The dataset included 176 practices with 2,569,153 eligible patients registered across the 16-year period (2003–2018), and 17.6 million person-years of follow-up (PYFU). Over the study, there were 264,127 incident anxiety codes (any anxiety code) recorded; 216,126 new episodes of anxiety diagnoses; and 197,217 new episodes of anxiety symptoms.
GP use of anxiety codes
GPs used a large number of READ codes (Table 1). The most frequently used diagnostic codes were ‘anxiety states’, ‘anxiety with depression’, and ‘panic attack’, totalling 82.6% (n = 178,488/216,126) anxiety diagnosis episodes. ICD-10 diagnostic codes were used infrequently, with ‘generalised anxiety disorder’ and ‘mixed anxiety and depressive disorder’ each representing less than 2% (n = 3,482/216,126; n = 3,735/216,126) of diagnostic codes. When diagnostic codes were grouped, codes relating to GAD accounted for more than half of diagnosis codes used by GPs, with a further 31% attributed to the MADD category. When recording anxiety symptoms, ‘anxiousness symptom’, ‘anxiousness’, and ‘worried’, were used in the majority (n = 192,243; 97.5%) of anxiety symptom episodes.
Trends in coding over time
The incidence of any anxiety code rose from 17.8/1000 PYAR in 2003 to 28.5/1000 PYAR in 2018. Between 2003 and 2008, the incidence of anxiety diagnoses fell from 13.2/1000 PYAR to 10. 1/1000 PYAR; after which incidence remained fairly constant, before increasing in recent years (Table 2 and Fig. 1). The incidence of anxiety symptoms more than doubled over the entire study period rising from 6.2/1000 PYAR in 2003 to 14.7/1000 PYAR in 2018 (Table 2 and Fig. 1).
The best-fitting joinpoint model for any anxiety codes included one joinpoint at 2011 [with 95% confidence that the joinpoint was between 2009 and 2014 (95% CI 2009–2014)], after which incidence substantially increased (Supplement 2). For diagnosis codes, the best fitting model included two joinpoints: one in 2008 (95% CI 2006–2011), after which incidence remained fairly constant, and one in 2013 (95% CI 2011–2016), after which incidence substantially increased (Supplement 4). For symptom codes, the best fitting model had one joinpoint at 2007 (95% CI 2005–2009), after which incidence continued to increase, but at a slower rate compared with earlier years (Supplement 5).
After adjusting for age and gender, the IRR for any anxiety code was 1.65 (95% CI 1.63–1.68) when comparing 2018 with 2003 (Supplement 5). For symptom codes, after adjusting for age and gender, incidence more than doubled [IRR 2.41 (95% CI 2.34–2.48)] when comparing 2018 with 2003 (Supplement 7).
Recorded incidence of anxiety in females was nearly twice that of males (Supplement 5–7). This was consistent across any anxiety code, anxiety diagnoses and anxiety symptoms [adjusted IRR: females compared with males: any anxiety code IRR 2.13 (95% CI 2.11–2.14); diagnosis codes IRR 2.07 (95% CI 2.05–2.09); symptom codes IRR 2.12 (95% CI 2.10–2.14)] (Supplement 5–7).
Recorded incidence of anxiety (any code) decreased with age, with incidence for those aged ≥ 85 being just over half (IRR: 0.58 (95% CI: 0.57–0.60)) that of those aged < 25 (Supplement 5). A similar pattern was found for recorded incidence for anxiety diagnoses (Supplement 7), with the incidence for those aged ≥ 85 being approximately half [IRR: 0.48 (95% CI: 0.46–0.50)] that of those aged < 25, and for anxiety symptoms (Supplement 6), with a 30% reduction in the incidence of anxiety for the oldest age group compared with the youngest age group [IRR: 0.67 (95% CI 0.65–0.69)].
Findings from sensitivity analyses examining the potential impact of clustering within GP practices were consistent with the results from models that did not allow for clustering (data not shown). IRRs were the same, although confidence intervals were slightly wider.
Trends in coding over time by gender and age
Whilst the recorded incidence of anxiety was more common in females, the overall pattern of trends over time (any anxiety code, diagnoses and symptoms) were similar for males and females (Supplement 8–10). There was no evidence of an interaction between year and gender for any anxiety code (p value for interaction = 0.38). Whilst there was evidence of an interaction between year and gender for diagnosis codes (p < 0.001) and (weakly for) symptom codes (p = 0.053), the differences were small and unlikely to be clinically meaningful.
When stratified by age, recorded incidence increased substantially in younger age groups in later years of the study (Fig. 2 and Supplement 11 and 12). There was strong evidence of an interaction between year and age for all models (any anxiety code, diagnosis, and symptoms: p value for interaction < 0.001).
There was a marked increase in the recorded incidence of anxiety diagnosis between 2013 and2018 in the two youngest age bands, increasing from 11.8/1000 PYAR to 24.4/1000 PYAR for under 25 s and from 13.1/1000 PYAR to 22.7/1000 PYAR for those aged 25–34. Incidence of anxiety diagnosis fell between 2003 and 2018 in the oldest age groups, decreasing from 10. 5/1000 PYAR to 8.1/1000 PYAR for those aged 75–84, and from 8.4/1000 PYAR to 6.1/1000 PYAR for those aged ≥ 85 (Supplement 11).
There was a marked increase in the recorded incidence of anxiety symptoms over the duration of the study for the two youngest age bands, increasing from 4.6/1000 PYAR to 22.2/1000 PYAR for under 25 s and from 5.7/1000 PYAR to 21.2/1000 PYAR for those aged 25–34. In contrast, whilst the incidence of anxiety symptoms increased over the first half of the study period for the oldest age groups (65–74, 75–84, and ≥ 85 years), incidence then decreased in the second half (Supplement 12).
