Changes in rates of recorded depression in English primary care 2003-2013: time trend analyses of effects of the economic recession, and the GP contract quality outcomes framework (QOF)
Tony Kendrick, Beth Stuart, Colin Newell, Adam W A Geraghty, Michael Moore
Abstract
Background: Depression may be increasing, particularly since the economic recession. Introduction of quality outcomes framework (QOF) performance indicators may have altered GP recording of depression.
Methods: Time trend analyses of GP recording of depression before and after the recession (from April 2008), and the QOF (from April 2006), were conducted onanonymised consultation data from142 English practices contributing to the Clinical Practice Research Datalink, April 2003-March 2013.
Results: 293,596patients had computer codes for depressive diagnoses or symptomsin the 10 years.Prevalence of depression codes fell from 44.6 (95% CI 44.2, 45.0) per 1000 person years at risk (PYAR) in 2003/2004 to 38.0 (37.7, 38.3) in 2008/2009, rising to 39.5 (39.2, 39.9) in 2012/2013. Incidence of first-ever depression codes fell from 11.9 (95% CI 11.7, 12.1) per 1000 PYAR in 2003/2004 to 9.5 (9.3, 9.7) in 2008/2009, rising to 10.0 (9.8, 10.2) in 2012/1203. Prevalence increased in men but not women following the recession, associated with increased unemployment. Following introduction of the QOF,GPs used more non-QOF-qualifying symptom or other codes than QOF-qualifying diagnostic codes for new episodes.
Limitations: Clinical data recording is probably incomplete. Participating practices were relatively large and not representative across English regions.
Conclusions: Rates of recorded depression in English general practices were falling prior to the economic recession but increased again subsequently, among men, associated with increasedunemployment. GPs responded to the QOF by switching from diagnostic to symptom codes, removing mostdepressed patients from the denominator for measuring GPperformancein assessing depression.
Key words: depression, prevalence, primary care, economic recession, pay for performance, QOF
Corresponding author:
Professor Tony Kendrick MD FRCGP FRCPsych
Professor of Primary Care
Primary Care & Population Sciences
University of Southampton
Aldermoor Health Centre
Southampton SO16 5ST
Tel: 02380 241083
Fax: 02380 701125
Email:
This is independent research funded by the National Institute for Health Research (NIHR) School for Primary Care Research (grant no. 214). The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
Background
Depression is very common, often chronic or relapsing, and costly, and is usually treated in primary care if it is treated at all. There is concern that depression might be increasing in prevalence worldwide, although the evidence is mixed. Epidemiological surveys suggest prevalenceincreased from the early 1990s up until 2004, at least in the USA (Hasin et al, 2005; Kessler et al, 2005; Eaton et al, 2007). Overall rates in the UK did not appear to have risenat least up until 2007 (Singleton et al, 2003; McManus et al, 2009), although there was limited evidence of an increase among women (Spiers et al, 2012). Major depressive disorder (MDD) moved up from 15th to 11thin the ranking of disorders by disability adjusted life years between 1990 and 2010 (a 37%increase) (Murray et al, 2012), butthis change in rankingwas due to population growth and ageing - prevalence rates for MDD were found to have decreased slightly over the 20 year period (Ferrari et al, 2013). Nevertheless, the King’s Fund estimated that 1.45 million peoplein England would have depression by 2026, and the total cost to the nationwould exceed GBP 12 billion per year, including prescriptions,inpatient and outpatient care, supported accommodation, social services, and lost employment (McCrone et al, 2008).
There are relatively few data onrates of depression in primary care compared to the large amount of epidemiological data from community surveys (Waraich et al, 2004). The diagnosis and treatment of depression in primary care is controversial, as on the one hand it has been found repeatedly that general practitioners (GPs)fail torecognisea proportion of disorders which might benefit from treatment (Kessler et al, 2002; Cepoui et al, 2008),while on the other they are accused of treating more and more people with antidepressants unnecessarily (Parker, 2007). Antidepressant use has been rising steadily since the early 1990s when the selective serotonin reuptake inhibitors were introduced, which could be due to more people being diagnosed year on year, but could be due to the prescribing of longer courses of treatment (Moore et al, 2009), or a combination of both.
