For this survey, investigators at the University of Texas Health Science Center at Houston (UTHealth) and “Carol Davila” University of Medicine and Pharmacy (UMF) collaborated to develop a questionnaire by modifying a previously established questionnaire used in Pakistan  and Turkey , to assess the knowledge, attitude, and practices of physicians regarding ASD in Romania. The questionnaire was translated to Romanian and back translated to English by the Romania team to ensure the accuracy of the translations. The study protocol was approved by the Institutional Review Boards (IRBs) of UTHealth (HSC-GEN-15-0844) and Ethical Review Committee (ERC) (PO-35-F-03) of Carol Davila-UMF.
Participants in this survey were selected from a listing of all the private practices and clinics of physicians in Romania. A volunteer team that included medical students from the “Professor Alexandru Obregia” Clinical Psychiatry Hospital were provided the aforementioned list to assist with distribution and collection of completed questionnaires from the physicians at each of their practices from January to July in 2017. As a result of this effort, they collected a total of 383 completed questionnaires.
Data management and quality assurance of data
The UTHealth team provided a Research Electronic Data Capture (REDCap)  bilingual database in Romanian and English for data entry from the questionnaires. We used the double-data entry method  to minimize discrepant data as part of our quality assurance procedures. Initially, the Carol Davila team in Romania entered data from the questionnaires into the REDCap database. Then the completed questionnaires were sent to the UTHealth team in the US for a second round of data entry. Since physicians responded to the open-ended questions in Romanian, we identified a graduate student at UTHealth, who is a native Romanian speaker, to translate the responses into English, which were then entered into the REDCap database.
The questionnaire for assessing knowledge, attitude, and practices of physicians about ASD
The Questionnaire contained three sections. Section A had 19 questions focused on assessing socioeconomic and demographic information including age and sex. This section also inquired about the participant’s background in the medical field and their current practice (e.g., location of clinic, number of patients seen in a day in their practice, etc.). This section ended by asking if the physician personally knew someone with ASD.
Section B had six questions to assess the participants’ knowledge and attitude about ASD. This section first asked if the participant had ever heard of “Autism or Autism Spectrum Disorder.” If the participant marked “Yes,” then in the following question they were asked to provide the sources where they heard about ASD. All participants were then asked to provide an estimate for the prevalence of ASD in Romania, the US, and globally. The physicians were subsequently asked that out of every 100 children that s/he sees in her/his practice, how many children have ASD. At the end of this section there were14 statements that were prepared to assess various aspects of the knowledge, attitude, and practices of the Romanian physicians about ASD, for which the response options were in a 5-point Likert-scale: “Strongly agree,” “Agree,” “Undecided,” “Disagree,” and “Strongly disagree.”
For developing an overall score to quantify various aspects of physicians’ knowledge about ASD, we have classified the Likert-scale responses to binary. Specifically, depending on whether the statement in the question was true or false, the “Undecided” groups were merged either with the “Agree” or “Disagree” groups. For example, since question 2, “Autism is a possible result of neglect by the parents,” is false, the physicians who marked “Undecided” for question 2 were merged with the “Agree” and “Strongly agree” groups. Conversely, because question 8, “Children with autism require special education,” is true, the physicians who marked “Undecided” for question 8 were merged with the “Disagree” and “Strongly disagree” groups. As a result of this reclassification, we derived binary variables based on whether the statements in the question were true or false. Since questions 1, 6, 8, 9, 10, 11, and 12 are true, we recoded “Strongly agree” and “Agree” as 1, and “Undecided”, “Disagree” and “Strongly disagree” as 0. For questions 2, 3, 4, 5, 7, and 14, which are false, “Strongly agree”, “Agree”, and “Undecided” were recoded as 0, and “Disagree” and “Strongly disagree” were recoded as 1. We determined that the responses to Question 13, “Parents in Romania tend to think their children are at risk for autism,” may not be reliable because the prevalence or risk of ASD in Romania is currently unknown. Additionally, we determined that the responses to Question 9, “Children with autism deliberately misbehave,” may not be reliable as they could fluctuate depending on how the physicians interpreted the statement. Due to their ambiguous nature, Questions 9 and 13 were excluded from further analyses. Based on the remaining 12 questions, we calculated the sum of all the question scores as a total score for physicians’ knowledge about ASD, which we used for further analyses.
Section C had six questions designed to assess the participants’ practices about ASD. This section assessed the physician’s knowledge of available tools to screen children for ASD, and to list the screening tools they had used in the past, if any. The physicians were then asked about early indicators of ASD in 2-year-old children, and what they do when they suspect a child has ASD. Section C continued by asking the physicians about which ASD diagnostic tools they had used in the past (“Have you ever used any of the following to diagnose a child with autism or Autism Spectrum Disorder”). Next the physicians were presented with a 4-point Likert-scale question (“In diagnosing children with autism, the following symptoms are”) comprised of 11 accurate or inaccurate symptoms used to diagnose ASD with response options of “Necessary,” “Not necessary, but helpful,” “Not helpful,” and “Don’t know.” The last question in this section asked physicians to provide their opinion on ways to reduce the prevalence of ASD in Romania.
