2.1 Anxiety Survey
The initial survey questions were created within a focus group of eight professionals (parents, physicians, psychologists) with extensive expertise in FXS and involved in the National Fragile X Foundation (NFXF). The initial draft survey included both structured, forced-choice questions as well as open-ended questions. In addition to questions related to anxiety, the survey included items detailing demographics (e.g., age, sex) of the individual with FXS. Example of questions pertaining to the presence of anxiety were: is the person with FXS, at times, anxious, worried, nervous or afraid; can the person with FXS state with their words that he/she is anxious; behavioral or verbal symptoms, and physical symptoms, reported when the responder believes the person with FXS is anxious, along with frequency, intensity and duration of such symptoms. For respondents identified as having FXS (self-advocate), the same questions were presented with a change to first person text, (e.g., “what do you do or say?” versus “what do you see or hear?”).
The draft survey was presented to a focus group including parents of individuals with FXS (eight parents of children ranging from 5 to 35 years of age), two people with FXS, and seven medical providers, with the purpose of confirming that the survey was comprehensive, clear, and respectfully worded. The results prompted the inclusion of duration and intensity to be separately rated for the symptoms observed. Subsequently, the final survey was sent via email to three families (two parents, one family of a teenager, one family of an adult with FXS, and one sibling of an individual with FXS) and reviewed extensively by employees of the NFXF who volunteered to pilot the process involved in completing the survey. No further adjustments were requested based on this pilot sample. No identifying information was included in the survey. The final survey can be seen in the Supplementary File S1 (PDF) and includes a total of 21 items.
The newly generated, anonymous survey was administered online through a survey software program. Under the Protection of Human Subjects section 45 CFR 46.101(b)(2) this study was exempted. The NFXF sent a survey link to 10,000+ emails subscribed to receive its general emails. The NFXF also posted a link to the survey on their social media sites, and a link was posted on their website August through December 2019 at the MyFXResearch portal site. Recipients of the email were eligible to participate if they were an individual with FXS or were a family member/caretaker of an individual with FXS. Participants caring for multiple individuals with FXS were eligible to complete the survey once per individual with FXS. While a total of 1,414 people clicked on the survey link, only 482 completed the survey.
456 caregivers of individuals with FXS between the ages of 2 and 81 years (87 female, 369 male) completed the survey. Among females with FXS, n = 4 were 2–5 years, n = 17 were 6–12 years, n = 19 were 13–17 years, n = 30 were 18–29 years, n = 16 were 30–49 years and n = 1 was >50 years. Among males, n = 33 were 2–5 years, n = 80 were 6–12 years, n = 70 were 13–17 years, n = 102 were 18–29 years, n = 74 were 30–49 years and n = 10 were >50 years. In addition, 28 self-advocates (24 female and 2 male) age 15–66 (Mean 41.8 ± 15.6) completed the survey.
2.3.1. Quantitative Analyses
Given the numerous behavioral or physical features of anxiety on the survey, a descriptive, rather than hypothesis-driven approach was taken. For both caregiver and self-advocate surveys, participants first indicated whether specific symptoms occurred during periods of anxiety. If the respondent answered “yes”, they were asked to make ratings pertaining to frequency, intensity, and duration of the symptoms. Frequency was rated using the following scale: 1 = Rarely when they are anxious, 2 = Sometimes when they are anxious, 3 = About half the time when they are anxious, 4 = Usually when they are anxious, 5 = Always when they are anxious. Duration was rated using the following scale: 1 = Less than I minute, 2 = 1–5 min, 3 = 6–15 min, 4 = 16–60 min, 5 = Greater than 60 min. Intensity was rated as scale of 1–5 (1 = mild, 2= slightly, 3 = moderately, 4 = very much, 5 = extremely). Because of the subjectivity of the intensity ratings the data were not reported, rather focusing on frequency and duration.
We first examined whether age was associated with the presence or frequency of specific behaviors related to anxiety within females and males using Spearman correlations. Given the exploratory nature of the analysis, uncorrected p-values are reported and thresholded using p < 0.05. Next, the proportion of participants (all females, given limited N; males within each age bin: 2–5, 6–12, 13–17, 18–29, and 30–49 years) responding “yes” was quantified and rank ordered to allow identification of commonly reported symptoms. The same approach was taken for determining the symptoms that occur most frequently and for duration. These rank-ordered variables were graphed and reviewed to identify prominent patterns.