Analysis of computerised medical record data from The Health Improvement Network (THIN) general practice database found that the prevalence of adults diagnosed with depression and treated with antidepressants roughly doubled between 1993 and 2004:prevalence among women increased from around 50 to around 100 per 1000 person-years at risk (PYAR), and among men from around 25 to around 50 (Morgan et al, 2008). We analysedcomputerised medical record data from the General Practice Research Database (GPRD) from 1993 to 2005 and also found that antidepressant prescribing approximately doubled during the study period (Moore et al, 2009). However this was due to a doubling in the average number of prescriptions given to each patient rather than an increase in the incidence of new cases of depression. The majority of antidepressant prescriptions were given as long term treatment or as intermittent treatment to patients with presumed multiple episodes of depression, and the recorded incidence of first-ever episodes actually declined over the period: in women from 15.8 per 1000 PYAR in 1993 to 10.1 in 2005, and in men from 7.8 per 1000 PYAR in 1993 to 6.0 in 2005 (Moore et al, 2009).
GPs do not usually use specific diagnostic criteria to characterise depression and they use symptoms as labels associated with antidepressant prescriptions as well as diagnostic labels. Furthermore, their use of labels seemed to be changing in the period up until 2006, perhaps as a consequence of being accused of over-diagnosis. A separate analysis of the THIN database found that the incidence of new episodes of diagnosed depression fell from 22.5 to 14.0 per 1000 PYAR between 1996 and 2006, butthe incidence of new episodes ofdepressive symptoms rose threefold from 5.1 to 15.5 per 1000 PYAR, and the combined annual incidence of diagnoses and symptoms remained stable at around 29 per 1000 PYAR, suggesting that GPs were increasingly using symptom rather than diagnostic labels to record depression (Rait et al, 2009).
These studies predate the economic recession starting inApril 2008, following which there was an increase in suicides in England (Barr et al, 2012) as in other western countries, associated with rising unemployment rates, particularly in men (Chang et al, 2013; Coope et al, 2014). Antidepressant use has continued to rise, doubling in the 10 years up to 2011 among the world’s richest nations (McCarthy, 2013) with a steeper trajectory in England since 2008 (Health & Social Care Information Centre, 2014) which may be due to an increased incidence of depression as well as more long-term prescribing.
The studiesabove also predate the introduction of performance indicators in the GP contract quality outcomes framework (QOF) which required the assessment with symptom questionnaires of the severity of depression in people diagnosed with categorical depressive disorders, at diagnosis from 2006 and also at follow-up from 2009 (BMA & NHS Employers, 2006;2009). This may have affected GPs’ willingness to label people with categorical diagnoses and further encouraged the use of symptoms as medical record labels instead, from 2006 onwards.
We analysed data from the Clinical Practice Research Datalink (CPRD), a longitudinal anonymised research database derived from nearly 700 primary care practices in the UK, formerly known as the GPRD(Medicines and Healthcare Products Regulatory Agency (MHRA), 2014; Williams et al, 2012), to determine how GP rates of recording of depression changed in England between April 2003 and March 2013, exploring possible effects of the recessionfrom April 2008 and the QOF from April 2006, using time trend analyses of quarterly rates.
Methods
The study protocol was approved by the Independent Scientific Advisory Committee of the MHRA. Data were obtained from all general practices which were in the CPRD continuously from 2003 to 2013, whose recording of data was judged by the CPRD to be up to standard (UTS) (MHRA, 2014), and for whom the Index of Multiple Deprivation (IMD) deprivation score (for the practice address) was available, thereby excluding practices outside England. Only general practices continuously in the CPRD and UTS for the whole of the ten year period were included, to provide a relatively stable denominator for calculating rates of recording of depression.The available routinely collected anonymised GPconsultation data includedclinical events (symptoms and diagnoses), therapy events (prescriptions), and referral events (including mental health and other referrals).
Eligibility for inclusion
The inclusion criteria were all patients who between 1st April 2003 and 31st March 2013 had clinical or referral eventsrecorded which included a Read code for non-psychotic depressive symptoms or diagnoses,or for assessment using depression symptom questionnaires. We excluded patients with psychotic diagnoses including bipolar disorder, psychotic depression, and schizoaffective psychosis, and patients prescribed antidepressants for other indications besides depression. The 179 Read codes we used as inclusion criteria were classified by us into four categories: diagnoses (n=88); symptoms (40); questionnaire assessments (36); and others (15). (See Web Appendix for the list of Read codes used, and our classification.)
