Ethics Statement

We received approval from the West of Scotland research ethics committee to undertake this work. Permission to analyse the anonymised data for the year 2008-9 was provided by the NHS Greater Glasgow and Clyde Enhanced Services data group, which was the authorised “guardian” of this data set.

The work involved analysis of a large routinely collected dataset. The data received was completely anonymised and the research team did not have access to patient identifiers.

In view of this, we did not seek individual patient consent to undertake this work. This was in agreement with the guidance given by the local ethics committee and the guardians of the dataset.

Data Collection

The data reported in this paper came from the West of Scotland, with a population of circa 1.8 million served by two different health boards. The local health boards oversee a programme of incentivised depression screening in chronic disease as part of a wider chronic disease management programme of ‘Local Enhanced Services’ (LES). The Quality & Outcomes Framework (QOF) is part of the UK wide, pay for performance, General Medical Services contract for family practitioners [20]. The QOF is an incentive scheme containing a group of clinical indicators, where primary care practices are paid based on how well they score for each indicator, which is dependent on their clinical performance against any given indicator. LES are contractual arrangements at a local health board level with family practices designed to augment the basic QOF specification by incentivising additional indicators that are deemed to be particularly important by a given area and there are no penalties for non-adherence. In the areas under investigation in our study, family practices were paid under the LES scheme to carry out a comprehensive annual health assessment for patients with three common chronic diseases, CHD, diabetes and stroke, including depression screening. Besides depression screening, the annual health assessment also included assessment and management of other health related behaviours such as smoking status, alcohol consumption, diet and activity levels. The annual health assessment was usually carried out by a practice nurse and lasted for one hour. The remuneration offered varied according to disease area with £31 each for patients with diabetes, £26 each for patients with previous stroke and £23 each for patients with CHD. The remuneration was dependent on the level of coverage of indicators achieved in the health assessment, with full payment offered for >90% coverage, 3/4 payment for 75-90% coverage and half payment for 60-75% coverage.

In 2008-9, the period on which the current study is based, the comprehensive annual health assessment was offered to all patients on the practice register with one of the three aforementioned chronic diseases and was usually carried out by practice nurses. The annual health assessment protocol was based on a universal template issued by the health board and it was specific for each of the three diseases (see appendix S1). The results of the assessment were entered into the template with “Read codes” assigned to each data entry. Read codes are the coded thesaurus of clinical terms, by which clinicians in the UK record patient findings and procedures [21]. The assessment included detailed history taking, various physical examinations and blood tests, and recording of certain drugs prescribed including antidepressants, anti-psychotics and cardiovascular drugs.

Importantly, the health assessment included screening patients for depression using the depressive subscale of the Hospital Anxiety and Depression Score (HADS-D) [22]. Patients, who were noted to be under treatment for depression, were exempt from depression screening. The template recommended that patients with HADS of 11 or more should be considered for cognitive behavioural therapy or pharmacotherapy, either by the practice nurse doing the assessment or via subsequent review by the family practitioner, whichever was deemed appropriate by the practice. Patients with HADS of 8-10 were to be advised about low intensity psychosocial interventions such as self-help services or useful online resources [23,24].

Data Processing and Analysis

The data consisted of Read codes, which identified specific measures, the value of each measure taken and the date recorded. Read code identification was based on a list provided by the health board (see appendix S1). Read code entries were converted into clinical values for analysis and independently validated by a second researcher. The validation process continued until both researchers were in agreement.

We restricted our analysis to adults aged from 18 to 90 and health assessments recorded between 01/04/2008 to 31/03/2009. We analyzed the prescription records of those patients, who did not have depression screening results recorded, for an antidepressant prescription in the 12 month observation period. We were not able to explicitly differentiate which of the patients in this category were newly prescribed antidepressants. However based on our situational knowledge of primary care practice in Scotland, that the average prescription duration is not usually longer than 90 days, we further classified these patients into those ‘likely’ to be newly started on antidepressants. If patients in this category were first prescribed antidepressants more than 90 days after the start of the observation period, they were labelled as ‘likely’ to be newly started on antidepressants without undergoing depression screening. Patients were labelled as ‘under treatment’ for depression and exempt from depression screening if: (a) they were noted to be on antidepressants based on their prescription record with no record of depression screening during the 12 month observation period (01/04/2008 to 31/03/2009) (b) the first prescription for antidepressant was issued in the first 3 months of observation period. The full list of antidepressants prescribed can be accessed from appendix S2.

