Discussion
Our results show that duration of treatment has a significant impact on the distribution of reported time-to-onset in spontaneous reporting systems in pharmacovigilance. Notably, the median of the median RTTO of ‘angioedema’ for long-term treatments (8 days) exceeded that of ‘hepatitis’ for short-term treatments (7.5 days). In addition, the RTTO for hepatitis for the short-term treatments (range 4-11 days) is substantially shorter than that observed for an overlapping set of drugs in the prospective study by Chalasani et al. (range 20-117 days). These findings suggest that aggregated time-to-onset for short-term treatments cannot be reliably evaluated in spontaneous reporting systems without taking into account the expected duration of treatment. The shorter than expected RTTO for hepatitis with short-term treatments must be interpreted bearing in mind that adverse events are less likely to be reported as suspected ADRs after the end of treatment. Thus, this does not rule out the possibility that they represent causal associations. Whereas we anticipated short-term treatments to exhibit shorter than expected reported time-to-onset, we had not anticipated the longer than expected reported time-to-onset of angioedema for the long-term treatments. Fluoxetine for example presented a median RTTO of 26 days, with RTTO exceeding 30 days for more than 25% of the reports. While long time-to-onsets may be associated with imprecise information more often, our sensitivity analysis indicated that if such an effect was present it did not have a major impact on our results.
When categorizing drugs into short term (up to 3 weeks) and long term (3 months or more) treatment groups, some drugs were not easily classified. The non-steroidal anti-inflammatory drug ibuprofen for instance was classified as a short term analgesic, while it is known to be used long term in the treatment of chronic inflammations for example rheumatoid arthritis and arthrosis [16]. In an attempt to circumvent the possibility of varying durations of treatments within each drug, a restriction was made to reports with a specified indication for the suspected drug. However, this approach proved intractable, as the numbers of reports on either angioedema or hepatitis for each drug were too low to allow for robust analysis. Despite the special care taken in selecting drugs from different pharmacological classes, the panel of drugs for short term treatments was dominated by antibacterials (ATC J01). The broader group of antimicrobials (J) has previously been observed as the single largest class of agents causing drug induced liver injury [13]. Nevertheless, two of six short term treatments did not belong to the ATC-class (J), confirming that the short observed RTTO were not likely due to a class effect.
Confounding by underlying infection (viral hepatitis) cannot be excluded as an explanation for the shorter RTTO for short term treatments. The studied treatments consist of antibacterial drugs, with the exception of ibuprofen (non-steroidal anti-inflammatory drug) and paracetamol (anilid). However, the RTTO for the two non-antibacterials are also short and it is unlikely that the majority of reports for the antibacterials should be for patients with a viral infection misdiagnosed as bacterial.
VigiBase pools data originating from a range of professional settings, geographical regions and time periods with varying legislation and tradition of reporting. The diversity of reporters includes health care professionals, patients and pharmaceutical companies. Different criteria and routines for receiving, recording and managing reports prior to submission to VigiBase [4] result in an intrinsically heterogeneous collection of data where it is likely that the RTTO varies with the type of reporter. A suggestive time-to-onset is a major argument for suspecting and reporting an adverse event as a suspected ADR. However, the level of suspicion for the reported drug-ADR pairs in VigiBase is not specified for all reports and will vary within the dataset.
A previously discussed phenomenon that may affect the output of aggregated time-to-onset information is how the reporter has defined the onset date of the ADR, which can be recorded as when the first symptom started, when the diagnosis was made, or when the ADR became serious and thereby reportable by the pharmaceutical companies [3]. As ADRs may develop certain seriousness criterion over time, e.g. hospitalization, the RTTO will inherently be longer than the actual occurrence of the ADR. The generally quick onset of angioedema requiring hospitalization would be less likely influenced by this reporting bias, whilst for hepatitis the latency would be more likely prolonged by this bias rather than shortened, as was the tendency for both short-term and long-term treatments in this study.
Our study of the influence of treatment duration on RTTO was limited to two ADRs: angioedema and hepatitis. Further work is required to evaluate to what extent the phenomenon affects other ADRs. Treatments without duration, i.e. vaccinations or anesthesia, obviously follow a different pattern where the likelihood that an adverse event is reported as a suspected ADR can be expected to decrease more gradually from the time of treatment.