Paediatric safety signals identified in VigiBase: Methods and results from Uppsala Monitoring Centre

Abstract Purpose The purpose of this study is to uncover previously unrecognised risks of medicines in paediatric pharmacovigilance reports and thereby advance a safer use of medicines in paediatrics. Methods Individual case safety reports (ICSRs) with ages less than 18 years were retrieved from VigiBase, the World Health Organization (WHO) global database of ICSRs, in September 2014. The reports were grouped according to the following age spans: 0 to 27 days; 28 days to 23 months; 2 to 11 years; and 12 to 17 years. vigiRank, a data‐driven predictive model for emerging safety signals, was used to prioritise the list of drug events by age groups. The list was manually assessed, and potential signals were identified to undergo in‐depth assessment to determine whether a signal should be communicated. Results A total of 472 drug‐event pairs by paediatric age groups were the subject of an initial manual assessment. Twenty‐seven drug events from the two older age groups were classified as potential signals. An in‐depth assessment resulted in eight signals, of which one concerned harm in connection with off‐label use of dextromethorphan and another with accidental overdose of olanzapine by young children, and the remaining signals referred to potentially new causal associations for atomoxetine (two signals), temozolamide, deferasirox, levetiracetam, and desloratadine that could be relevant also for adults. Conclusions Clinically relevant signals were uncovered in VigiBase by using vigiRank applied to paediatric age groups. Further refinement of the methodology is needed to identify signals in reports with ages under 2 years and to capture signals specific to the paediatric population as a risk group.


| INTRODUCTION
To minimise harm from medication use, healthcare professionals and patients need to know about the risks. Postmarketing reporting systems provide opportunities to increase our knowledge of risks that were not recognised in the premarketing clinical trials. The knowledge gained will constitute the basis for prevention and mitigation of patient harm by manufacturers, regulatory authorities, and healthcare institutions.
To detect previously unrecognised rare adverse drug reactions To use the opportunities of the broad coverage in VigiBase, UMC has begun to screen for safety signals in subgroups, eg, a specific type of reporters such as patients, 4 geographical areas, and drug groups.
Disproportionality analyses (used to highlight statistical associations for further evaluation 5,6 ) in subgroups of reports have been shown to uncover previously unknown associations 7 and even improve performance compared with using the complete data set of a postmarketing reporting system. 8 The first subgroup to be the subject for signal detection screening at UMC was reports within the paediatric age group.
Information on the safety and efficacy of a medicine used for neonates, infants, children, and adolescents is limited if individuals with these ages were not included in the premarketing clinical trials. Drug toxicity is poorly reported in paediatric clinical trials, 9,10 particularly where clinical trials involve both adults and children. 11 As a consequence, information on dose recommendations, precautions, warnings, and ADR profiles specific to paediatric age groups can be lacking when prescribing and administering medicines to these patients. 12 Children experience a wide range of ADRs, as described from national pharmacovigilance databases, 13 and the reporting pattern differs both from reports for adults and between paediatric age subgroups. 14 In order to increase knowledge for the safer use of medicines in the paediatric population, VigiBase reports were screened to uncover previously unrecognised risks of medicines in this age group.

| Scope
Signal detection and assessment at UMC follow a three-step process as described in Table 1. The adjustments made to capture signals in reports on paediatric ages are further described for each of these steps in this section.
"Drug-event pairs" refer to clusters of ICSRs (denoted "reports" in this paper) with the same suspect or interacting drug and the same event. The preferred base level (active ingredient) in the WHO Drug dictionary was used to classify the "drug," and the preferred term from the WHO-Adverse Reaction Terminology (WHO-ART) was used to classify the "event" in the signal detection screening of reports in the paediatric age group.
If the in-depth assessment suggests that a signal should be communicated, the hypothesis is presented with data and arguments 16  and was restricted to reports with ages less than 18 years. Figure 1

