This study included 33,031 children and young people in England who were prescribed an antidepressant in primary care between January 2006 and December 2017. Of these, 12,149 (37%) had a record of visiting a secondary care paediatric or psychiatric specialist less than 12 months before or 6 months after their first primary care antidepressant prescription. The percentage of 5–17-year-olds prescribed antidepressants with a record of visiting a specialist increased between 2006 and 2012, before levelling off at approximately 39%. Considering only those aged 12–17 years old prescribed SSRIs, 24.4% had a record of visiting a child and adolescent psychiatrist. Although the proportion of visiting a child and adolescent psychiatrist initially increased, it began to decrease after 2013.
These results suggest a significant proportion of children and young people receive prescriptions for antidepressants without the involvement of a relevant specialist, such as a mental health specialist or paediatrician. This is contrary to the NICE guidelines for depression [6] and obsessive-compulsive disorder [20], which recommend assessment and diagnosis by a child and adolescent psychiatrist before the initiation of antidepressants, and nocturnal enuresis [30], which recommends referral for further review and assessment of possible underlying factors before initiation of imipramine. There could be several reasons for this, including the lack of awareness of relevant guidelines by GPs. However, one small qualitative study suggested good understanding of the depression guidelines amongst GPs in interviews [31]. Another explanation for GPs initiating antidepressants without the involvement of a specialist could be a lack of prompt access to secondary care services or evidence-based non-pharmacological treatments such as cognitive behavioural therapy. There is limited availability of specialist services leading to long waiting times (average 57 days in 2017/2018) [8] and strict acceptance criteria [32]. According to reports by the Children’s Commissioner for England, provision of low-level (non-specialist, preventative, and early intervention) mental health services is complicated and variable across the country [33]. If a GP feels a child or young person needs urgent treatment for their mental health condition, and other non-pharmacological options and specialist assessment are not immediately available, they may choose to prescribe antidepressants despite the recommendations of clinical guidelines. In addition, patients or their parents may decline other treatment options or request antidepressants. For example, some of the adolescents in one qualitative study described that without SSRIs, they felt too low to benefit from alternative therapies [34].
We also explored the indications associated with antidepressant prescribing to children and young people. Where indications were identified, depression was the most frequently recorded indication overall (38%), echoing the findings of other studies utilising UK EHR [5, 16, 35]. John et al. [16] found over half of new antidepressants prescribed to 6–18-year-olds between 2003 and 2013 were associated with a depressive disorder or depression-related symptoms. Like this earlier study, we used a broad code list to capture depression which included symptom codes (e.g. ‘low mood’) as well as diagnostic codes. The figure we report is therefore likely to be higher than the true proportion of patients with a diagnosis of depression. Furthermore, the depression code list included mixed anxiety and depression codes (not included in the anxiety code list). Depression and anxiety often co-exist, particularly in older children and adolescents.
We were unable to identify an indication for a large proportion of patients in the cohort (46% overall). This was the case for around a third of patients first prescribed SSRIs (31%) and much higher for those first prescribed TCAs (77%). The proportion of patients with a record of one of the indications of interest increased over the study window, particularly for those prescribed SSRIs. For patients prescribed SSRIs in the final year of the study period (2017), 57% had a record of depression and 25% had a record of anxiety. In this study, we used a pre-specified list of indications, and it is possible that some indications were missed. Sarginson et al. explored antidepressant prescriptions to 3–17-year-olds between 2000 and 2015 using UK EHR [5]. They reported a number of indications in addition to those in our study, but all were associated with a small percentage of prescriptions: eating disorder (0.8%), headache disorder (2.3%), and abdominal pain/irritable bowel syndrome (2.3%) [5].
Overall, missing information about antidepressant indications makes it difficult to assess whether antidepressants were prescribed for evidence-based indications (whether licensed or ‘off-label’). Despite this, it may be reasonable to assume the majority of SSRIs were prescribed to treat mental health conditions, particularly towards the end of the study period. As discussed later, several factors may influence clinical coding. The use of codes in UK primary care is generally at the clinician’s discretion, and the lack of a coded indication does not necessarily imply a gap in care provision.
Strengths of this study include the use of QResearch, a large, population-based primary care database which captures all prescriptions issued from participating general practices, and linkage with a secondary care database which captures all NHS hospital visits made by patients in England. The QResearch database includes data from a sample of practices using EMIS Web software, the most widely used and widely distributed clinical computer system in England [36]. As a result, the study cohort is likely to be a representable sample of children and young people in England, and the study findings are likely to be generalisable to primary care across the country. Another strength is that our time window for specialist visits includes the immediate period following as well as before primary care initiation of prescribing. This allows for the possibility that in urgent cases, prescribing may be initiated in primary care together with a secondary care referral. Hence, we are unlikely to have underestimated the true proportion of antidepressant prescribing where there has been no secondary care involvement.
The following limitations should be considered when interpreting the results of this study. First, it is possible that we did not capture all interactions with specialists. While the majority of specialist consultants in paediatrics and CAMHS are likely to be based in secondary care, it is possible some may be based in a community setting which would not be captured within HES. The HES dataset does not include data about privately funded medical care. However, in the UK, the vast majority of children and young people receive healthcare provided by the NHS. This study will not have captured any hospital visits that occurred more than 12 months before or 6 months after the first primary care prescription was issued. Misclassification within the consultant specialty field of the HES dataset could also have resulted in some visits being missed. However, in this study, the association between antidepressant prescriptions and specialist visits was based entirely on their relative timing, and many of the visits (particularly to paediatrics) may have been for unrelated issues. On balance, therefore, we believe these figures are more likely to be overestimates.
Second, we identified indications based on the diagnostic codes recorded by clinicians. Indications would only be captured if coded in the patient’s record and if the code lists we applied captured all relevant cases. UK general practice records contain a mixture of coded and free-text information, and coding can depend on factors such as severity, whether the consultation is the first presentation of symptoms, and whether the condition is included in the Quality and Outcomes Framework (QOF) which provides financial incentives for the management of certain conditions [37]. In this study, it is possible that conditions were recorded in free-text records but never coded, and thus unavailable in the datasets used. Originally, we planned to supplement the primary care data with diagnosis information from HES. Upon examination, however, the secondary care data identified very few additional cases (e.g. 3% increase in the number of patients with a depression diagnosis code), and so only the primary care data were used for this purpose. The code lists used in this study were created by experienced clinicians with expertise in UK primary care and psychiatry. The majority of code lists (except depression, as discussed earlier) included diagnostic codes but not symptom codes. It is possible that the frequency of indications with these narrower definitions was underestimated, depending on the coding preferences of clinicians. In summary, it is possible that relevant information about antidepressant indications is available to clinicians while providing clinical care. However, based on the information typically available at scale to researchers, it is difficult to identify antidepressant indications for a significant proportion of the population of interest.
Third, the prescription data only represent prescriptions issued within primary care. The prescriptions included in the study may not be the patients’ first-ever antidepressant if the first prescription was issued in secondary care. However, when the first prescription occurs in secondary care, prescribing will normally be transferred to primary care, meaning these patients are likely to be represented in our analysis. The full clinical picture will be more complex than presented in this study and conclusions about prescribing behaviour should be limited to primary care.
This paper does not provide any new evidence about the rates of antidepressant prescribing in this population or possible changes to the relative frequency of prescribing of different antidepressants. Several existing studies have demonstrated increases in antidepressant prescribing in the years up to 2015 [5, 16, 35, 38, 39], and in future work, we will explore whether these trends have continued in more recent years (see protocol [17]).