Sample size justification and statistical power
In order to estimate the proportion of physicians who are knowledgeable about ASD within a 5% margin of error with 95% confidence, we needed to survey at least 384 physicians. Considering that around 5% of the surveys were expected to be incomplete, we planned to survey 400 physicians. In our survey, a total of 383 physicians completed the survey. This sample size is also sufficient to detect an effect size of 0.3 or greater with at least 80% power at a 5% level of significance, assuming the ratio of physicians in the two groups compared (e.g., male vs. female physicians) with respect to their knowledge sub-score ranged from 0.40 to 0.60.
We used descriptive statistics to summarize various characteristics of the study population, including age and gender of the participants. Some of the open-ended questions were categorized before analysis. Physicians’ age was categorized as a dichotomous variable with categories of ≤ 35 & > 35 years. Since we received a variety of responses for names of the medical schools where the physicians earned their degrees, these responses were reduced to three categories: 1) Carol Davila University of Medicine and Pharmacy, 2) Grigore T. Popa University of Medicine and Pharmacy, and 3) Other Medical Schools. The physicians that responded to the survey had clinics in various cities of Romania, and the responses were categorized based on the number of survey responses received for each city, leaving us with four categories: Bucharest, Suçeava, Brăila, and other cities. We also asked physicians to write the year in which they completed their most recent continuing education course, for which the responses were divided into before 2007, between 2007 and 2014, and after 2014. The physicians were also asked to write down how many years they had been practicing medicine, and the responses were categorized as more than 30 years, between 16 and 30 years, and 15 years or less.
The responses for the average number of patients each physician saw in his/her practice daily and the average number of patients under the age of 12 seen by each physician in a day were categorized as: ≤ 20 patients, 21–40 patients, and > 40 patients per day. The self-reported responses for the average time a physician spent at his/her practice each day were categorized based on an 8-h workday (≤ 8 h and > 8 h per day). Likewise, responses for the average time a physician spent with each patient was classified as ≤15 min and > 15 min. Only binary (yes or no) responses were allowed for other variables such as whether the physician completed specific ward rotations in or after medical school, and if the physicians had heard of ASD in the past, and were reported as such.
All physicians were asked whether they had heard of ASD in the past. If a physician left this question blank, we interpreted that the physician did not know, hence included this in the group that have not heard of ASD. Physicians were then asked to mark where they had heard of ASD as their source of ASD knowledge (SAK). The 10 SAK options provided for selection were: medical school, continuing education, conferences, primary literature, colleagues, television, newspaper, internet, radio, and other sources not included in the list. Physicians were allowed to choose more than one SAK. Responses to the SAK question from physicians that had indicated that they had not heard of ASD or did not respond to the question were included in the analysis as an additional category. This resulted in a three-level variable for each SAK: those who heard of ASD from the selected SAK (Yes); those who heard of ASD but did not hear of ASD from this SAK (No); and those who had not heard of ASD.
Developing sub-scores to quantify various domains of physicians’ knowledge about ASD
To investigate the possible involvement of independent sub-domains in the overall knowledge score based on the 12-statement Likert-scale questions, we performed Exploratory Factor Analysis to determine these latent factors . We used the criteria specified in Samanic, et al.  for extracting variables for various factors. That is: Eigenvalue of 1.00 (i.e., % of variance explained by each factor is equivalent to the variance explained by only one variable) or more were retained for the analysis. Questions with an absolute factor loading value of ≥ 0.40 were included in the composite score (weighted score or factor score) for further analysis. The Factor Analysis resulted in five composite scores with factor-like sub-scores derived from questions included in each factor, with equal weights of 1.
Analysis of total score and sub-scores of physicians’ knowledge about ASD
The total score and factor-like sub-scores were used as dependent variables in univariable and multivariable General Linear Models (GLMs) or linear regression models to determine the physicians’ characteristics that were associated with their ASD knowledge sub-scores. Demographic, socioeconomic, and current clinical practice of the physicians were treated as categorical variables and were considered independent variables in the GLMs. Based on results from the univariable GLMs, we selected independent variables with a P-value ≤ 0.05 to include in the multivariable GLMs. We also explored possible interactions of these variables with each other in relation to the knowledge total score and sub-scores using GLMs. Variables were considered significantly associated with the total score and sub-scores if they had a P-value ≤ 0.05. Each of the final GLMs for the total score and each sub-score were used to obtain and report the Least Square Means (LSMeans) as the adjusted mean sub-scores (AMS), after adjusting for other variables in their final respective models. All analyses were performed using the SAS 9.4  statistical software package.