Adverse Drug Reactions in Children—A Systematic Review

Background Adverse drug reactions in children are an important public health problem. We have undertaken a systematic review of observational studies in children in three settings: causing admission to hospital, occurring during hospital stay and occurring in the community. We were particularly interested in understanding how ADRs might be better detected, assessed and avoided. Methods and Findings We searched nineteen electronic databases using a comprehensive search strategy. In total, 102 studies were included. The primary outcome was any clinical event described as an adverse drug reaction to one or more drugs. Additional information relating to the ADR was collected: associated drug classification; clinical presentation; associated risk factors; methods used for assessing causality, severity, and avoidability. Seventy one percent (72/102) of studies assessed causality, and thirty four percent (34/102) performed a severity assessment. Only nineteen studies (19%) assessed avoidability. Incidence rates for ADRs causing hospital admission ranged from 0.4% to 10.3% of all children (pooled estimate of 2.9% (2.6%, 3.1%)) and from 0.6% to 16.8% of all children exposed to a drug during hospital stay. Anti-infectives and anti-epileptics were the most frequently reported therapeutic class associated with ADRs in children admitted to hospital (17 studies; 12 studies respectively) and children in hospital (24 studies; 14 studies respectively), while anti-infectives and non-steroidal anti-inflammatory drugs (NSAIDs) were frequently reported as associated with ADRs in outpatient children (13 studies; 6 studies respectively). Fourteen studies reported rates ranging from 7%–98% of ADRs being either definitely/possibly avoidable. Conclusions There is extensive literature which investigates ADRs in children. Although these studies provide estimates of incidence in different settings and some indication of the therapeutic classes most frequently associated with ADRs, further work is needed to address how such ADRs may be prevented.


Introduction
Adverse drug reactions (ADR) are a major health problem to the individual as well as for society [1]. The World Health Organisation's definition of an ADR is ''a response to a drug which is noxious, and unintended, and which occurs at doses normally used in man for prophylaxis, diagnosis or therapy of disease, or for the modification of physiological function'' [2]. The frequent occurrence' of ADRs in children has been reported in three previous systematic reviews of observational studies covering the period from 1966 to 2010 [3,4,5]. The reviews provided estimates of ADR rates causing hospital admission, in hospitalised children and in outpatient children and demonstrated that ADRs in hospitalised children are a considerable problem. Two of the reviews [4,5] provide data on the clinical presentation of the ADR and the drugs involved. In addition, the more recent review [5] provides information on the methods and persons involved in identifying ADRs.
There are however, a number of limitations to the previous reviews. Each review [3,4,5] applied a search strategy, using a limited number of keywords to just two electronic bibliographic databases -MEDLINE and EMBASE. Importantly, as a consequence, relevant studies may have been excluded. In addition, the reviews excluded studies that included adults as well as children, thus reducing the number of eligible studies, and the more recent reviews excluded studies that evaluated adverse drug events (medication errors as well as ADRs).
These reviews do not provide information about the drugs involved in ADRs or about which methods were used for detecting, or assessing the causality and subsequent of an ADR [6]. Establishing the relationship between the drug and suspected reaction is fundamental to drug safety and being able to determine the avoidability [7] of an ADR in order to try to prevent its future occurrence is crucial to reducing the burden of ADRs.
We therefore undertook this systematic review to provide a more comprehensive assessment of all relevant studies and to understanding how ADRs might be better detected, assessed and avoided.

Criteria for considering studies for this review
Included studies. Observational studies that estimate the incidence of ADRs including retrospective and prospective cohort studies of children.
Excluded studies. Studies which focus on ADRs in relation to a specific drug (e.g. antibiotics or carbamazepine), clinical condition (e.g. epilepsy, asthma) or specific clinical presentations of ADRs (anaphylaxis); case control studies; those carried out exclusively on a neonatal intensive care unit; studies reporting medication errors, therapeutic failures, non-compliance, accidental and intentional poisoning and drug abuse.
Participants. Children as defined by the original study authors.
Studies included three defined populations: 1) children admitted to hospital, 2) children in hospital and 3) children within the community.
Interventions. Exposure to any systemic or topical medicinal product including herbals and aromatherapy, as defined by researchers.
Types of outcome measure. Any clinical event described as an adverse drug reaction or non-avoidable adverse drug event to an individual or group of drugs.