The depressive subscale of HADS (HADS-D), which gives a total score of 21, was used as a screening tool. A score of ≥11 is considered a clinically significant disorder, whereas a score between 8 and 10 suggests a mild disorder [22]. We used the threshold of HADS ≥8 as a cut-off for a ‘positive screening result’ as there is evidence to suggest that this offers an optimal balance of sensitivity and specificity [25-27] and such an approach has been endorsed by national guidelines [28]. All patients who underwent depression screening were checked for a new prescription of antidepressants in the six months after the date of assessment. No reliable information was available on the number of patients who were referred for psychological therapies following their depression screening. We also did not have information on history of previous episodes of depression or history of antidepressants prescribed in the past, prior to the observation period. We considered the effects of screening for each of the diseases individually and in combination. The socioeconomic status was provided to us in the form of the Scottish Index of Multiple Deprivations (SIMD) score, which identifies small area concentration of multiple deprivations across all of Scotland in a consistent way. The SIMD score was divided into quintiles from 1-5 with Q1 representing the most deprived area [29]. Smoking status was divided into current non-smokers and smokers; alcohol status was classified into moderate (< 21 units men, <14 units women), hazardous (21-50 units men, 14-35 units women) and harmful (>50 units men, >35 units women) based on their weekly units consumption. This classification was adapted from the latest report of the Scottish Health Survey [30]. The other variables for health related behaviour such as diet and physical activity levels all had a large number of missing values and hence were excluded from all analyses.

Statistical Analysis

The incidence of a positive screening was calculated for single diagnoses and for both 2 and 3 co-morbidities. The incidence rate was calculated as the number of positive results found as a result of depression screening (defined as HADS ≥8) of the total number of patients screened in each diagnostic or demographic subgroup. Similarly, the rate of new antidepressant treatment was calculated by dividing the number of subjects with a raised HADS who started treatment, by the total number with a raised HADS. Logistic regression models were used to examine whether age, gender, deprivation quintile and multimorbidity were associated with the odds of screened patients having a raised HADS; and on the odds of starting a new antidepressant treatment given a raised HADS following screening. Results of the regression models are presented as the odds ratios associated with either a fixed increase in the value of the predictor (for continuous predictors) or compared to the stated reference group (for categorical predictors), with 95% confidence intervals (CI) and p-values. We visualise the logistic regression analysis by presenting the corresponding predicted probabilities of new treatment depending on a patient’s age, gender and HADS, and adjusted for deprivation and comorbidity, to allow inference for other patients meeting these criteria.

We present a Kaplan-Meier plot to illustrate the time from start of monitoring to starting new treatment dependent on whether patients were screened for depression or not. All subjects were followed up for 6 months and patients who did not start treatment were censored at 6 months. Patients in the screened population were stratified by whether they had a HADS<8, 8 to 10 or ≥11.

Analysis was carried out using the R statistical software, version 15.1 [31]

Section 1. Uptake of Depression Screening in Routine Practice

A total of 125,143 patients were listed as having CHD, diabetes or stroke in the year 2008-09. Figure 1 shows the distribution of patients across the three diseases and their respective proportional combinations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Distribution of patients across three diseases for the year 2008-09.

https://doi.org/10.1371/journal.pone.0074610.g001

Figure 2 shows the distribution of patients according to their depression status. Of the total sample, 10,670 (8.5%) patients were under treatment for depression, having received their first antidepressant prescription of the observation period within the first 3 months and were thus exempt from screening. The remaining 114,473 (91.5% of total sample size) patients were eligible for depression screening. However, depression screening was only undertaken in 35,537 (31.1% of those eligible) and 78,936 (68.9%) were not screened.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Flow chart showing the outcome of depression screening.

https://doi.org/10.1371/journal.pone.0074610.g002

Table 1 provides the details of the demographic characteristics of those with known depression status and those with unknown depression status, along with 95% confidence intervals (CIs) for the differences between them. These show that, due to the extremely large sample size, there were significant differences between the two groups for all characteristics except smoking (for which there is a high level of missing data, likely to be biased since most of the data on non-smokers was missing). However the CIs show that the magnitude of all of the differences was extremely small, suggesting that those screened were similar to those not screened in terms of their demographics.

