Topic Analysis of Datasets
Since each abstract can be represented by its topic with maximum weight, we analyzed the topic distribution in each dataset to provide an overview of opioid-related research aspects.
1) Prescription Opioid: Fig. 4 shows the distribution of assigned topics within the prescription opioid dataset. The five most prevalent topics were T6, T4, T21, T30, and T8. The corresponding word clouds for these topics are shown in Fig. 5. T4 shows the “patient”, “dose”, and “prescribe” with the general terms associated with opioid prescription. T21 was primarily concerned with “physician”, “prescribe”, and “practice”. T30 focuses on “chronic pain” and “patients”, and T8 may demonstrate topics associated with the increased risk for death from opioid overdose. The most dominant topic, T6, is assigned to 16.5% of the abstracts in the dataset which is nearly three times the number of abstracts assigned to the second most populous topic (T4). This trend indicates that many opioid-related research efforts have focused on the abuse and health risks of prescription opioids.
2) Codeine: According to the distribution of assigned topics for the codeine dataset (Fig. 6), the five most popular topics were T4, T33, T12, T35, and T13. The corresponding word clouds for these topics are shown in Fig. 7. T4 and T33 contain high-weighted terms such as: “extraction”, “detection”, “drug”, “clinical”, and “treatment”; T12, T13, and T35 show that the research on codeine has been closely related to pain-patient associations, such as the post-operative pain following surgery.
3) Morphine: According to the distribution of assigned topics for the morphine dataset (Fig. 8), the five most popular topics were T19, T26, T12, T13, and T6. The corresponding word clouds for these topics are shown in Fig. 9. The most dominant topic, T19, was assigned approximately 9% of the abstracts in the dataset, with focus on “patients”, “post-operative pain”, and “treatment”; while the second top topic, T26, encompassed “cancer patient pain treatment”, and T12 primarily involved “epidural morphine” and “analgesics”. Topic T13 involved the “clinical drug treatment” and T6 contained aspects related to “opioid withdrawal”, including “naloxone”, “morphine” and “rat experiments”.
4) Hydrocodone: According to the distribution of assigned topics for the hydrocodone dataset (Fig. 10), the five most popular topics were T7, T9, T10, T12, and T2. The corresponding word clouds for these topics are shown in Fig. 11. They include not only the opioid-related common terms such as “prescription opioid: (T7), “drug” and “pain treatment” (T10), and “patient pain” (T12), but also focus on the “opioid metabolites” and “urine/specimen sample detection” (T9) and “comparison” of “hydrocodone” with “other analgesic medications” on “pain relief” (T2).
5) Oxycodone: According to the distribution of assigned topics for the oxycodone dataset (Fig. 12), the five most popular topics were T18, T25, T27, T14, and T23. The corresponding word clouds for these topics are shown in Fig. 13. T18, T25, and T14 were similar to some topics of other opioid datasets; T27 includes the comparison between “oxycodone” and “tapentadol” on the effect of “pain relief treatment”; and T23 contained the terms of “drug abuse” along with increased use of “prescription opioids”.
6) Methadone: According to the distribution of assigned topics for the oxycodone dataset (Fig. 14), the five most popular topics were T11, T32, T35, T3, and T15. The corresponding word clouds for these topics are shown in Fig. 15. The top three most popular topics, T11, T32, and T35, encompassed common topics such as “addiction”, “drug dose”, and “abuse in prescription opioid use”; T3 included a special concern regarding the possible effects of “methadone use” on “pregnant women” or “infants” such as “Neonatal Abstinence Syndrome (NAS)”. T15 shows that much research has been pursued on the relationships between “methadone maintenance treatment” with “HIV-related risk behaviors”.
Similarities
Although word clouds for the top 5 topics were analyzed separately for each dataset as above, checking word clouds for all topics (Supplemental Information Fig. 1) suggested that most topics for all six datasets could be grouped further into the following categories
-
Opioid/Drug Prescription: topics containing general search terms (drug name) and opioid/drug prescription-related issues, such as increasing trend of dose.
-
Patient/Pain: topics involved in opioid-use patients with pain resulting from conditions such as cancer, chronic, surgery post-operative pain.
-
Misuse/Abuse: topics related to problems in using these opioids such as overdose leading to death.
-
Adverse/Side Effects: topics associated with side effects from using these opioids such as RLS (Restless Legs Syndrome), NAS (Neonatal Abstinence Syndrome).
-
Physician/Clinical Treatment: topics indicating that research on these opioids focused on practical experiments and treatments, especially those addressing opioid use disorder treatment.
-
Gender/Age or Woman/Child: topics suggesting that these opioids may have gender and age disparities in pregnant women and infants.
-
Genotype/CYP (Cytochrome P450): topics related to research on opioid metabolism and genotype.
-
Review: topics related to current research reviews focused on prescription usage.
-
Differences: We note that although most topics in all six datasets shared common categories/themes as described above, they were assigned different ratios of abstracts in each dataset. This explains why the top 5 topics were different for each dataset. For example, although the topic “pregnant women and infants” appeared in all datasets, this topic appeared in the top 5 topics only for the methadone dataset.
Furthermore, compared to other opioids, methadone was the only opioid which had several topics related to other factors (e.g., alcohol, smoking, heroin and cocaine use, HIV/AIDS, and Hepatitis C Virus (HCV)). Therefore, topics in the category of Physician/Clinical Treatment for methadone involved alcohol/smoking cessation/intervention, or cocaine treatment, or Opioid Substitution Therapy (OST) for HIV prevention. The association of these factors with methadone might partially be due to methadone being a synthetic opioid, which associates it with illegal drugs (e.g., heroin, cocaine). This is in contrast to codeine and morphine which are natural opioids or hydrocodone and oxycodone which are semi-synthetic opioids.
Another difference among the six datasets is in the category “Adverse/Side Effects”. The side effects in the codeine dataset contained more detailed symptom keywords such as cough, headache, bowel, renal, and side effects. The oxycodone dataset also contained more detailed symptom keywords such as bowel, renal, sleep, and RLS. Therefore, we may possibly conclude that different opioids may cause different side effects and/or the consequences vary between opioids and patients.