Search methods for identification of studies
A range of electronic bibliographic databases were searched (Table 1) using a search strategy of text words and indexing terms ( Table 2). In addition, we examined references in relevant studies and those cited by previous systematic reviews. Contact with experts was made to identify other potentially relevant published and unpublished studies. We did not apply language restrictions to the search.

Selection of studies
Screening on title, abstract and full publication stage. Duplicate citations were removed. A study eligibility screening proforma based on pre-specified inclusion criteria was used. Two reviewers (RMDS, EG) independently screened each title and categorised as include, exclude or unsure. The two independent categorisations for all titles were compared and the title categorised again following discussion if two reviewers disagreed. Where there was agreement to exclude, the citation was excluded at this stage. All other citations were reviewed at abstract level. This process was repeated and where there was disagreement, discussion took place between reviewers and citations were re-categorised. Those with agreement to include or as unsure were reviewed at full publication level. The process was repeated at full publication stage. Studies considered as unsure or included at full publication stage were reviewed by a third reviewer (JJK). Reasons for exclusion were documented at the abstract and full paper stage of the screening process.
Checking for correct exclusion at each stage. At title stage, two reviewers (RMDS, EG) independently viewed the abstracts for a proportion (2%) of studies excluded. Independent categorisation were compared (as above). This process was repeated at abstract stage where a third reviewer (JJK) reviewed 10% of full papers for studies excluded based on abstract. This was repeated at full publication stage where the same reviewer (JJK) reviewed 20% of excluded full papers. If any studies were excluded incorrectly at any stage, additional checking was performed.

Data extraction
We extracted the following data from each study: 1) Study characteristics: country; year completed; duration; number of sites; design (prospective or retrospective); clinical setting; number of children. 2) Identification of ADR: definition of ADR, including definition of drug exposure; incidence definition and calculation (numerator and denominator, either at patient or episode level); assessment of causal relationship to drug; person who assessed and categorised ADRs; any method (e.g. case record review) or reporting system used (e.g. Yellow Card). 3) Information relating to the ADR: clinical presentation; associated drug(s)/drug classification; associated risk factors (including age, gender, polypharmacy); ADR considered avoidable.

Assessment of methodological quality of included studies
As we were unable to find a validated assessment tool for critically appraising observational studies of adverse drug reactions, we developed a quality assessment form specifically for the review. The following aspects were deemed important when assessing study quality: study design; methods for identifying ADRs; methods used to establish the causal relationship between drug and effect; tools for assessing avoidability of the ADR; and tools for assessing severity of the ADR. Criteria were graded as yes, no, unclear, or not reported. Two reviewers (RMDS, EG) independently assessed methodological quality of each study (Table 3).

Statistical analysis and data synthesis
For each of the three defined populations; children admitted to hospital, children in hospital and children within the community, a forest plot was produced to present the ADR incidence rate and 95% confidence interval for each relevant study. Studies were sub grouped according to whether the incidence rate was reported at the patient and/or episode level and whether or not all patients had been exposed to a drug. Further, for rates reported at the patient level, a distinction was made between studies that had included one admission per patient and those that had included multiple admissions per patient. All results provided per study were included. Pooled estimates were calculated if the variability in incidence rates was not considered too large.
Univariate meta-regression was used to determine if study level characteristics (setting, gender, age, oncology and number of drugs used) are associated with ADR incidence. Incidence rates for ADRs causing admission and occurring in hospital, calculated at the patient level for a single episode were included. Multivariate meta-regression was not undertaken due to the paucity of covariate data. Risk factor analyses reported by any study were collated.

Results
The search was originally undertaken in November 2009 and retrieved 20 906 potentially relevant citations. An update search was subsequently performed in October 2010 and retrieved an additional 3234 citations. Combining both searches we identified 24 140 potentially relevant citations, of which 5 039 duplicate citations were removed. Screening at title and abstract stage excluded a further 18 592 and 251 citations respectively. Full papers were reviewed and 96 citations met the inclusion criteria. Agreement between reviewers at each stage of the review is described in Figure S1. Additional citations were identified through checking for correct exclusion at each stage (n = 3), reference checking (n = 13) and personal communication with authors (n = 5). In total, 117 citations relating to 102 studies were included in the review ( Figure S1).

