O3205: “googling” Cerebral Palsy and Autism: Utilizing Search Engine Machine Learning Algorithms to Identify Questions Being Asked and the Available Online Resources

Background: Cerebral Palsy (CP) and Autism Spectrum Disorder (ASD) are the leading causes of childhood physical and global disability, respectively. As with many conditions that affect children, there are myriad sources of information available online of varying reputability. Parents access these websites via search engines such as Google, which uses a proprietary machine learning algorithm to compile commonly asked questions for a particular search term. We aimed to study these questions along with the websites returned by the search algorithm to better understand what types of questions parents of patients with CP and ASD were asking and to review the quality and accuracy of information readily available. We further aimed to identify differences between the types of questions asked and resources available for each condition.

Methods: Google searches of “Cerebral Palsy” and “Autism” were performed; questions suggested by language processing algorithms were recorded and analyzed. Questions were categorized by Rothwell type (fact based, policy based or value based) and subtyped by two independent reviewers. Resources related to each question were categorized (academic, government, commercial). Each resource was scored according to JAMA benchmark quality criteria. Nominal and discrete data were analyzed using Chi-squared and student’s t-tests, respectively.

Results: The top 100 suggested questions for CP and ASD were included for analysis. There was no difference in Rothwell type between groups (p=0.07), but the subtypes were significantly different between groups (p=0.04). There were more treatment related questions for CP (9% versus 2%), while there were more holistic-understanding related questions for ASD (21% versus 13.5%). There was a greater share of commercial sites for ASD compared to CP (41% v 22%, p< 0.001) and more sites published by law firms for CP (32% v 2%, p< 0.001). Mean JAMA benchmark quality scores were higher in CP compared to ASD (2.198 vs 1.767, SD: 1.19, 1.25, P = 0.01). Mean Fleish-Kincaid Reading Level was above the 11th grade for both the ASD and CP sites, far exceeding the recommended 6th grade level for patient information.

Conclusion: There was a significant difference between the type of questions and resources identified in CP and ASD searches. Searches related to CP tended to be more medical in nature, while ASD queries identified more questions regarding the societal impact of the diagnosis. While there were discrepancies in website publishers between cohorts, both searches yielded similarly limited academic resources. As these two conditions affect a substantial population of children cared for by pediatric physicians, it is imperative that we work to increase the availability of appropriate, reliable, evidence-based, high-quality online resources for our patients and their families. Moreover, parents and caregivers should be educated on evaluating the quality and source information available for these and other diagnoses.