JMIR Publications

ABSTRACT

Background:

Understanding how individuals think about a topic can help to significantly improve communication. This is especially true when it comes to the medical domain where emotion and implications are high. Neurodevelopmental disorders (NDD) represent a group of diagnoses, affecting up to 18% of the population, involving differences in the development of cognitive or social functions and including attention deficit hyperactivity disorder (ADHD) as well as autism spectrum disorders (ASD). Both are complex conditions involving multiple symptoms and interventions where parents and health professionals interact. There is a gap in our global understanding of how each of those stakeholders differs in their preoccupations, making it difficult to address needs in knowledge mobilization.

Objective:

We aim to use Natural Language Processing techniques to build the Knowledge Graph from online information related to each stakeholder to help accelerate the identification of shared concerns and points of divergence between them. Ultimately, online information could be used to target knowledge mobilization and improve communication and care for individuals with ADHD and ASD.

Methods:

We created two datasets by collecting the posts from ASD and ADHD related online forums and PubMed abstracts and utilized the Unified Medical Language System (UMLS) to detect the biomedical concepts. Positive Pointwise mutual information (PPMI) followed by truncated Singular Value Decomposition (SVD) was applied to obtain the corpus-based UMLS concept embeddings for forums and PubMed. Property graph models were used for the Knowledge Graph representation of forums and PubMed. Semantic relatedness between concepts and the ASD condition or ADHD condition was calculated to rank the related concepts and stored as weight of edges. Additionally, UMLS semantic types were used to group concepts as well as to provide additional categorical information about concept’s domain.

Results:

Public forums on ADHD and ASD provide us with a wide range of concepts across multiple domains. Using Knowledge Graphs allows us to illustrate overlapping concepts between health professional literature (PubMed) and parental concerns (forums) with similar relevance scores, as the edge weight, and different co-occurrence frequency with the condition in each corpus, as the node size. Further, Knowledge Graphs also identify concepts with significantly different relevance scores between the stakeholders.

Conclusions:

Understanding the complex needs of families dealing with ASD or ADHD plays an important role in better communication between health professionals and families. Online public data, which is a source of information from large numbers of individuals, can provide significant insights into a condition. Moreover, it allows us to capture diversity in preoccupations and identify most relevant concepts for each stakeholder. Future research will be needed to identify how overlapping concepts may interact differently between each other for each stakeholder.