Thank you for listening to my talk at ACMT. Here are the slides. Please check out the rest of my research

Introduction

Social media provide a ubiquitous and readily accessible stream of real-world data. We have shown that social media activity predicts nonfatal opioid use (Chary, Genes, Giraud-carrier, Hanson, Nelson, & Manini, 2017), identifies patterns of combination of novel psychoactive substances2, and identifies dose-effect associations of rare ingestions.

A key limitation of these analyses is that they do not assess whether the variables under study causally interact. A crucial element of pharmacological investigations is the dose-response curve, which provides phenomenological evidence that a substance mediates a certain effect. If we could extract dose-effect associations from social media, we could directly compare these real-world data with those from more controlled investigations and, ultimately, infer results about substance efficacy from real-world data with the confidence now reserved for controlled trials.

The goal of this project is to develop a method that extracts dosage information from social media.

Introducing Jarvis

Jarvis is a combination of rules-based and unsupervised learning. We begin with rule based on human communication¹

The Roadmap

B –>

1. Extract Doses from Online Commentary (*We Are Here)
2. Extract Associated Effects
3. Estimate Dose-Effect Associations, the Social Media analog of Dose-Response Curves

The Code Repository

The Python code underlying Jarvis is in a private GitHub repository. Please contact us for access².

Bibliography

1. 1.Chary, M. et al. Epidemiology From Tweets: Estimating Misuse Of Prescription Opioids In The Usa From Social Media. journal Of Medical Toxicology 1–9 (2017).
2. 2.Chary, M., Yi, D. & Manini, A. F. Candyflipping And Other Combinations: Identifying Drug–Drug Combinations From An Online Forum. Frontiers In Psychiatry 9, 135 (2018).

---