0 points.' We then added together all of the points for each subsection and divided by the total number of questions of each section to get a composite score from 0 to 1. As all items in each section were optional, only questions that were answered by each participant were factored into this calculation. However, we also ran a sensitivity analysis excluding all participants with missing data. The overall results did not change and they can be found in Additional file 1: Tables S2 and S3. Second, we used the same procedure to calculate a total score across all answered questions from any of these sections—once again, excluding unanswered questions from this analysis. Using these scores, we than ran both unadjusted and adjusted Binomial Logistic Regression models (‘glm’ function of the ‘stats’ package) to determine whether likelihood of being diagnosed with autism differed based on each of the subsection healthcare inequality scores or the total healthcare inequality score. We ran this in the whole population, as well as in sex-stratified AFAB and AMAB groups, respectively. As above, intersex participants were included in the overall analyses but not in either of the sex-stratified groups.

We tested whether we could use these healthcare inequality scores alone to correctly classify individuals as autistic or non-autistic. We used the following procedure for each subsection and the total score iteratively. First, we used the ‘createDataPartition’ function from the ‘caret’ package to section off 20% of the population (including 20% of autistic and non-autistic adults each) as a training dataset. We then used the used the ‘predict’ function from the ‘stats’ package to test whether the chosen healthcare inequality score correctly predicted autism diagnosis on the remaining 80% of the sample. The accuracy, specificity, and sensitivity of each of these tests has been reported as well. The purpose of using binomial logistic regression and machine learning methods to predict autism status based on health inequality scores was to determine whether self-reported healthcare quality was markedly different between autistic and non-autistic individuals.

Autism specific questions

If participants self-reported an autism diagnosis at the outset of the survey, we also asked them a few additional questions about their autism-specific experiences of healthcare with their main healthcare provider (exact wording has been provided in Additional file 1: Figure S16. As these questions were not asked to the non-autistic participants, we did not conduct any statistical analyses on this section. However, we have provided the summary data for the questions in Fig. 3.

Effects of the Covid-19 pandemic on healthcare quality

The questionnaire asked participants to answer questions about their most recent healthcare experience (screenshots of these questions have been included as Additional file 1: Figures S14 and S15). As part of this section of the survey, participants were asked to include the date of this healthcare appointment. The WHO officially designated the Covid-19 infection as a pandemic on March 11th, 2020. The present study attempted to understand whether the coronavirus pandemic has affected healthcare quality by designating healthcare appointments as ‘during the pandemic’ if they occurred on or after March 11th, 2020. Only participants who included full date information (e.g., including Date, Month, and Year of their most recent healthcare appointment) were included in these analyses. Full demographic information on this subsample can be found in Additional file 1: Table S5. We then employed Fisher’s exact tests (using the CrossTable function from the ‘gmodels’ package) to determine whether healthcare quality was different between autistic and non-autistic adults before and during the pandemic, respectively; in addition, we considered whether healthcare quality changed before and during the pandemic for autistic adults and non-autistic adults, respectively.

留言 (0)

沒有登入
gif