Abstract
BACKGROUND: Methylphenidate is an effective first-line treatment for attention-deficit/hyperactivity disorder (ADHD). However, many adverse effects of methylphenidate have been recorded from randomized clinical trials and patient-reported outcomes, but it is difficult to determine abuse from them. In the context of COVID-19, it is important to determine how drug use evaluation, as well as misuse of drugs, have been affected by the pandemic. As people share their reasons for using medication, patient sentiments, and the effects of medicine on social networking services (SNSs), the application of machine learning and SNS data can be a method to overcome the limitations. Proper machine learning models could be evaluated to validate the effects of the COVID-19 pandemic on drug use. OBJECTIVE: To analyze the effect of the COVID-19 pandemic on the use of methylphenidate, this study analyzed the adverse effects and nonmedical use of methylphenidate and evaluated the change in frequency of nonmedical use based on SNS data before and after the outbreak of COVID-19. Moreover, the performance of 4 machine learning models for classifying methylphenidate use based on SNS data was compared. METHODS: In this cross-sectional study, SNS data on methylphenidate from Twitter, Facebook, and Instagram from January 2019 to December 2020 were collected. The frequency of adverse effects, nonmedical use, and drug use before and after the COVID-19 pandemic were compared and analyzed. Interrupted time series analysis about the frequency and trends of nonmedical use of methylphenidate was conducted for 24 months from January 2019 to December 2020. Using the labeled training data set and features, the following 4 machine learning models were built using the data, and their performance was evaluated using F-(1) scores: naïve Bayes classifier, random forest, support vector machine, and long short-term memory. RESULTS: This study collected 146,352 data points and detected that 4.3% (6340/146,352) were firsthand experience data. Psychiatric problems (521/1683, 31%) had the highest frequency among the adverse effects. The highest frequency of nonmedical use was for studies or work (741/2016, 36.8%). While the frequency of nonmedical use before and after the outbreak of COVID-19 has been similar (odds ratio [OR] 1.02 95% CI 0.91-1.15), its trend has changed significantly due to the pandemic (95% CI 2.36-22.20). Among the machine learning models, RF had the highest performance of 0.75. CONCLUSIONS: The trend of nonmedical use of methylphenidate has changed significantly due to the COVID-19 pandemic. Among the machine learning models using SNS data to analyze the adverse effects and nonmedical use of methylphenidate, the random forest model had the highest performance.