The study used 2 data sources to collect information about tramadol-related AEs: (1) the US Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS); and (2) posts in a health forum, WebMD. First, we used FAERS data as traditional pharmacovigilance data to find new AE signals. Globally, the United States is the country with the highest number of opioid analgesics consumed [20], and FAERS is the representative spontaneous reporting system with the largest number of publications related to tramadol-related AEs [21,22]. FAERS contains information about AEs and medication error reports submitted to the FDA, coded in the Medical Dictionary for Regulatory Activities (MedDRA) terminology. Reporting to FAERS is voluntary for health care professionals and consumers, while it is mandatory for manufacturers [23]. Next, WebMD, one of the popular health forums, was chosen to obtain more detailed patient narratives of AE symptoms. The AEs were investigated using various types of social media such as social networking sites (Facebook, Twitter), blogs, forums, and comments [24]. Among them, the health forums have the advantage of effectively obtaining patient experiences online because they are specialized medical topics and have many patient users [25]. WebMD is a well-known health information service website in the United States that provides information on health and wellness topics, including drug information [26]. It also provides bulletin boards for posting personal reviews on specific drugs, frequently used as data sources for AE-related research [24,25].

We used personal information on age and sex for FAERS and WebMD data sets; additionally, occurrence country was used for FAERS data in the analysis. FAERS provides various personal information, including age, sex, occupation, and country where the AEs occurred in the demographic file. All data sets were released after anonymization according to the privacy policy of the US Department of Health & Human Services [27]. WebMD displays the patient reviews of specific medications with nicknames, ages, and sex. WebMD also has its privacy policy fully complying with data protection regulations and developed to provide a safe space for sharing health-related information online [28]. We carefully considered data security and user privacy, and this study was ethically approved by the KNU Institutional Review Board (KNU-2021-0401)

We downloaded AE reports for January 2016 to December 2020 from the FAERS website [29]. From the 7,843,727 reports, we identified duplicate reports by case ID and selected the most recent to eliminate duplicate data. This left us with 6,874,999 reports. From these, tramadol-related AEs were retrieved for both single- and multi-ingredient drugs, including searches of the various brand names used in different countries as well as the generic drug name (n=92,135). Finally, we excluded AE reports in which tramadol was not reported as the primary or secondary suspected drug to rule out the potential misclassification of other drug-related AEs (n=29,345). The specific drug names and numbers of AE reports by year are summarized in Multimedia Appendix 1.

Online patient reviews of tramadol from September 2007 to August 2020 were gathered from the WebMD website using a Python web crawler. Patient review data were collected for tramadol HCL as well as combination drugs using both generic and brand names. A total of 3917 posts were automatically collected. Most of these pertained to tramadol HCL (n=2762). For the combination drugs, the majority of posts pertained to Ultram (Ultram, n=704; Ultram ER, n=134; Ultracet, n=177; tramadol HCL-acetaminophen, n=132; Conzip, n=8). Some of the posts only provided a rating of the drug. These were excluded and we retained only those posts with detailed descriptions. We then excluded posts that did not relate to drug-related AEs, such as drug prescription information and advertisements. Finally, we reviewed the remaining 2062 posts and assigned annotations according to the types of AEs described. The collected data were independently reviewed by 2 researchers (SP & J-WK), and symptoms considered as AE were manually assigned preferred terms (PTs) for AE types from MedDRA. If the PTs assigned by the 2 researchers did not match, a PT was agreed upon through discussion.

To compare the distribution of tramadol-related AEs reported on FAERS with that of those reported on WebMD, we first categorized reports into respondent age groups. The age groups were children and adolescents (ages <19 years), adults (ages 19-64 years), and older adults (ages ≥65 years). The number of AEs was counted based on MedDRA’s PTs in both data sets. Two or more PTs reported in a patient were counted as different AEs. The distribution of AEs was explored using the system organ classes (SOCs) of the MedDRA.

Additionally, we performed a disproportionality analysis of the FAERS data to identify the detection of tramadol-related AEs in traditional pharmacovigilance data. Disproportionality analysis is a comparison of observed and expected values; in this case, for tramadol-related AEs [30]. We calculated the proportional reporting ratio (PRR), the reporting odds ratio (ROR), and the information component (IC), which are commonly used as disproportionality measures. The threshold criteria for signal detection of adverse drug reactions were defined as PRR ≥2, ROR ≥2, and the observed number of tramadol-related AEs ≥3. The signal criterion for the IC measurements was when the lower limit of the 95% CI was greater than 0.

New signals were defined as AEs satisfying the signal criteria that were previously unknown or incompletely described. Their status as new was ascertained from tramadol label information. The drug labels were retrieved from a regulatory authority database (FDALabel, FDA, USA) and medical resources software systems (UpToDate, Wolters Kluwer Health; Micromedex, IBM). Medical events such as death, disability, hospitalization, or life-threatening consequences were classed as serious AEs. All statistical analyses were performed using SAS, version 9.4 (SAS Institute Inc.).

It is necessary to identify unknown AEs to reduce drug-related health risks [31]. Many efforts have been made to systematically detect unknown AEs using both social media platforms and traditional pharmacovigilance systems [32,33]. The rich amount of publicly available health information that patients post online is an invaluable additional data source for postmarket safety surveillance [24,32]. Therefore, researchers have used data from both social media and pharmacovigilance systems to detect AEs, as each type of data source has its strengths and limitations [34-36].

