Although adverse drug reactions (ADR) represent a major source of morbidity and mortality in adults, the nature and frequency of ADRs affecting children and young people remains poorly defined. Published data suggest that ADRs account for 1.5-2.1% of paediatric hospital admissions and affect 2.6-9.3% of paediatric inpatients and 1.5-11.1% of paediatric outpatients [1–10].

In the past children and young people have frequently been excluded from clinical trials and where trials are undertaken, they are often insufficiently powered to detect even relatively common ADRs. This means that children and young people are often prescribed medication with limited information about a treatment’s ADR profile, at doses which are not evidence based, and “off–label”; a practice which itself is associated with increased risk of ADRs [2].

Post marketing surveillance is an essential tool in enabling the detection of potentially serious ADRs; however, to detect an ADR with a frequency of 1:10,000, at least 30,000 individuals need to be treated with the drug for at least 1 case to be identified. Therefore, post-marketing surveillance requires the monitoring of large patient cohorts exposed to a medication to detect ADRs. Currently, pharmacovigilance systems rely primarily on spontaneous reporting of ADRs. However under-reporting of ADRs is a significant issue affecting all European ADR monitoring systems. It is widely accepted that only 10% of UK and European healthcare professionals take part in the process of ADR reporting and that only 6-9% of ADRs affecting adults, which should be reported, are actually reported to the regulatory authorities [11–15]. Despite extending spontaneous ADR reporting in the UK to the public, and a wider group of healthcare professionals, public reporting rates still remain low, and significantly lower for children and young people than adults [16].

A growing area of medical concern and drug treatment is that of obesity [17, 18]. Obese adolescents tend to be become obese adults, and it is well recognised that obesity in adults is associated with increased morbidity and mortality [19, 20].

In the UK, orlistat, is currently the only anti-obesity drug licensed for use in severely obese (Body Mass Index, BMI ≥30 kg/m²) adult patients or obese patients with comorbidities. Use in adults is associated with a 30% discontinuation within 3 months due to intolerable ADRs [21–24]. In pediatric/adolescent clinical trials, orlistat use has also been associated with a high level of ADRs (97-100%) and medication discontinuation (3.4-30%). However the majority of these orlistat studies have included less than 30 subjects so the ADR profile in children and young people is poorly understood [25–27].

In the UK, the majority of children and young people who require medical intervention for obesity, even when assessed in secondary care, are routinely reviewed and prescribed their medication in primary care. Primary Care electronic administrative data provide a record of relevant clinical information, service contacts and prescribing information and can, therefore, be used to assess the effectiveness of medications and potentially to generate pharmacovigilance signals. Such datasets which have been previously used for pharmacoepidemiology t have not been widely used for pharmacovigilance and have the potential to offer a complementary approach to routine pharmacovigilance methodologies.

Unplanned medication discontinuation (UMD), as a signal for the identification of possible ADRs, has been previously assessed in the adult population using Primary Care electronic administrative data. This approach is based on the recognition of an early, unplanned discontinuation of the prescription for a medication, possibly accompanied by switching to alternative medication. A close agreement has been previously reported between the UMD rate and ADR rates reported in the literature for a variety of adult medications [28].

The primary objective of this study was to assess the utility of UMD as a signal for possible ADRs in children and young people. Using a large Primary Care electronic administrative dataset and orlistat as an exemplar, we assessed prescribing pattern, medicine use and the potential to generate pharmacovigilance signals for medicines prescribed with low frequency to special populations such as children and young people.

From the study population of 166,726 children and young people (49% female) aged 0–16 years and registered with a participating primary care practice in Jan 2006, a cohort of 79 young people newly prescribed orlistat during the study period were identified. In 2006, at cohort inception, the incidence of orlistat was 0.06/1000/year for children aged 0–16 years.

We demonstrate that using UMD as a surrogate for possible adverse events, routinely collected primary care datasets may be used for pharmacovigilance signal generation, although validation of generated signals and confirmation of potential ADRs will require appropriately designed prospective studies. We believe that this approach will be complementary to the spontaneous reporting systems currently in use and enhance the capability of these systems to widen the breadth of ADRs detected while reducing cost and time to ADR detection. Furthermore such an approach permits assessment of prescribing appropriateness together with real life prescribing patterns. In this study, one in every two patients prescribed orlistat discontinued within a month, and more than three quarters discontinued within three months, of starting treatment.

For medicines such as orlistat, continued use is likely to reflect patient acceptability and perceived effectiveness. In this context it is likely that early discontinuation reflects the appearance of unwanted poorly tolerated ADRs, while later discontinuation represents a perceived lack of response [35]. The levels of UMD in this study are similar to those reported by Viner at al, who noted orlistat discontinuations of 45% at one month, 75% at 3 months and 90% at 6 months [36].

Discontinuation at three months was significantly more frequent in the young people aged 17 years and over, possibly reflecting the effects of emerging autonomy and loss of parental influence [37–39]. One month discontinuation was significantly associated with social class and the presence of other co-prescribed medication, with patients from the most affluent social groups and those co-prescribed other medications less likely to discontinue orlistat therapy. While the association between social deprivation and poor adherence is well recognised the observation that medication discontinuation was less likely in those co-prescribed other medications runs counter to the perceived wisdom that higher levels of non-adherence are observed among adolescents and young adults prescribed multiple medicines [40, 41].

In this study orlistat was prescribed predominantly to adolescents females, and those from the most socially deprived areas. Although the relationship between obesity and social deprivation is well recognised, this is the first study to report a link between social deprivation and orlistat prescribing [42–46].

