Candyflipping and Other Combinations: Identifying Drug-Drug Combinations from an Online Forum

Abstract

Novel psychoactive substances (NPS) refer to synthetic compounds or derivatives of more widely known substances of abuse that have emerged over the last two decades. Case reports suggest that users combine substances to achieve desired psychotropic experiences while reducing dysphoria and unpleasant somatic effects. However, the pattern of combining NPS has not been studied on a large scale. Here, we show that posts discussing NPS describe combining nootropics with sedative-hypnotics and stimulants with plant hallucinogens or psychiatric medications. Discussions that mention sedative-hypnotics most commonly also mention hallucinogens and stimulants. We analyzed 20 years of publicly available posts from Lycaeum, an Internet forum dedicated to sharing information about psychoactive substance use. We used techniques from natural language processing and machine learning to identify NPS and correlate patterns of co-mentions of substances across posts. We found that conversations mentioning synthetic hallucinogens tended to divide into those mentioning hallucinogens derived from amphetamine and those derived from ergot. Conversations that mentioned synthetic hallucinogens tended not to mention plant hallucinogens. Conversations that mention bath salts commonly mention sedative-hypnotics or nootropics while more canonical stimulants are discussed with plant hallucinogens and psychiatric medications. All types of substances are frequently compared to MDMA, DMT, cocaine, or atropine when trying to describe their effects. Our results provide the largest analysis to date of online descriptions of patterns of polysubstance use and further demonstrate the utility of social media in learning about trends in substance use. We anticipate this work to lead to a more detailed analysis of the knowledge contained online about the patterns of usage and effects of novel psychoactive substances.

Keywords: computational biology; computer simulation; natural language processing; psychedelic drug use; toxicology.

Figure 1

Top 20 most frequently mentioned substances. x-Axis denotes number of posts in which the substance was mentioned at least once. MDMA, 3,4-methylenedioxymethamphetamine, also known as ecstasy; DMT, N,N-dimethyltryptamine; DXM, dextromethorphan; LSA, lysergic acid amide, also known as ergine.

Figure 2

Heat map of correlation coefficient of substance–substance co-mention pairs whose correlation was statistically significant. Each tiny box represents one pair of substances. Drug names on x- and y-axis specify the pair associated with each box. Color of tiny box indicates correlation, according to scale up in lower right.

Figure 3

Heat map of substance-substance co-mentions by class. Each tiny box represents one pair of substance classes. Labels on the x- and y-axes specify the substance classes associated with each box. Color of tiny box indicates absolute frequency of mentions, according to colorbar scale in lower right.

Figure 4

Social network of drug discussions. Each node (text) represents a substance. Each edge (connecting line) represents the correlation between mentions of the two connected substances. The thicker the line the stronger the correlation.