Included studies
A total of 102 studies (117 citations), were included in the review. Eighty (80/102) studies described the clinical event as an ADR. In 10 of these studies, ADR was a category within 'drug related' problems/admissions; three studies described ADRs as drug induced disease/illness. Sixteen described an ADE where the non-preventable ADE was the same as our definition and two studies used the term iatrogenic disease to describe an ADR. Some studies included multiple settings; 42 studies investigated ADRs as the cause of admission to hospital, 51 studies investigated ADRs in the hospital setting, and 36 studies investigated ADRs in the community setting. Studies included in our review were conducted in 31 different countries, mostly Europe (40/102) and America (32/102 Table 4.

Assessment of methodological quality of included studies
All studies, including those that evaluated ADEs, explicitly stated that they had used either the WHO ADR definition [8] or a comparable one and that they excluded drug errors. Methodological features of each individual study are provided in Table 4.

Study design
The majority of studies were carried out prospectively (n = 85; 83%), which included 13 in those causing admission, 26 studies with the ADR occurring in hospital, 24 in the community, 16 in hospital and causing admission and 6 in mixed hospital and community settings. Fourteen studies were carried out retrospectively, which included six causing hospital admission, two in hospital studies, and four in the community, one causing admission and in the hospital setting and one the study that considered ADRs that resulted in any medical care contact. Two studies (one in hospital, and one in hospital and causing admission), used both study designs. For the remaining study we were unable to determine the study design (Table 4).

Persons involved in identifying ADRs
Sixty-four studies reported that a clinician; either a medical doctor, nurse or pharmacist, was involved in the identification of ADRs. Thirty studies reported also involving either the child or parent. Eight studies did not provide information about who identified the ADRs.

Methods for identifying ADRs
Several methods were used to detect ADRs. Multiple ADR detection methods were employed in 58/102 studies; these consisted of a combination of case record review, drug chart review, laboratory data, computerised ADR reporting system,