Section 2. Impact of Depression Screening on Antidepressant Prescribing

Figure 2 shows that, of the screened patients, 7080 (19.9%) were identified as screen positives based on HADS ≥8, of which 572 (8%) (relative risk against those with HADS<8, 3.3 (CI 2.97-3.67), p value <0.0001) were initiated on new antidepressant treatment within the observation period. Only 2925 (8.2% of screened) patients had an indication of moderate to severe depression with HADS ≥11 and of these 303 (10.4%) received a new anti-depressant prescription within 6 months of their assessment. This was the threshold recommended by the health board for considering pharmacotherapy. Finally, 696 (2.4%) of 28,457 patients with a normal HADS (<8) were also started on new antidepressants during the observation period.

Figure 2 also shows that 4989 patients in the non-screened population of 78,936 received their first antidepressant prescription in the last 9 months of observation period; these patients were ‘likely’ to be newly treated based on the assumption outlined in the methods section. Thus, the estimated overall new antidepressant prescription rate for the non-screened population was 6.3%. The overall new antidepressant prescription rate for the screened population was 3.5% (1268 patients out of 35,537). The new antidepressant prescription rate was higher for the non-screened population than for the screened population but it was lower than the rate for those with a raised HADS (≥8), which was 8%.

Figure 3 presents the time to starting new treatments for the screened and non-screened populations. This illustrates the association between a raised HADS (≥8) recorded at screening and a greater proportion of patients receiving treatment more quickly. It is also notable that subjects identified with normal range HADS (<8) receive the lowest rate of new treatments; even lower than the overall rate of the non-screened population.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Antidepressant initiation time for the screened and non-screened population.

https://doi.org/10.1371/journal.pone.0074610.g003

As a sensitivity analysis, we undertook the analyses with and without amitriptyline, since it is often used for other indications, such as diabetic neuropathy. While there were lower proportions of patients recorded as being on antidepressants when amitriptyline was excluded, the distribution across the HADS categories was similar. Appendix S2 reports the full list of antidepressants prescribed with details of number of patients for the non-screened and the screened population.

Section 3. Multimorbidity and Depression Screening

Table 2 compares the uptake of depression screening and the rate of positive depression screening (HADS ≥8) between patients with one of the three chronic diseases under study, those with any combination of two diseases and those with all three chronic diseases. The proportions for uptake and positive screening result increase along with increasing number of chronic conditions. Patients with multimorbidity were significantly more likely to have a positive result on screening (OR 1.48 (95% CI 1.39-1.58) (Table 3).

Section 4. Patient demographics, depression screening and antidepressant initiation

Figure 4 illustrates the clear socioeconomic gradient in the rate of having a positive depression screen (HADS ≥8), with rates being highest amongst those who were most deprived. Logistic regression results shown in Table 3 confirm that patients from the most deprived areas were more than twice as likely to have a positive result (HADS ≥8) as compared to those from the least deprived areas (OR 2.56, 95% CI 2.34-2.79). However, socioeconomic status did not have a significant impact on probability of starting a new antidepressant among those with a positive depression screen.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Title-Depression Screening and Socioeconomic deprivation.

Rate of positive depression screening found from total screened (upper plot); and rate of anti-depressant initiation from those with raised HADS ≥8 (lower plot), versus socioeconomic deprivation status.

https://doi.org/10.1371/journal.pone.0074610.g004

Figure 5 shows that patients from older age groups were less likely to have a raised HADS (≥8) on depression screening as compared to their younger counterparts. The youngest and oldest patients appeared to have lower rates of initiation of antidepressants given a raised HADS (≥8) than those in middle age. Table 3 confirms that the patients aged 45-64 were at least 1.5 times more likely to be treated with antidepressants given a raised HADS than the oldest patients (OR 2.00, 95% CI 1.58-2.54).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Title- Depression screening and Age.