KEY POINTS
• Signal detection of global individual case safety reports for paediatric age groups uncovered previously unrecognised risks of medicines.
• Three signals were further evaluated and subsequently added to the product label, providing new information for patients, caregivers, and healthcare professionals to consider prior to and during therapy. Exclusion and inclusion criteria are applied to a designated data set, and data-driven methodologies and filters are implemented to a list of drug-event pairs (coupled with or without a subgroup) or drug-drug events.
II Initial manual assessment UMC assessors manually review the list of drug-event pairs to identify potential signals to undergo in-depth assessment. The review includes the following: • checking of whether the event is already well-described in the product information; and • a brief review of the individual reports to exclude report series that displays an obvious alternate and more likely explanation for the association or lacks sufficient information for assessment.
III In-depth manual assessment UMC staff or external experienced scientists and clinicians conduct causality assessment of the individual case reports and review the literature on the topic to compile evidence for or against a signal. 15 Then, a decision is made whether or not the strength of the report series supports the communication of a signal.
displays the number and type of reports that were included and excluded from the data set. Suspected duplicate reports were excluded by using the vigiMatch, an algorithm for automatic duplicate detection. 18,19 Reports on vaccines were excluded because vaccine reports were designated to a separate screening to allow for capturing age-independent reports, hence not restricted to paediatric ages.
Reports recording harm of newborns resulting from in utero exposure are not sufficiently captured in a data set restricted to paediatric ages, since these reports can be given with the mother's age or with no age specified; therefore, reports indicating in utero exposure were excluded and designated to a separate review.

| Filters applied to the drug-event paediatric age-group list
The DEAG list included 55 108 posts for review, so we applied and tested four different filters in the initial manual assessment as presented in Table 3. At previous screenings of VigiBase, WHO-ART critical terms (considered as being indicative of seriousness) had been used to prioritise report series for review. 22 We were uncertain whether serious problems specific to paediatric ages were captured with critical terms. Diarrhoea, which can be a life-threatening event for an infant, was, for example, not designated to be a critical term. Therefore, we identified report series referring to serious events using the International Conference of Harmonisation (ICH) seriousness criteria, 23 flagging DAEGs where all reports in E2B format fulfilled one of the ICH seriousness criteria.

| Initial manual assessment
The aim of the initial manual assessment was to identify potential signals that should proceed to in-depth assessment. A multidisciplinary team of pharmacists, nurses, data scientists, and physicians (including a paediatrician/clinical pharmacologist) manually assessed the DEAG list. The assessors could select all paediatric age groups with the same drug event in the same assessment and thereby occasionally deviate from the prioritisation by vigiRank. The product information was scrutinised to determine whether the drug-event pair should be considered "known" for the age group. If the event was labelled but the drug was not approved for the specific age group, additional sources,  The lower threshold was set to enable enough reports for assessment, and the limit to the maximum number of reports was set to increase the likelihood of capturing previously unknown problems (ie, signals) as well as rare adverse drug reactions, which holds the primary purpose of the international compilation of reports in VigiBase.
Single-reported country To complement national centres by focusing on global problems No reports received in VigiBase after 1 January 2012 To be relevant and capture current problems used by paediatricians, were reviewed to check whether these sources had listed the event and therefore could be considered to be known.
For safety labelling and approval status, the UK Summary of Product Characteristics 26 and product labels of drugs approved by the US Food and Drug Administration (FDA) 27 were consulted. The British National Formulary for Children 28 or NeoFax and Paediatrics 29 were referenced to represent sources used by paediatricians. Whenever the drug could not be found in the aforementioned sources, DrugDex, 30 Martindale, 31 or other national product information was reviewed.
Each post in the DEAG list was categorised according to a decision tree ( Figure 3) and recorded with any of the following outcomes: • Potential signal, needing further in-depth manual assessment • Known, considered well-described for the specific age group in the product information • Nonsignal, report series suggests alternative more likely explanations for the event, such as coreported drugs, or lacks sufficient/relevant data for assessment • Keep under review (KUR), needing time to gather more/better documented reports FIGURE 2 Extract from drug-event age group (DEAG) list used during initial manual assessment. Twenty-four thousand two hundred fifty one denotes unique drug-event pairs for the complete paediatric data set with ages less than 18 years after applying exclusion criteria specified in Table 2 [Colour figure can be viewed at wileyonlinelibrary.com] The negative disproportionality measure referred here is based on the negative lower end point of the 95% credibility interval of the Information Component 5,6 (IC025 < 0) and denotes less reporting than expected in the full VigiBase data set. During the assessment, a system for graphical overview and access to the global individual reports was used in addition to statistics displayed for the DEAGs via a separate software interface with information such as the following: • selected information tabulated for each individual report, including narratives (when available) • summary statistics for the report series such as • number of reports • age range • dates for when the first report on the drug and for the drug event was entered into VigiBase in the specific age group