Studies estimating the proportion of paediatric hospital admissions related to ADRs
Description of studies. There were 42 studies, where ADRs have been investigated as the cause of admission to hospital. The period under study varied widely and ranged from 1 week to 11 years. The majority of studies were described as being performed in a general paediatric unit or ward (n = 22) 34]. Four studies included general medicine [30][31][32][33] one study in a hospital emergency department [35]. Two studies covered general medicine and a hospital emergency department, [36,37], and one study an integrated primary care information database [38]. Two studies were performed in the paediatric intensive care setting [39], one in combination with general paediatrics also [40]. Seven studies covered a combination of clinical settings [41][42][43][44][45][46][47]. The remaining three studies were performed in dermatology and venereology [48], Infectious diseases [49] and an isolation ward [50].
ADR incidence. We do not have ADR incidence rates for 12/42 of these studies as the child only data was not available (n = 4), data were not split by clinical setting (n = 5), data provided for ADRs in hospital but not causing admission (n = 2) and data were provided for the total number of ADRs but not the ADR 1 st Concept -general terms used to describe the participants -infants and children. 1. exp Child/ 2. exp Adolescent/ 3. (young adj (person$ or people or adult$ or individual$ or women or woman or men or man)).ti,ab. 4. (child$ or adolescen$ or kid or kids or youth$ or youngster$ or minor or minors or teen$ or juvenile$ or student$ or pupil$ or boy$ or girl$).ti,ab. 5. exp Students/ 6. Puberty/ 7. Pediatrics/ 8. (infan$ or newborn$ or new born$ or baby$ or babies or child$ or schoolchild$ or kid or kids or toddler$ or adoles$ or teen$ or boy$ or girl$ or minor$ or juvenil$ or youth$ or kindergar$ or nurser$ or puber$ or prepuber$ or pre puber$ or pubescen$ or prepubescen$ or pre pubescen$ or pediatric$ or paediatric$ or schoolage$).ti,ab.
2 nd Concept including terms relating to adverse drug reactions 9. side effect$.ti,ab. 10. (drug induced or drug related or drug safety).ti,ab. 11. tolerability.ti,ab. 12. toxicity.ti,ab. 13. Harm$.ti,ab. 14. adrs.ti,ab. 15. (adverse adj2 (effect or effects or reaction or reactions or event or events or outcome or outcomes)).ti,ab. 16. (toxic adj3 (effect$ or reaction$ or event$ or outcome$)).ti,ab. 17. exp product surveillance, postmarketing/ or exp adverse drug reaction reporting systems/ or exp drug toxicity/ or exp abnormalities, drug induced/ or exp drug hypersensitivity/ 3 rd Concept -terms relating to the occurrence of ADRs 18. incidence/ or prevalence/ 19. (incidence$ or prevalence$ or occurrence or admission$ or admitted or visit$ or hospitalisation or hospitalised or hospitalization or hospitalized).ti,ab. The terms within each concept were ORed, and then all 5 concepts were combined using the AND Boolean operator. This search strategy was translated as appropriate for the other databases. doi:10.1371/journal.pone.0024061.t002 frequency at the patient or episode level (n = 1). Figure 1 presents data from all studies that provide incidence rates for ADRs causing admission to hospital (n = 30). These rates range from 0.4% to 10.3% of children (single admission). One study was an extreme outlier [20] and if this was excluded we found a reduction in the upper limit of this range to 4%, and a pooled incidence estimate of 2.9% (2.6%, 3.1%).
ADR incidence. We do not have ADR incidence rates for 18/ 54 of these studies as the child only data was not available (n = 3), the data were not split by clinical setting (n = 7), data were provided for the total number of ADRs but not the ADR frequency at the patient or episode level (n = 5), data provided for ADRs and ADEs combined (n = 2), and data provided for ADRs causing admission but not in hospital (n = 1). Figure 2 presents data from all studies that provide incidence rates for ADRs in hospital (n = 36). These estimates range from 0.6% to 16.8% of patients (at a single episode and with prior drug exposure). A pooled estimate has not been calculated since the rates are considered too varied.
ADR incidence. We do not have ADR incidence rates for 19 (19/36) of these studies as the child only data were not available (n = 10), the data were not split by clinical setting (n = 3), data not available for the total number of children/visits (n = 4), data were provided for the total number of ADRs but not the ADR frequency at the patient or visit level (n = 1) and data were provided for errors only (n = 1). Figure 3 presents data from studies that provide incidence rates for ADRs in the community (n = 15). Two studies were not included in this figure due to their method of ADR ascertainment,

All Settings
Drugs and clinical presentation associated with ADR. We do not have information on the drugs involved in ADRs for 50/102 studies, as the child only data were not available (37 studies), ADRs were a subset of events looked at and ADR specific data were not reported (10 studies), and drug data were not available in the publication (3 studies). For studies that provided data (52/101) ( Table 5); anti-infectives were the drug class most commonly reported across the three settings. Proportions ranged from 3.5%-66.6% for causing admission studies (17 studies); 8.6%-100% for in hospital studies (24 studies); and 17%-78% for community studies (13 studies). The most common associated clinical presentations reported were nausea, vomiting, diarrhoea and skin rash. Anti-epileptics were the second most common reported drug class in both the causing admission and in hospital studies; proportions ranging from 0.8%-30% (12 studies); and 3.9%-46.6% (14 studies) respectively. Reported clinical presentations were ataxia, skin rash, increased fitting, and drowsiness. Non-steroidal anti-inflammatory drugs (NSAIDs) were      In addition, corticosteroids were commonly reported across the three settings. Proportions ranging from 5.5%-41.0% for causing admission studies (7 studies); 1.7%-23.4% for in hospital studies (10 studies); and 0.05%-5% for community studies (3 studies). The most common associated clinical presentations reported were immunosuppression, post-operative bleeding, gastric irritation, and diarrhoea.
The distribution of drugs implicated in ADRs reflect the prescribing practices for the individual settings. For example; vaccines were commonly reported in causing admission studies (7 studies) and community studies (5 studies). Proportions ranged from 1.7%-20.0% and 9.2%-25% respectively, with rash and fever being the most common associated clinical presentations. Cytotoxics were reported in both causing admission (8 studies) and in hospital studies (7 studies), proportions ranged from 14.2%-50%, and 1.7%-66.6% respectively. The remaining studies reported a variety of drugs implicated in ADRs, for some more than one drug was the cause of a single ADR (Table 5).