Rate of positive depression screening found from total screened (upper plot); and rate of antidepressants initiation from those with raised HADS ≥8 (lower plot) versus age.

https://doi.org/10.1371/journal.pone.0074610.g005

Table 3 also shows that males were significantly less likely than females to have a positive result on depression screening (OR 0.83, 95%CI 0.79-0.89). Amongst positive depression screens with HADS ≥8, males were also at least 25% less likely than females to be initiated on antidepressants (OR 0.63, CI 0.53-0.75).

An increase in HADS was associated with an increase in the rate of initiation of antidepressants, with each unit increase above 8 in HADS increasing the odds of treatment by 11% (OR 1.11, CI 1.07-1.14). Figure 6 displays the predicted probabilities of a subject starting new treatment given a raised HADS (≥8) against age for each gender and HADS category, showing how much more likely middle-aged and female patients were to start treatment than the oldest, youngest and male patients, for both categories of raised HADS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Relationship between probability of new antidepressant treatment, gender, and age and HADS categories, adjusted for deprivation and comorbidity.

https://doi.org/10.1371/journal.pone.0074610.g006

Summary of Findings and Comparison with Existing Literature

Despite the fact that primary care practitioners were being incentivized to screen for depression in chronic disease, a majority of patients (68.9%) were not documented as having been screened during the observation period. Depression screening was low in our study. While inaccuracies in recording the results on the health assessment template could account for some of the missing results for depression screening in the database, it seems unlikely that this could provide a full explanation of our findings. Potential explanations include lack of family practitioner confidence in the value of HADS-D, it has lower overall accuracy when compared to other depression screening tools [17]; or lack of family practitioner support for this type of screening due to lack of evidence that systematic depression screening leads to improvement in clinical outcomes [15,16,32]. Alternatively, previously reported barriers to discussing depression (or mental health) in patients with chronic disease in primary care, such as stigma associated around the ‘label’ and physicians’ preconception of normalizing depression in patients with chronic disease, could be influencing factors [33,34].

The rate of positive screens identified as a result of depression screening in our study of routine practice ranged from 17-21% for those with a single condition to 26% for those with multimorbidity. This is consistent with rates of 6-22% which have been reported in clinical trials of depression screening [15]. However, epidemiological studies have reported varying results. In a study of routine primary care involving more than 10,000 patients in the United States, a yield of 20.1% positive screens was reported using the PHQ-2 score [35]; while a similar study of 532 patients in Norfolk reported a much smaller yield of 2.3% using the two stem questions (‘During the last month, have you often been bothered by feeling down, depressed or hopeless?’ and ‘During the last month, have you often been bothered by having little interest or pleasure in doing things?’) as a screening tool [36]. Interestingly, Burton et al reported a much lower rate of 3% of new depression cases found from depression screening using the validated two stem questions [37], based on their analysis of a large population study from a different database [38]. However, it is difficult to make a direct comparison of our results with those of Burton et al as their caseness criteria for depression relied on diagnostic coding entered by health professionals, which in turn depends on how health professionals interpret and manage a positive result on depression screening and data on the rate of positive screens is unavailable for that study. HADS-D has a specificity of 81%, when used for depression screening in patients with chronic physical health problems in a primary care setting [17]. It is likely that some of the ‘positive screens’ identified in our study include false positive cases and hence we over-estimate the yield identified from depression screening. On the other hand, there is evidence to suggest that family practitioners often avoid using specific diagnostic codes while recording depression, which underestimates the actual occurrence in clinical practice [39,40].

There is growing evidence internationally, that multimorbidity is the norm rather than the exception, particularly in ageing populations [41-43]. Our study shows a clear relationship between the number of chronic diseases and number of positive screens found on depression screening. Similar associations between multimorbidity and depression screening have been reported from a study based on results from a self reported postal survey in Australian primary care [44]. This further strengthens the argument put forward by policy makers and researchers, both in the UK and US, for the need to consider physical and mental health needs together [45-47].

