Methods
In order to evaluate the relationship between duration of treatment and reported time-to-onset, we contrasted the distribution of reported time-to-onset of angioedema (with an expected short latency) and hepatitis (with an expected longer latency) for a set of drugs expected to be used as short-term treatments with a set of drugs expected to be used as long-term treatments.

VigiBase
The WHO Global ICSR database, VigiBase [4], holds reports on suspected ADRs from more than 100 countries around the world. The reporting countries are members of the WHO Programme for International Drug Monitoring and contribute reports from their respective national ADR reporting system. This study was based on reports up until February 5th 2010, when a total of 4,978,565 reports were accumulated in VigiBase. The reported drugs are encoded using the WHO Drug Dictionary Enhanced, which uses the WHO Anatomical Therapeutic Chemical (ATC) classification. The selected drugs were studied at the preferred base level. ADRs are encoded in the WHO Adverse Reactions Terminology (WHO-ART) and the Medical Dictionary for Regulatory Activities (MedDRA) in parallel. The latter is built up by five levels of hierarchy [5], and this study was carried out at the level of MedDRA preferred terms (PT), which is the fourth of those.

Reported time-to-onset
In the context of this study we refer to the time span between reported start of drug therapy and reported onset of the ADR as the reported time-to-onset (RTTO), acknowledging the variable information that this may represent [1,3].

Selection of adverse drug reactions
To examine the effect of duration of treatment on RTTO, angioedema and hepatitis were selected as two well-studied ADRs, providing examples of mechanistically diverse conditions which differ in expected latency.
Angioedema is typically an abrupt ADR, classified according to the underlying pathophysiology [6]. Allergic histamine-mediated and pseudoallergic cyclooxygenas inhibition-mediated angioedema occur within hours of exposure to the offending drug, the former being dependent on prior sensitization. However, bradykinin-mediated angioedema associated with angiotensin-converting enzyme inhibitors (ACEI) is an exception, as it may occur during the first week of therapy or be delayed up to several months [7]. None of the other drugs studied in our analysis (see section 2.4) have been documented to induce angioedema by a bradykinin-mediated mechanism. Therefore the time-to-onset for angioedema is expected to be short for the majority of the studied drugs.
Drug induced hepatitis is commonly divided into intrinsic and idiosyncratic reactions [8], where the dose dependent reactions in the intrinsic category mainly are caused by acetaminophen toxicity [9,10]. In contrast, idiosyncratic hepatitis has no immediate relationship to dose, although a dose threshold has been suggested [11,12]. Among the proposed underlying mechanisms, immune-mediation is suggested to cause liver injury with latencies as short as 1-8 weeks, while non-immune mediated reactions may take up to one year [9]. The median time-to-onset of drug-induced liver injury was found to be 42 days (range 20-117 days) in a prospective observational study [13]. However, biochemical signs of drug induced hepatitis by acetaminophen is known to present within 24 hours from intoxication [14].
The basis for our analysis were the MedDRA Preferred Terms (PTs) describing the two selected ADRs: ‘Angioedema’ and ’Hepatitis’. These PTs contain a range of Lowest Level Terms (LLTs) reflecting the same medical concept as the corresponding PT, expressed by synonyms and lexical variants [15]. After a review of all LLTs for both of the PTs in MedDRA, LLTs indicating that the ADRs had been aggravated were excluded e.g. ‘Angioedema aggravated’, ‘Angioneurotic edema aggravated’, ‘Angioneurotic oedema aggravated’ and ‘Hepatitis aggravated’. In addition, LLTs indicating too unspecific conditions such as ‘Edema vascular’, ‘Oedema vascular’ and ‘Hepatitis in other infectious diseases classified elsewhere’ were removed from the dataset, as well as LLTs referring to a specific inappropriate subtype of the ADR e.g. ‘C1 esterase deficiency acquired’ and ‘Syncytial giant cell hepatitis’. An overview of the studied MedDRA PTs and included LLTs after applying the exclusion of the above mentioned are presented in Table S1.

Selection of short-versus long-term treatments
A review of the top reported drugs for angioedema and hepatitis was performed. Among these, six substances (or combined substances reported as combination products) representing expected short- and six representing long or continuous duration of treatment were selected, with a known potential of causing both studied ADRs. Short-term treatments were considered therapies with a recommended duration of therapy of three weeks or less, while long-term treatments were therapies with a recommended duration of therapy of three months or longer. Each included drug was checked for these durations of treatment according to the literature [16]. Only substances reported for both studied ADRs were chosen, to enable a comparison of the impact of duration of treatment on the RTTO. Special care was taken to select substances from different pharmacological classes and thus mechanisms of ADR induction. Drugs chosen as representatives of short-term treatments were erythromycin (macrolid), ibuprofen (nonsteroidal anti-inflammatory drug, NSAID), clavulanate potassium/amoxicillin trihydrate (penicillin/ β-lactamase inhibitor), sulfamethoxazole/ trimethoprim (sulfonamide/synthetic antibiotic agent), paracetamol (acetaminophen, anilid) and ciprofloxacin (fluoroquinolone). The representative substances chosen for long-term treatments were enalapril (angiotensin-converting enzyme inhibitor, ACEI), fluoxetine (selective serotonin reuptake inhibitor, SSRI), isotretinoin (retinoid), methotrexate (antimetabolite), simvastatin (HMG-CoA reductase inhibitor), and ticlopidine (platelet aggregation inhibitor).

Data extraction and criteria
Reports containing the defined terms were extracted from the database and subject to the following selection criteria. Firstly we focused exclusively on reports with a single suspected drug, to circumvent the complexity of multiple suspected drugs with different start dates. Secondly, suspected duplicates were removed using a previously described probabilistic record matching algorithm [17]. Furthermore, reports with imprecise relevant dates, i.e. lacking specified day or month for start of drug treatment or onset of suspected ADR(s) as well as reports with potentially imprecise dates that referred to 01 and/or 15 of a month were excluded. The latter dates are known to sometimes represent estimated dates by the national pharmacovigilance centre, where only the month or the year is known. Finally, to reduce the impact of data quality issues, reports with unspecified gender or age were excluded in addition to reports with negative RTTO or RTTO exceeding patient age.

Ethics statement
De-identified individual case reports have been routinely collected as a public health service internationally since 1968, through the WHO Programme for International Drug Monitoring. The protection of the identity of the patient and the reporter has been routine from the outset.

Statistical analysis
The reported time to onset for different drugs with angioedema and hepatitis was visualized in box plots, using the “bwplot” function in the lattice package of R [18]. In these plots, the box demarks the interquartile range, the medians are marked with a dot, and whiskers extend to the most extreme data point that falls within 1.5 times the interquartile range of either end of the box; outliers beyond the whiskers are not displayed. Comparisons between long- and short-term treatments in the reported time-to-onset for each ADR were made with non-parametric Mann-Whitney tests. The null hypothesis is that there is no difference in reported time-to-onset between the two groups of drugs, for each respective ADR. The “wilcox. test” R function for a two-tailed and unpaired Wilcoxon rank sum test [19] was used, which is equivalent to the Mann-Whitney test [20]. Formal survival analysis was not possible, lacking information on the number of patients at risk at a given point in time after initiation of treatment.