| In-depth manual assessment
Clinical experts from the UMC or external expert volunteers, who are part of the UMC Signal review panel, assessed the individual reports (representing the potential signals) in depth and reviewed the literature to compile evidence for or against a signal. 15 The assessors could choose to restrict their assessment to reports on paediatric ages or include reports for other ages as well.

| RESULTS
A total of 55 108 DEAGs were retrieved after applying the inclusion and exclusion criteria described in Figure 1 and Table 2. The median number of reports representing the DEAG posts was three reports, and 45% contained one to two reports (83% and 69% for the neonates and infants and 36% and 37% for the children and adolescents, respectively).
During the initial manual assessment, 472 DEAG posts (392 unique drug-event pairs) were reviewed (neonates = 13; infants = 37; children = 217; adolescents = 205). The number and proportion of potential signals, as well as known ADRs for the age group/ nonsignals/KURs, are displayed in Table 4. A total of 27 DEAGs and 21 unique drug-event pairs were classified as potential signals requiring in-depth manual assessment. The potential signals were all identified in the two older paediatric age groups.
The filters applied to the DEAG list and that generated the greatest rate of potential signals were DEAG posts with a negative disproportional pattern in the full VigiBase data set and with labelled reactions excluded and serious events with new drugs (≥2009). A total of 10% and 7% of the DEAGs, respectively, were classified as potential signals when these filters had been applied in comparison with approximately 4.5% to 5% for the other filters, see Table 5.  Table 6. In all but the two patient safety signals concerning off-label use and accidental overdose, the drug-event report series was listed with a negative disproportional pattern in the full VigiBase data set.