Meta-regression
Univariate meta-regression results (Table 6) suggest that the incidence rate for ADRs occurring in hospital is higher than for ADRs causing admission (OR = 2.73 (0. 93, 8.03)). In addition, the results suggest that the incidence rate is higher for studies with a relatively high mean/median number of drugs per patient (OR = 1.49 (1.14, 1.94)), high percentage of females (OR = 1.13 (0.91, 1.40)), high percentage of oncology patients (OR = 1.15 (0.89, 1.50)) and low mean age of patients (OR = 0.71 (0.39, 1.27)). However, only the variable representing the mean/median number of drugs per patient achieves statistical significance.

Risk factors
Risk factor analyses reported by all studies were collated. Consistent with the meta-regression results, evidence is provided, from 10/19 studies that consider gender as a risk factor, that boys are less likely to have an ADR, and, from 16/17 studies, that risk increases with the number of drugs taken. In addition, 3/3 studies suggest that the risk of ADRs is greater with off-label use. Only two studies considered oncology as a risk factor. The results for the age analyses do not follow a clear pattern and are difficult to interpret due to the variety of age categorisations used.

Tools for assessing causality
Nearly three quarters of the studies (72/102) mentioned a causality assessment, of which the Naranjo algorithm was the most frequently used tool (30/72). Of the 72 studies, seven used a selfassessment method rather than a published causality tool. Despite the majority of studies mentioning a causality assessment, only half of these studies (36/72) reported causality data that were complete for all identified ADRs, specific to the paediatric population and did not include errors as part of the assessment (Table 4).

Tools for assessing severity
Thirty-four (34/102) studies performed an ADR severity assessment. Rates ranged from 0%-66.7% of reported ADRs considered to be severe. By setting, the proportion of ADRs occurring in hospital assessed as severe ranged from 0% to 66.7%, compared with 0% to 45.5% of ADRs causing admission, and 0% to 32.6% of ADRs occurring in the community. Twenty studies provided a reference to indicate the severity tools used, however tools differed widely. Examples of ADRs assessed as severe were those that caused death or were directly life-threatening, caused hospital admission, prolonged hospitalisation or caused transfer to higher level of clinical care.

Assessment of avoidability
Nineteen (19/101) studies performed an avoidability assessment, however, data were only available for 14/19 studies as child only data were not available in 4/19 and ADR specific data were not provided in 1/19. For these 14 studies 7%-98% of ADRs were designated either definitely/possibly avoidable. Three studies provided the rationale for sixty-two avoidable ADRS; inappropriate selection or indication for use of drug (n = 14), inadequate patient education (n = 14), prescribing not rationale (n = 11), lack of appropriate prophylaxis for known ADR (n = 9), lack of appropriate monitoring of drugs (n = 5), previous known ADR to medication (n = 3), dose prescribed was too high (n = 3), inappropriate duration of treatment (n = 1), drug was not prescribed per treatment protocol (n = 1), inappropriate duration of drug and monitoring of treatment (n = 1). Ten studies used a recognised avoidability assessment; of which half used Schumock and Thornton [111] (Table 4).  Table 4. Cont.