2.3.2. Factor Analysis
Both male and female participants aged 5–49 were included in the factor analysis. This subset of 418 participants was 80% male (n = 333) and 20% female (n = 83) with less than 1% unknown (n = 2 missing sex). The mean age of the sample was 20 years (SD = 10) (quartiles = 12, 18, 27 years). For the measurement invariance analyses, the subgroup was split into participants younger than 18 years (n = 196, 80% male, mean age 12 ± 4) and 18+ years (n = 222, 79% male, mean age 28 ± 7).
Each item had four response scales: yes/no, intensity, duration, and frequency. For the purpose of exploratory factor analysis, the yes/no scale was used. A response was not required for any of the items, and some respondents indicated “no” by leaving the item blank. To account for this, all non-responses to the yes/no scale were coded “no.”
The goal of this analysis was to explore a candidate factor structure for the development of an instrument to measure observable behaviors related to anxiety in FXS, and to evaluate its measurement invariance across age groups. The first stage was exploratory data analysis to evaluate cell size and interitem correlations. Second, exploratory factor analysis was conducted. Given the binary response scale, the estimation method was weighted least square mean and variance adjusted (WLSMV). Oblique (geomin) rotation was used. Solutions with up to four factors were requested.
Model fit was evaluated using the relative fit indices root mean square error of approximation (RMSEA), comparative fit index (CFI), and the Tucker-Lewis index (TLI). Smaller values of RMSEA and larger values for CFI and TLI Indicate better fit, and commonly used thresholds are 0.06 for RMSEA and 0.95 for CFI and TLI . However, these fit indices are for continuous data and when used in categorical data they are biased . In addition to the fit of the solution, the quality was evaluated by noting the strength and pattern of item loadings. The selected solution was carried forward to the measurement invariance analysis. For this confirmatory method, items were adopted onto a factor if the loading was greater than 0.40.
Measurement invariance was evaluated in three stages. First, the fit of the model in the subgroups independently was assessed, evaluating whether requiring the same factor structure in both groups was associated with poor model fit. Next, the factor loadings were constrained to be equal across groups and assessed how this impacted model fit. Finally, the item thresholds were constrained to equality across groups. Theta parameterization was used because the data were binary. For this reason, the equivalence of residual variance was not tested. The same relative fit indices used for the exploratory analysis were used, but without thresholds, given that typical conventions do not likely apply when using binary data .
2.3.3. Qualitative Analysis
Free-text survey responses about behaviors seen, vocalizations heard, and physical symptoms related to anxiety were analyzed using a qualitative content analysis approach. Content analysis is a systematic approach to text analysis, using codes and categorization to label and organize content trends and patterns within a qualitative data set . Responses were uploaded to ATLAS.ti analytic software. A team-based consensus code was developed by a multidisciplinary team with extensive FXS clinical and research expertise including, parents, physicians (pediatricians, geneticist, neurologist, and psychiatrist), and psychologists . During the coding sessions 20% of the responses were discussed, one at a time, until consensus was reached on which code(s) to apply to each piece of text. The process was iterative; codes were collapsed and combined, and responses were recoded as needed to best represent the data. Next, two team members (psychologist and physician) working as primary coders used the codebook to independently co-code an additional 10% of the data. Intercoder agreement was calculated using Krippendorf’s α and a high rate of agreement (α = 0.96) between the coders supported the trustworthiness of the coding structure and the overall study . The remaining results were divided and coded independently between the primary coders. Next, code frequencies for specific emergent observable behaviors and physical symptoms were calculated and proportions were compared by group (male versus female; children versus adolescents/adults) using Pearson’s Chi-square analysis with significance set at p < 0.05. Caregiver responses to the open-ended question: “What makes you confident it is anxiety related?” were coded, categorized, and networks were developed to examine relationships and code co-occurrences. Several broad themes were developed to represent overarching ideas about anxiety in FXS, using direct quotes to support the themes. Final themes were discussed and approved by the entire multidisciplinary research team.