Seventy percent of the cohort prescribed orlistat was co-prescribed medications including antibiotics, topical treatments for skin conditions and analgesia. Five percent of patients were also prescribed metformin, licensed for use in individuals aged 10 years and older for type II diabetes mellitus, but which is being increasingly used for the treatment of obesity and insulin resistance [46–51].

In the present study, one fifth of all patients (predominantly female) prescribed orlistat were also prescribed antidepressants at some point during the study period. In the general adolescent population antidepressants are reportedly prescribed to 1-5/1000 individuals per year [52–54], therefore the high level of use observed in our study suggests a significantly greater prevalence of depression in the obese t population a finding consistent with studies in both young people and adults [54–56].

Read code, describing clinical symptoms and treatments, were recorded by a primary care physician for approximately a fifth of patients around the time of orlistat discontinuation. Respiratory, genitourinary, endocrine, skin, gynaecological and gastrointestinal symptoms were reported to occur in 10%, 8.9%, 7.6%, 7.6%, 5.1% and 5.1% of patients respectively. Data from pediatric/young person orlistat trials suggest that adverse effects, primarily diarrhoea, affect 97-100% of individuals, leading to a discontinuation rate of 2% - 30% [26, 27, 57]. In this study a Read code for gastrointestinal symptoms (all of which were abdominal pain) was only recorded for 5.1% of patients. The reasons for this apparent difference between our data and that from clinical trials may be that patients were pre-warned about possible gastrointestinal symptoms and therefore did not report the ADR, or that gastrointestinal side effects were so common that primary care physicians did not record them as a reason for discontinuation.

An unexpected finding was the number of individuals for whom urological system Read codes, urinary catheterisation and urinary retention, were recorded. The reasons for this finding are not clear. While urinary retention is frequently associated with neurological disorders and or constipation, assessment of co-prescribed medicines and items for these individuals, identified only one subject prescribed laxatives and none prescribed medicines or items commonly used for the treatment of neurological, bowel, bladder, malignant or infective conditions during the study period. Although orlistat use has been associated with nephrolithiasis and acute renal injury [58, 59], there are currently no data in the literature linking orlistat with urinary retention. However as of November 2013 (current reporting period), the Medicines and Healthcare products Regulatory Agency, the UK government medicine regulatory agency, had received two “Yellow Card” reports for urinary retention possibly associated with orlistat use [60], raising the possibility that the association between orlistat use and urological symptoms observed in this study may be real. Although the approach used in this study doesn’t allow us to attribute causality, it is reassuring to observe that the symptoms and diagnoses recorded by primary care physicians around the time of orlistat discontinuation, such as influenza infection, upper respiratory infection, headache and menstrual irregularities are in close agreement with those reported for adults in the real world [61].

Current clinical guidelines make it clear that prior to the initiation of orlistat therapy, height and weight should be recorded. In our study, 70% of the study cohort had this information recorded, and of these only a third had measurements recorded within a month of their initial prescription. The failure to undertake appropriate BMI assessment prior to orlistat initiation means that it is not possible to determine for the majority of patients whether orlistat was prescribed appropriately, according to the UK guidelines.

We have previously reported that electronic prescribing databases can be used to identify UMD in adult patients, and that such UMD rates can be used as a surrogate for ADR/AE rates [28, 35]. However, the use of UMD does not identify the reasons for discontinuation or the nature of possible ADRs/AEs. The UMD and ADR results we obtained for orlistat are in close agreement with previously published UMD and ADR rates obtained using more traditional methods of ascertainment, indicating that UMD obtained from routinely collected data may represent a surrogate marker for ADR/ADE occurrence. Using this approach together with more traditional spontaneous reporting systems [62], may reveal potential pharmacovigilance signals at an earlier stage following introduction of a medicine and permit a more accurate assessment of ADR incidence than is currently possible.

The use of large primary care electronic datasets enables the linkage of whole population prescription information with recorded clinical symptoms, diagnoses and encounters. As we have demonstrated, this approach permits the assessment of prescribing trends and identification of individuals who discontinue their medication at an early stage before clinical effectiveness might be expected. Assessment of clinical symptoms and diagnosis recorded around the time of orlistat discontinuation identified unexpected findings which might indicate, although not prove, the occurrence of an ADR to the study medication.

There were several limitations to our study. It is possible that the small number of patients prescribed orlistat in this study may reduce the generalisability of the orlistat specific findings; however it does not alter the overall conclusion that the linkage of UMD with medical Read codes identifies possible pharmacovigilance signals of interest. Moreover, the PCCIU-R database, like most other prescribing/dispensing databases, only reflects primary care prescribing and does not provide information on secondary care prescribing or over the counter purchase of orlistat. It is therefore possible that patients for whom an initial prescription was issued in secondary care or who purchased their medication OTC may have been missed. However this is unlikely to be a major issue in the Western world, where primary care remains the major source of medication prescriptions. A further limitation is the lack of a specific recorded reason for orlistat discontinuation in the PCCIU-R database, and this is the reason why all Read codes at or about the time of discontinuation were identified and examined.

Using orlistat as the exemplar, we have demonstrated that when unplanned medication discontinuation are linked to clinical codes for events occurring at the time of discontinuation, it is possible to generate both predictable and unpredicted pharmacovigilance signals. These findings support the use of this approach together with traditional pharmacovigilance methodologies to identify possible signals and plan more specific and focused pharmacovigilance studies with the aim of validating suspected ADRs.