Anonymised data were obtained from the start of the patient’s registration with the practice, which in most cases predated the 10-year period during which they had to be labelled with a depression Read code to be included in the study. In addition to clinical, therapy, and referral events, the dataset included patients’ dates of birth, gender, marital status, family size, GP practice code (also anonymised), NHS Region, and IMD score for the practice address.
Analysis
We used patient years at risk (PYAR) as the denominator to calculate the level of recording of depression, so someone who was registered in the CPRD for only one quarter of the year would contribute 0.25 years to the denominator. In order to be able to identify significant changes in the annual incidence or prevalence of depression, we calculated a minimum sample size which would be sufficient to estimate a rate of approximately 5 per 1000 PYAR with a 95% confidence interval of plus or minus 0.5 per 1000 PYAR. This required a sample of 19,112 patients per year, i.e. a total of at least 191,120 for the ten year period.
Rates per 1000 PYAR of recorded depression, defined in terms of the presence of the inclusion criteria Read codes, were calculated for each of the 10 years, and for each of the 40 quarters, of our study period, together with 95% confidence intervals. We identified three different numerators:
(i)prevalence of depression:which included all inclusion criteria Read codes recorded in the patients’ clinical or referral events file during the year, or quarter;
(ii)rate of new episodes of depression: a sub-set of the total recorded codes, limited to those patients who had no previous depression codes recorded within the previous four quarters;
(iii)incidence offirst-ever depression: a sub-set of the total recorded codes, limited to those patientswho had no previous codes for depression diagnoses, symptoms or antidepressant treatment recorded within the 10 year study period, and no previous record of depression or antidepressant treatment recorded in their past history prior to April 2003. (We were unable to determine whether patients had had previous unrecorded depression, as the data were anonymised and so we could not ask the patient or their GP about any previous unrecorded episodes prior to their registration with a CPRD practice).
We analysed changes over time in the quarterly prevalence and incidence of first-ever depression codesfor the total cohort, and broken down by gender;by age group (adolescent 16-17 years, younger working age 18-29, older working age 30-64, and retired 65 years and over); and by quintile of IMD deprivation score. We also subdivided new episodes of depression into those which had one or more diagnostic category codes which qualified patients for inclusion in the denominator for the QOF performance indicators, and those which had other depression codes (usually symptom only codes) which did not qualify patients for inclusion in the QOF.
We conducted time trend analyses to determine whether significant changes in the quarterly rates of GP recording of depressioncodes (for both prevalence and incidence of first-ever depression) were found followingthe economic recession fromApril 2008; and whether the rates correlated with subsequent unemployment rates from the Office of National Statistics(2014). We also analysed whether significant changes in the types of coding (QOF-qualifying versus non-QOF-qualifying codes) were found for new episodes of depression following the introduction of QOF indicators for the assessment of depression with severity questionnaires at diagnosis fromApril 2006.Quarterly time periodswere chosen for the time series analyses because yearly periods would have given too few time periods to look at changes either side of the events in 2006 and 2008. We took into account seasonality as well as the underlying trend, as rates of depression are known to vary with the seasons of the year.
Data were analysed as an interrupted time series using segmented regression (Wagner et al, 2002). This divided the time series into two periods, before and after the event of interest, and tested whether there was a significant step change, or change in the slope of the line following the event. Our regression model was: Υt = β0 + β1 x timet + β2 x eventt + β3 x time after eventt + et where: β0 = the level of depression at the start of the observation period in 2003, β1= the secular trend in level of depression, β2= the change in the level of depression in the quarter of the event, β3=the change in trend after the event, and et= the error term, which included an allowance for autocorrelation. We tested for autocorrelation using the Durbin-Watson statistic(Durbin and Watson, 1971) which indicated it was present,so the regression model was fitted using the Stata ‘prais’ command to fit a Cochrane-Orcutt transformation and control for seasonal effects.
