Exploring perceptions about Paracetamol, Tramadol and Codeine in Twitter: A machine learning study. (Preprint)

BACKGROUND Paracetamol, codeine and tramadol are used to control mild pain. These drugs can be dispensed without a prescription or medical consultation and can be the gateway to the opioid addiction. OBJECTIVE Identify the perceptions and experiences of Twitter users in relation to these drugs. METHODS In this study we have focused on tweets posted in English or Spanish between January 2019 and December 2020 that mentioned paracetamol, tramadol or codeine. A total of 152.056 tweets were collected, but 49.462 tweets were excluded. We analyzed the content of the tweets according to the codebook we created for the study. In terms of type of user, we distinguished between patients, family members and friends, health care professionals or institutions. In terms of the content, if it was of a medical nature, we classified it depending on whether it made reference to the efficacy of a drug or its adverse effects. We also analysed if the information stated was scientifically correct identifying those tweets that included links to scientific papers. Moreover, in the non-medical content, we distinguished between four themes: 1) Commercial issues; 2)Economic aspects; 3) Solidarity; and 4) Trivialization. Finally, looking at the drug we classified the tweets in three categories: paracetamol, tramadol, and codeine. We selected a total of 1000 tweets for each drug which were classified manually. The goal of this initial tagging was to provide the data to train, test and validate Machine Learning classifiers so that all the extracted tweets classification could be inferred. RESULTS Of the total of tweets considered classifiable, 42,840 tweets mentioned paracetamol and 42,131 tweets mentioned weak opioids (tramadol or codeine). 73.10% of the tweets were posted by patients. However, tweets posted by healthcare professionals and health institutions obtained the highest ratio of like-tweet and tweet-retweet. The distribution of medical content for each drug was statistically different (p<0,001). Non-medical content was predominant (73.9%) in tweets regarding opioids while in paracetamol tweets medical content was predominant (66.8%). Of those tweets that included medical content, 80.8% mentioned the efficacy of the drug and of those, only 6.9% described a good or sufficient efficacy. Among the non-medical content, we also found statistically significant differences in the distribution between the different drugs (p<0.001). CONCLUSIONS Twitter users show great interest in finding relief from their pain, with their posts focusing more on the effectiveness of the painkiller than its potential side effects. The number of tweets posted in the first person about trivialization of the drug, specifically about recreational use, and the lack of awareness about the possible side effects is alarming.