Using spontaneously generated online patient experiences to improve healthcare : A case study using Modafinil

Background Acknowledged issues with the RCT focus of EBM and recognition of the value of patient input have created a need for new methods of knowledge generation that can give the depth of qualitative studies but on a much larger scale. Almost half of the global population uses social media regularly, with increasing numbers of people using online spaces as either a first- or second-line health information and exchange resource. Estimates suggest the volume of online health related data grew by 300% between 2017 and 2020. As a data source, this unstructured freeform textual data is a form of patient generated health data, containing a mass of patient centred, contextually grounded detail about the perceptions and health concerns of those who post online. Methods for analysing it are at an early stage of development, but it is seen as having potential to add to clinical understanding, either by augmenting existing knowledge, or in aiding understanding of real-world usage of healthcare interventions and services. Objectives To explore how large-scale analysis of SGOPE can help with understanding patient perspectives of their conditions, symptoms, and self-management behaviours, assess the effectiveness of interventions, contribute to the process of knowledge and evidence creation, and consequently help healthcare systems improve outcomes in the most efficient manner. A secondary aim is to contribute to the development of methods that can be generalised across other interventions or services. Methods Using Modafinil as a case study, a multistage approach was taken. First, an exploratory study, comparing both qualitative and basic NLP techniques was undertaken on a small sample of 260 posts to identify topics, evaluate effectiveness and identify perceived causal text. An umbrella scoping review was then undertaken exploring how and for what purposes SGOPE data is currently being used within healthcare research. Findings from both then guided the main study, which used a variety of unsupervised NLP tools to explore the main dataset of over 69k posts. Individual methods were compared against each other. Results from both studies were compared and for evaluation. Results In contrast to the existing inconclusive systematic review evidence for Modafinil for anything other than narcolepsy, both studies found that Modafinil is seen as by posters as effective in treating fatigue and cognition symptoms in a wide range of conditions. Both identified the topics mentioned in the data, although more work needs to be done to develop the NLP methods to achieve a greater depth of understanding. The first study identified eight themes within the posts: reason for taking, impact of symptoms, acquisition, dosage, side-effects, comparison with other interventions, effectiveness, and quality of life outcomes. Effectiveness of Modafinil was found to be 68% positive, 12% mixed and 18% negative. Expressions of causal belief were identified. In the main study, effectiveness was measured with sentiment analysis, with all methods showing strong positive sentiment. Topic modelling identified groups of themes. Linguistic techniques extracted phrases indicating causality. Various analysis methods were compared to develop a method that could be generalised across other health topics