| DISCUSSION
Clinically relevant signals were identified by screening paediatric age groups in VigiBase. Of the eight signals, 32-39 two concerned harm in connection to off-label use and accidental overdose of medicines by young children. Six signals referred to new potentially causal associations, of which labelling changes have subsequently been made for three. [40][41][42] The signals suggesting new associations for atomoxetine, temozolamide, deferasirox, levetiracetam, and desloratadine were also relevant for adults. None of these signals had a positive disproportional reporting pattern in the full VigiBase data set and had not previously been reviewed when using the full data set and disproportionality analyses as the screening method. So, in addition to finding signals relevant for the paediatric population, the focus on paediatric age groups enabled issues to be uncovered that are important for any age group.
In two signals, young children were identified as a particular risk group. These signals demonstrated the need for further action to prevent parents from administering dextromethorphan to young children despite warnings raised by authorities and to initiate measures to prevent accidental intake of antipsychotic medicines by young children.
Historically, signal detection and assessment at UMC have focused on finding previously unknown associations between events and drugs, ie, compiling evidence for or against a causal relationship. In the review based on the paediatric ages, the patient group and its context were in focus as a possible risk group in addition to exploring a causal relationship. This required a different mindset when assessing the report series. In setting up for the current signal detection review, much effort and time were put into defining the scope and adjusting current statistical screening methodology to the paediatric population.
In retrospect, however, more effort should have been made on guiding the assessors to capture signals specific to paediatric ages as a particular risk group.
No signals originated from the two youngest age groups (0-27 d and 28 d-23 mo). The time spans of these age groups (1 and 23 mo, respectively) are much shorter than for the two older age groups (10 and 6 y, respectively). Hence, the younger age groups have fewer reports overall, fewer drug-event pairs, and fewer reports per drugevent pair (83% and 69% of the pairs in these age groups were based on only one to two reports), resulting in lower vigiRank scores. Consequently, the vigiRank scores for the older age groups were competing with the scores for the younger age groups when the drug-event pairs for the paediatric age groups were combined into the DAEG list. To better account for the younger ages in the future, these reports could be combined into one age group (<2 y), and/or less granular medical terms could be used to increase the number of reports per drug-event pair. Alternative solutions could be to review drug events in these age groups separately to avoid a dominance of drug events from the older age groups or randomise an equal number of drug-event pairs from each age group to undergo assessment.
Drug events by the four paediatric age groups were prioritised by vigiRank, and well-described ADRs for the age groups were excluded from further evaluation. The proportion of identified signals based on the paediatric data in VigiBase was lower than a previous signal detection screening using the vigiRank on the full VigiBase data set 43 (2.0% vs 3.1%). Known ADRs had a higher frequency in the paediatric data set (63% of all unique drug-event pairs) compared with a previous screening of the full data set (41%), 43 but when excluding the known ADRs from the denominator in the current and previous data sets, a similar proportion of identified signals was identified (5.5% vs 5.2%). The higher proportion of known ADRs for the paediatric drug events could be explained by that they represented large report series in the adult data set, hence increasing the likelihood of being known ADRs.
In the paediatric signal detection screening, we restricted the number of reports to 30 per drug-event pair within the ages 0 to 17 years for reasons set out in Table 2. This restriction in the number of reports might have been too conservative, possibly resulting in missing important signals among paediatric reports. In future screenings, this limit to the number of reports should be reconsidered.
New initiatives are encouraged within paediatric pharmacovigilance. The European Medicines Agency (EMA) recently developed a statistical query that can be applied to EudraVigilance to highlight imbalances in reported drug-event pairs for children as compared with adults. The query has been used to support EMA Scientific Committees in their investigation of specific paediatric safety concerns. 44 The current description of signal detection of paediatric age groups refers to safety concerns identified via large-scale hypotheses generation of VigiBase data and presents how these findings were identified using the vigiRank.
It should be noted that the signals presented in this paper are preliminary in nature and their status can change over time when more data on the problems identified are available. Also, the basis for the signals is a global pharmacovigilance reporting system, which has known limitations, such as the information being from a variety of sources, and the likelihood that the suspected adverse reaction being drug related is not the same in all cases. 45 Spontaneous reporting systems have known strengths and limitations. 46 Underreporting causes signals to be missed because ADRs are not always recognised or reported by caregivers, patients, or health Serious events a (no restriction to drug) 377 19 Serious events a and new drugs in age group later than or equal to 2005 b 221 10 Negative disproportionality measure overall c and not labelled d 126 13 The same drug-event pairs (coupled with any or several of the four paediatric age groups) can be accounted for in more than one screening filter. a Reports series referring to serious events using the ICH seriousness criteria, 23 flagging pairs where all reports in the E2B format fulfilled one of the ICH seriousness criteria.
b New drugs were defined as drugs first reported to VigiBase in the specific paediatric age group on/after 1st of January 2009 or on/after 1st of January 2005. c The negative disproportionality measure referred here was based on the negative lower end point of the 95% credibility interval of the Information Component (IC025 < 0) 5,6 and denoted less reporting than expected in the full VigiBase data set.  • Twenty reports with atomoxetine as the sole suspected drug.
• Time to onset (eleven cases) ranged from 14 days to 10 months.
• Seven cases recovered after drug withdrawal (based on available records).
Dextromethorphan and serious neurological disorders 36 April 2015/2015 no. 5 Used in many cough, cold, and flu products approved from age 4 and 6 years in the United States and the United Kingdom, respectively 15 vs 1.6 reports with ataxia/ 2-11 y/ positive 110 reports with neurological disorders for ages less than 6 y • Continuous to be reported also after 2008/2009 when a recommendation that these ages should not be given this medicine was issued.
• Serious reactions such as ataxia, convulsions, dyskinesia, and coma had been reported.
•Majority of reports were not reported with accidental intake of drug or overdose.
• Better warnings for parents suggested.
Olanzapine and accidental drug intake by children 37 April 2015/2015 no. 5 For schizophrenia and bipolar disorder in adults 11 vs 0.2 reports with miosis (majority coreported with accidental exposure)/ 2-11 y/ positive 20 reports with medicationerror-related problems (eg, accidental exposure)/ below the age of 6 y • Reminder to consider the risk of accidental exposure to young children when prescribing these drugs to patients with potential decreased risk awareness.
• Packaging is crucial to protect children from accidental exposure to these very potent drugs. • Coreported with other drugs and terms.
• Oesophagitis plausible when considering other labelled events such as stomatitis, dysphagia, and gastroenteritis.
Deferasirox and pancreatitis 34 June 2015/2015 no. 6 Chronic iron overload because of frequent blood transfusions in patients with beta thalassaemia major 9 vs 2.7 reports/ 12-17 y/ negative 14 reports/ 4-16 y/ five countries (any ages: 62 reports) • Eleven reports with deferasirox as the sole suspected drug.
• Time to onset (nine cases): 17 days to 5 years with a median of 11 months.
• Six cases recovered after drug withdrawal (based on available records).
• Four isolated cases from published clinical trials (two concerned elevated amylase levels).
• Company responds by adding pancreatitis to the label. 42 Levetiracetam and impaired renal function 38 June 2015/2016 no. 2 Epilepsy 13 vs 2.9 reports with acute renal failure/ 12-17 y/ negative 14 paediatric reports/ 6 wk-17 y/ five countries/ twelve well-documented reports for adults assessed/ more than 100 reports with any ages • Signal covered reports with acute renal failure, azotaemia, renal function abnormal, and/or interstitial nephritis.
• Several published case reports had been described for paediatric ages and adults.
• Three of twelve well-documented reports for adults described increasing creatinine levels after starting levetiracetam followed by a reduction to normal after drug withdrawal.
• Acute kidney injury added to the label. 40 (Continues) professionals. Another weakness of the system is the possibility of various reporting biases, for example, unexpected overreporting because of media attention or because the medicine is undergoing intensive monitoring, which can influence quantitative analyses. Also, the in-depth assessment of individual reports can be restricted because of poor quality data or lack of data. However, the method used for identifying drug-event pairs for further evaluation at UMC, vigiRank, prioritises not only disproportionate reporting patterns but also informative report series and thereby increases the chances of a conclusive case assessment. 21,43