Results
The dataset included 142 practices, at which a total of 2,326,673 patients had been registered at some point during the study period April 2003 to March 2013 inclusive. Compared to national practice data (Health & Social Care Information Centre, 2011), participating practices were broadly representative in terms of age and gender profiles and levels of deprivation, but larger than the average for England and not representative regionally, being significantly over-represented in the North-West, and significantly under-represented in the North-East, Yorkshire and Humber, and East Midlands regions.
Prevalence of depression codes, and incidence of first-ever depression codes
A total of 293,596 registered patients (12.6%) had received at least one non-psychotic depression Read code during the 10 year study period. Annual prevalence of depressioncodes fell from 44.6 (95% CI 44.2, 45.0) per 1000 person years at risk (PYAR) in 2003/2004 to 38.0 (37.7, 38.3) in 2008/2009, rising to 39.5 (39.2, 39.9) in 2012/2013. Annual incidence of first-ever depression codes fell from 11.9 (95% CI 11.7, 12.1) per 1000 PYAR in 2003/2004 to 9.5 (9.3, 9.7) in 2008/2009, rising to 10.0 (9.8, 10.2) in 2012/1203. Annual rate of new episodes in patients with no record of depression in the preceding four quartersfell from 29.4(95% CI 29.1, 29.7) per 1000 PYAR in 2004/2005 to 26.8 (26.5, 27.1) per 1000 PYAR in 2008/2009, rising to 29.2 (28.9, 29.5) in 2012/2013.
Figure 1 shows the quarterly prevalence of depression codes, and incidence of first-ever codes, per 1000 person quarters at risk (PQAR), over the 10 year study period (see Web Table 1 for detailed figures). We tested the hypothesis that the economic recession from April 2008 may have changed the trajectory of the prevalence of depression codes, and the incidence of first-ever depression codes, using time series analysis, also shown in Figure 1. Rates of prevalence and incidence showed a similar pattern. The trend prior to quarter 2 of 2008 was a statistically significant decline in both prevalence and incidence rates, but from that date, there was a significant change in the slope of the trend for prevalence (p<0.001), which was not significant for incidence (p= 0.074). There was no evidence of a significant step change butlevels of both prevalence and incidenceflattened out after April 2008 and the slopes were no longer significantly different from zero.
Quarterly prevalence of depression codes by gender
Figure 2 shows quarterly prevalence rates for depression codes per 1000 PQAR, by gender (see Web Table 2 for detailed figures). The interrupted time series models for both genders (Figure 2) showed a similar pattern as that for the total cohort prior to quarter 2 of 2008: levels of both prevalence and incidence of first-ever depression codes were declining. Following that time point, for prevalencein females the slope flattened and was not significantly different from zero. However for males the analysis showed a significant increase in the prevalence of depression codes. There were no significant changes to the slopes of the lines for incidence of first-ever depression codes for either gender (not shown).
Quarterly prevalence of depression codes by age group
Figure 3 showsthe quarterly prevalence of depression codes and incidence of first-ever depression codesby age group (see Web Table 3 for details).Figure 3 also shows the interrupted time series modelsby age groups. The prevalence of depression codes, and incidence of first-ever depression codes, were falling prior to quarter 2 of 2008 in all age groups, except forthe incidence of first-ever depressioncodes in 16-17 year olds. From that point onwards, there was a significant change in trend forthe prevalence of depression codes in all age groups: the trajectory flattened, becoming not significantly different from zero. Changes in the incidence of first-ever depression codes varied between age groups, with 18-29 year olds showing a significant rise after Q2 2008, whilethe trend for 30-64 year olds flattened out, and trends for the 16-17 year olds, and 65 and over, age groupsremained unchanged throughout.
Correlation with the unemployment rate
Figure 4 shows comparisons between levels of unemployment(Office of National Statistics, 2014) and the prevalence of depression codes for males and females respectively. There was anapparent increase in the unemployment rate after quarter 2 of 2008, which appeared to coincide with changes in trajectory for the prevalence of depression codes in both males and females.In the period prior to quarter 2 of 2008, there was a negative correlation between overall prevalence and unemployment (Figure 4). However after quarter 2 of 2008, the correlation was moderately positive and statistically significant. The relationship for males was very similar to the overall cohort, but for females there was no significant correlation after quarter 2 of 2008.