Importantly, we have shown that higher HADS on depression screening was associated with higher rates of new antidepressant prescriptions in routine primary care practice. This is contrary to previous evidence which has suggested that depression screening does not lead to increases in antidepressant prescribing [13,32]. However, our study reports what happens in routine clinical practice, whereas the existing evidence draws on findings from randomized controlled trials with small sample sizes and studies that usually did not focus on screening for depression comorbid with chronic disease. The majority of those prescribed antidepressants within six months of depression screening had either a negative screening result (HADS <8) or a positive screening with a probability of mild depression (HADS 8-10). There is little evidence of additional benefit of antidepressants compared with placebo in this group of patients [28] and hence this finding has important clinical and resource implications. However, this should be considered in the context of recent evidence of rising antidepressant prescribing in Scotland and the UK, and the uncertainty whether this rise should be attributed to increase in case-identification of depression or increase in duration of depression treatment or both [48-50]. The new antidepressant prescribing rate in the non-screened population was lower than those with raised HADS but higher than those with a normal HADS.

Various studies have shown that females and patients who are from socioeconomically deprived backgrounds are more likely to develop depression [51,52] but this study is one of the first to study their impact on results of depression screening in chronic disease. Our study showed that among the new depression cases, there was no difference in the rate of antidepressant initiation amongst different socioeconomic groups. However, we did find that males were less likely to be prescribed antidepressants, to the extent that a man with HADS ≥11 had a similar chance of being prescribed antidepressants as a woman with HADS of 8-10. This is contrary to the previous evidence which shows that family practitioners were more likely to prescribe antidepressants to male patients presenting with depression [53,54] suggesting there needs to be further research in this area to study the impact of gender on depression treatment. We also showed that middle-aged patients were more likely than the oldest patients to be initiated on antidepressants, given a raised HADS. Previous evidence has shown that primary care professionals often have ‘therapeutic nihilism’ when dealing with depression in elderly patients [55], which could be one of the explanations for this observed difference.

Strengths

This is the first study, to our knowledge, to examine the impact of implementing systematic depression screening for patients with chronic disease in routine clinical practice. The study has a large sample size with good representation of patients from different age groups and socioeconomic backgrounds. Processing of the large and complex dataset was cross validated by two researchers independently for quality control and robustness. A standardized and validated method of depression screening was used.

Limitations

This study used routinely collected data that was not initially intended to be used for research purposes. Since only a minority of the patients were actually screened, depression status was unknown for a large number of patients; also information on health related behaviour was incomplete in the majority of patients. There may be important differences between patients with known depression status and those whose depression status was unknown, which are not evident from their baseline demographic data. Practitioners may intuitively screen those patients where they are more likely to get a positive result, for instance patients with multimorbidity. Also, there is a possibility of reverse causality with GPs reviewing a patient whom they consider to have depression and offering screening subsequently. The overall accuracy of depression screening in our study was reliant on HADS-D which may not be the best available screening tool for patients with chronic disease in a primary care setting [17]. The uptake for depression screening was much lower in our study when compared with uptake for depression screening in the QOF programme in Scotland (90.5%) for the year 2008-09 [56]. However, the QOF programme had target driven incentivisation where practices were only paid if they achieved 90% coverage in depression screening. The incentivisation for the LES programme used in our study instead depended on achieving a proportion of clinical indicators across the health assessment protocol. Secondly the screening tool used for the QOF programme is two stem questions which are much quicker to administer when compared to the HADS-D questionnaire which consists of 7 items. Finally we did not have information on the presence of chronic conditions outside the ones studied, thus multimorbidity count is likely to be underestimated, nor on the number of patients referred for psychological therapies following their depression screening and assessment. We also did not have information on previous episodes of depression or previous treatment with antidepressants prior to the observation period, which are likely to be important influencing factors on antidepressant initiation rate.

Implications for Practice

Implementation of systematic depression screening for patients with chronic disease needs careful consideration. Our study shows that even within the context of incentivisation, universal coverage can be difficult to achieve in routine practice, unless simpler screening tools such as two stem questions are used. Importantly, depression screening identified large numbers of new cases which has clear resource implications, especially since it was clear that depression screening was associated with increases in anti-depressant prescribing and antidepressants were prescribed to patients where there is little evidence of additional benefit above placebo (HADS<8 and HADS 8-10).

Our data suggests that younger individuals, those with multimorbidity, and those from more deprived backgrounds were more at risk of depression in this population, thus suggesting that targeting depression screening at high risk groups could be an alternative approach. Males and elderly patients were less likely to be initiated on antidepressants, which needs further investigation.