| CONCLUSIONS
Clinically relevant signals were uncovered in VigiBase by using the vigiRank applied to paediatric age groups. Three of these safety concerns were subsequently added to the product label, providing new information for patients, caregivers, and healthcare professionals to consider prior to and during therapy. Further refinement of the methodology is needed to identify signals in the youngest paediatric age groups and to capture signals specific to the paediatric population as a risk group.

ETHICS STATEMENT
The authors state that no ethical approval was needed. 115 reports with loratadine, of which eleven cases aged 4-10 y • Assessment covered reports with desloratadine and loratadine with any age, with a focus on ages less than 12 years (paediatric cases in this signal).
• Reports covered weight increase, obesity, and appetite increased.
• Paediatric cases with desloratadine: five reports with sole suspect drug; two cases with a positive dechallenge and rechallenge.
• Paediatric cases with loratadine: nine reports with sole suspected drug; five cases recovered after drug withdrawal, of which one was reported with a positive rechallenge.
• Desloratadine and increased appetite were signalled by The Netherlands Pharmacovigilance Centre Lareb 48 in 2011.
• Weight increased 41,49 and appetite increased 41 added to the label. The positive/negative disproportionality measure referred here was based on the positive/negative lower end point of the 95% credibility interval of the Information Component 5,6 and denoted more/less reporting than expected in the full VigiBase data set.