Trends in coding over time of diagnosis subtypes
Trends over time in the diagnosis subtype groups were also examined (Fig. 3 and Supplement 13). Between 2003 and 2008, the incidence of GAD fell from 7.0 to 5.3/1000 PYAR; increasing over subsequent years to 8.2/1000 PYAR in 2018 (Fig. 3 and Supplement 13). The incidence of MADD gradually decreased from 4.8/1000 PYAR in 2003 to 2.9/1000 PYAR in 2011; and then increased to 6.2/1000 PYAR in 2018 (Fig. 3 and Supplement 13). Between 2003 and 2018, the incidence of Panic gradually declined from 2.4/1000 PYAR to 1.0/1000 PYAR (Fig. 3 and Supplement 13).
Fifteen GPs from six GP practices were interviewed between September 2018 and March 2019. The mean age was 44.9 years (SD 7.7) and eight interviewees were female (53.3%). Those interviewed had been consulting in general practice between 4 and 27 years. The analysis focused on identifying factors that may explain the trends seen in GP coding of anxiety. Possible reasons for these trends are detailed below.
Recent increases in anxiety
GPs commented that the number of patients presenting with anxiety had increased over time, and this had increased GP awareness of anxiety. They suggested that increasing internet use for shopping, working, and interacting meant “people are becoming more isolated…they’re not having to go out…[and] interact with people as much” (GP 15) and lacked ‘real-life’ social support. GPs stated that social media may skew perceptions of what real life is like and made it much easier to make comparisons with others’ lives, and that this could be a factor in the increase in anxiety.
“Social media…this perception that everyone should have this perfect life, perfect looks, perfect body, perfect house, perfect holidays...the reality is that not everyone has…I think that’s what’s feeding an anxiety boom.” GP 1
GPs reported that they had seen a recent increase in patients aged “18–25” (GP 11) presenting with anxiety, notably in the past five years.
“I’ve been a GP for 20 years and the incidence of anxiety seems to be on the increase, especially in the last five years, especially in younger people.” GP 12
They suggested this could be driven by social media, and by “an awful lot more pressure, or perceived pressure…to either perform or to do things” (GP 12), such as “exam…social…work performance” (GP 7). GPs commented that this could be compounded by online gaming. They are living in a “virtual world [that] is not the real world…losing social and physical contact…it makes them anxious about going out and [having] social contact.” (GP 13).
GPs stated that they felt in recent years there had been increased awareness of anxiety in the media and by celebrities, with “greater recognition from the public of their symptoms, less stigma, and [therefore more likely to] seek help” (GP 1). They explained that this meant patients often knew they ‘had’ anxiety and would “specifically raise the question themselves” (GP 4). Therefore, there was potentially an expectation that the GP ‘had’ to “medicalise” (GP 11) their symptoms and give a label of anxiety.
“By the time it gets to us we’re probably over-pathologising it, because we’re seeing it so we’re kind of feeling we have to do something about it…it’s quite difficult just to say that’s normal, don’t worry about it.” GP 2
Coding choice and influences
GPs commented they that used codes such as “anxiety states” (GP 9) to cover a general sense of anxiety, rather than ICD-10 codes. GPs talked about progressing to other diagnostic codes when they had more information during follow-up consultations, such as “generalised anxiety…or…chronic anxiety if they’d had [past] episodes” (GP 10). However, some GPs talked about using codes interchangeably, with a tendency to select whichever anxiety code presented first on the list—“whatever comes up first, that’s a code for anxiety, that’ll do” (GP 2). When talking about anxiety and depression presenting co-morbidly, GPs reported a tendency to code for both conditions “under the umbrella of depression” (GP 11).
“It can be difficult if someone’s depressed and having panic attacks, and I think the majority I do put as depression, but if someone has predominantly anxiety then I will classify them often as depression with anxiety.” GP 12
GPs preferred to use symptoms codes rather than label patients with an anxiety disorder because they thought they could be stigmatising, or because patients “don’t want to be given a diagnosis” (GP 4). Several GPs stated that coding for anxiety was particularly unhelpful if “it makes it easier for them to assume the sick role…that they’re not getting better” (GP 13).
Some GPs mentioned the depression QOF as influencing the decision to code for a symptom rather than a disorder. Although they referred to depression rather than anxiety, there was a sense that the QOF had led to GPs being more cautious about using diagnostic codes across all mental health conditions.
“I think QOF…has skewed prevalence rates…because now if we write depression [rather than low mood] they chastise us if we haven’t done so much within a number of weeks… so I tend to be rather cautious about labels.” GP 6
Threshold for coding symptom versus diagnosis
GPs reported that severity and chronicity of symptoms were used to determine “when to change [the code] to [an] anxiety disorder, where they have chronic…anxiety, like long-standing” (GP 11), rather than using a symptom code. Some GPs suggested they would delay coding for a disorder until “six weeks, a month…if they’re still managing to work then I probably would delay the diagnosis longer” (GP 1).
Some GPs commented that there was an association between coding for an anxiety disorder and prescribing medication to treat it. That is, if a patient had reached a threshold for being prescribed medication, then they would have also reached the threshold for an anxiety diagnosis, rather than an anxiety symptom.
“If I was prescribing an SSRI for anxiety without depression, I would certainly make a formal diagnosis [with a diagnostic code] then.” GP 6