FAERS is the national pharmacovigilance system used in the United States, while WebMD is among the social media platforms most frequently used in research to retrieve patient reviews of drug experiences. Thus, we used these 2 data sets to detect new tramadol-related AEs in this study. Our 2 data sets evidenced several AEs known to be symptomatic of serotonin syndrome, including altered mental status (eg, agitation, anxiety, hallucinations), autonomic instability (eg, arrhythmias, tachycardia, labile blood pressure, hyperthermia), neuromuscular abnormalities (eg, hyperreflexia, tremor, rigidity), and gastrointestinal symptoms (eg, nausea, vomiting, diarrhea) [37,38]. However, there were qualitative differences in the reported tramadol-related AEs between the 2 data sets.

The FAERS data set produced a more diverse range of AEs than WebMD, in terms of both the range of severity and the affected organ systems. For example, reports of overdose, death, and completed suicide were only found on FAERS. FAERS reports also included 4 more SOCs in which tramadol-related AEs were reported than WebMD. Further, all but 1 (pain) of the new tramadol-related AEs detected was only found in the FAERS data set. Patients’ reviews on WebMD were primarily concerned with mild AEs such as pruritus, constipation, and hyperhidrosis. These differences can be explained by the difference in sample sizes and AE reporters on the 2 sites. FAERS data had 10 times more reports of tramadol-related AEs than WebMD, allowing the inclusion of a wider range of AEs. This demonstrates the value of larger data sets when selecting social media platforms for drug-induced AE detection. Because AE reports on social media are only self-reported, serious AEs such as suicide and death cannot be reported on such platforms. However, because the risk perceptions associated with a given drug can differ between patients (mainly for WebMD) and health care professionals (mainly for FAERS), combining data from both populations may contribute to improved drug safety assessments.

The tramadol-related pain that was frequently reported in both data sets may be a previously unknown tramadol-related AE. The FDA drug label does not list pain among the potential tramadol-induced side effects. However, the statistical measures of disproportionality were over the threshold criteria of signal detection (PRR=2.2; ROR=2.2; IC=1.1). In addition, the patients’ descriptions indicated increases in pain after taking tramadol. This is to be distinguished from both the persistence of existing pain due to an ineffective analgesic effect and pain caused by withdrawal following drug discontinuation. Increased pain might indicate opioid-induced hyperalgesia, which is increased sensitivity to pain caused by exposure to opioids [39]. There have been several reported cases of tramadol-induced hyperalgesia in previous studies [40,41]. Our result suggests that clinical assessments that consider opioid-induced hyperalgesia may be of value when patients complain of increased pain following tramadol treatment [42].

We detected 20 new serious AEs possibly related to tramadol. A clear distinction between the different types of serious AEs was difficult due to the diversity of AEs and related organ systems; however, most seemed to be broadly symptomatic of vascular diseases (eg, splenic artery aneurysm, femoral artery aneurysm, iliac artery occlusion) and their complications (eg, cerebellar infarction, quadriplegia, hemiplegia, paraplegia, paralysis). Our analysis also detected various coronary-related AEs. These results are not presented here because myocardial ischemia is already listed on the FDA drug label. The biological mechanisms that might explain an association between tramadol and vascular disease are not clear. A possible explanation is that tramadol mediates vascular homeostasis and thrombosis formation by inhibiting the reuptake of serotonin, affecting the platelet aggregation process [43]. Previous in vivo and in vitro studies have shown that tramadol use may enhance plasma coagulation and inhibit platelet disaggregation [44,45]. This finding from our analysis suggests that additional caution may be indicated before the use of tramadol in patients with vascular diseases as well as coronary artery disease.

Our study had some limitations. First, our data may have produced nonrepresentative figures describing the occurrence of AEs due to limitations of spontaneous reporting systems such as under-reporting, selective reporting, biases in patient’s drug preferences, and heterogeneity of the reports of different reporters [46]. Second, the results of our disproportionality analysis cannot prove a causal relationship. This method used the occurrence of AEs related to other drugs in the data as a proxy for the background incidence of AEs [47]. Thus, it may have been influenced by absolute report numbers. Third, posts from WebMD may have been misclassified into the wrong PTs due to a lack of clinical information. For example, if a patient complained that a coma occurred after taking tramadol, it cannot be medically confirmed that tramadol caused the coma without clinical information. Fourth, personal information from our data sets is based on voluntary reporting, and information has not been verified. Therefore, we could not rule out misinformation on demographics. Fifth, the user coverage may differ between 2 data sets, especially for the country profiles.

Despite the several study limitations, our study has 2 main strengths. First, this study found new tramadol-related signals. Based on the essential feature of pharmacovigilance signal [14], these findings have political implications for preventing drug-related health harms through the early detection of adverse drug reactions. Second, this study found the additional possible AE, pain, by comparing data from social media and conventional pharmacovigilance. Because pain is the main indication for analgesics, including tramadol, it is difficult to classify it as an AE without considering the patient’s detailed symptom descriptions. Although FAERS is representative data for signal detection, detailed descriptions of AEs are not collected in this system. Thus, this study can be an example of the usefulness of patient drug reviews in social media in detecting unexpected AEs. However, the small sample size from WebMD prevented the detection of rare AEs by this means. To expand the utility of rare AE detection using social media, it would be necessary to use large amounts of social media data from several different sites, such as Twitter and Reddit. Social media data analysis using automatic natural language processing is a subject worthy of further research.