Combined settings (in hospital & in community)
Adverse Drug Reactions in Children-A Review Discussion This is the largest systematic review of ADRs in children to date and shows clearly that ADRs are an important clinical problem for children and have been the subject of a large number of studies.
Unlike previous systematic reviews [3,4,5], our review searched for studies using a comprehensive search strategy of a large number of databases, including those specific to toxicology and pharmacology. Nineteen databases were searched of which eight retrieved eligible studies. When compared with the previous reviews this resulted in an additional 73 studies being included in our review, of which, in 24, we were able to extract data. We included studies where ADEs had been evaluated, and that included both adults and children. In addition, we contacted authors of studies to obtain unpublished information. As a result, we were able to obtain unreported ADR incidence data for an additional 24/102 studies. This allowed us to make a more informed judgement regarding ADR incidence estimates.
In agreement with previous studies, including those specific to adults [112], this review found that ADR incidence rates were generally higher in hospitalised children than ADR rates causing hospital admission or in an outpatient setting. One of the main difficulties when comparing ADR incidence rates, particularly from observational studies, is that the studies differ in a number of ways, such as clinical setting, population characteristics and study duration. This may explain the large variation in the incidence rates reported. However, since the numerators and denominators used to calculate ADR incidence were not consistent across studies it was not possible to apply statistical methods to comprehensively explore the heterogeneity. Due to the large amount of heterogeneity, a pooled estimate of the incidence rate has been provided for ADRs causing admission only.
Concerning risk factors associated with ADRs, we found evidence, from both univariate meta-regression and the collation of risk factor analyses from individual studies, that the use of multiple drugs is an important predictor of ADRs. This may be due to the additive risk of an ADR when receiving several drugs or to drug-drug interactions.
We report where possible the drugs associated with ADRs and the clinical presentation, although information regarding drugs involved was poorly reported. The types of drugs associated with ADRs differed substantially between studies due to differences between patient populations there were a number of similarities, and many of the drugs analysed in this review are commonly used in children. The results of this review will facilitate a greater understanding of prescribing practices, thus ultimately reduce drug harm. This may help in the development of interventions to improve drug prescribing and monitoring.
We examined the methods used for detecting, and assessing the causality, severity and avoidability of an ADR. The assessment of     causality in individual cases of ADRs is required to establish whether there is an association between the untoward clinical event and the suspected drug [6]. The detection of ADRs depends on the validity and reliability of the tests employed and if sensitive methods are performed, in theory, all ADRs should be detected. We found a third (31/102) of studies did not report which causality assessment they used, with an additional six not using a recognised algorithm. As a consequence there may be either an underestimation or over estimation of ADRs in these studies. Over a third of studies (34/102) assessed ADRs for the severity of the reactions; just eight of which did not report any severe ADRs. Severe ADRs were described as those that caused either death or were directly life-threatening, caused hospital admission, prolonged hospitalisation or caused transfer to higher level of clinical care [113]. The ability to classify ADRs by severity provides a mechanism for clinicians to identify problem areas and implement interventions to inform paediatric pharmacovigilance practice. The absence of avoidability data was most noticeable in this review; with only fourteen studies (14/102; 14%) providing avoidability data. Therefore it is not possible to consider this important aspect of drug safety in order to prevent future ADRs [7]. Further studies are clearly required to determine which ADRs are potentially avoidable. These studies could provide the necessary data in order to enable clinicians to administer medications in the safest and appropriate way.
The reporting quality of some of the included studies was poor, which may have affected the results. Not all provided a clear definition of the term 'adverse drug reaction'; often insufficient information was in the publication in order to determine whether ADRs included medication or prescribing errors. ADR incidence data were not always clearly described in the publications. In many studies (n = 48/102) reporting was unclear regarding whether the incidence rate was reported at the patient and/or episode level and whether or not all children had been exposed to a drug.
It is disappointing given the large number of studies we identified which addressed this problem that most did not include these important methodological aspects. We recommend researchers should consider the approach which we have taken to assess the quality of these studies, although we recognise that further work is needed to develop a quality assessment tool which meets rigorous standards of development. We recommend that future studies provide information on the avoidability of ADRs; this may help in the development of interventions to improve drug prescribing and  monitoring. There are several outcomes that warrant further investigation or require more detailed information to be collected. Important risk factor data and the number of medications each child received needs to be reported fully in order to explore possible sources of heterogeneity between studies. Future studies need to use clear, unambiguous terminology to describe how ADR incidence rates are calculated. This would improve understanding of the clinical relevance of individual study findings and allow comparisons between studies for the purposes of systematic review, enabling more robust conclusions and recommendations. This review confirms previous studies which have shown ADRs to be an significant problem in children and has highlighted therapeutic classes of drugs most commonly associated with them. We strongly recommend further work to address prescribing practices in different settings and avoidability of ADRs is needed to indicate how such ADRs may be prevented.