A study of the safety and harms of antidepressant drugs for older people: a cohort study using a large primary care database

Self-controlled case-series study design

The self-controlled case-series analyses included only patients who had the outcomes of interest. Cases with each type of adverse event were identified; these were cases with a diagnosis of the adverse event between 1 January 1996 and 31 December 2008. We used only the first recorded diagnosis of the outcome of interest rather than recurrent events. Patients for whom the outcome was recorded as occurring on the same day as their first antidepressant prescription were distinguished in the analysis. Cases without any antidepressant prescriptions were included in the analyses to improve adjustment for age. 57

We used the extracted information on antidepressant prescriptions during the study period for cases with each outcome to identify periods of exposure to antidepressants and a baseline period. We accounted for multiple periods of exposure in the analysis, defining a period of antidepressant treatment as one without gaps of more than 90 days between the end of a prescription and the start of the next prescription. A prescription after more than 90 days counted as a new treatment episode. We categorised the time periods of exposure as 0 days (day of first prescription in each treatment episode); 1–28 days after the first prescription; 29–84 days and 85+ days (remaining treatment period); and periods after stopping treatment (1–28 days, 29–84 days and 85–182 days after stopping). The day of stopping treatment was taken as the date of the last prescription in the treatment episode plus the duration of the prescription. The 28 days before the first prescription in each treatment episode was considered as a separate category, as occurrence of the outcome of interest in this period could affect the probability of receiving an antidepressant prescription. All time periods outside these specified risk periods contributed to the baseline unexposed time periods. These risk periods were selected as they enable examination of short-term and longer-term effects of antidepressants on the risks of adverse events and are similar to those used in other studies of antidepressants. 18,20 Figure 1 illustrates the time periods used.

FIGURE 1.

Risk periods in the self-controlled case-series design.

We used conditional Poisson regression to estimate the relative incidence of each of the outcomes of interest for the defined time periods of risk compared with the baseline period. We adjusted for age in the analyses (65–69, 70–74, 75–79, 80–84, 85+ years). We used a p-value of < 0.01 (two-tailed) to assess statistical significance. Where the outcome was a fatal one we used only time from the first prescription in the observation period for analysis, as otherwise the method is invalid.

Sample size for self-controlled case-series study

To detect a rate ratio of 2.0 in a risk period of 1–28 days after the first prescription for an antidepressant with 80% power and 5% significance, with a proportion of 0.015 (28/1825) in the risk period of 28 days compared with an average observation period of 5 years, 1002 exposed cases would be required for each outcome. 58 We anticipated having at least this number for all-cause mortality and falls. To detect a rate ratio of 3.0 in a risk period of 1–28 days then 231 exposed cases would be required. We anticipated having around this number for rarer outcomes such as GI events (incidence rate 5/1000/year).

Objective

We sought to estimate the costs of health-care resource use in older people who had been diagnosed with depression and compare these for patients who had been prescribed different antidepressant drugs and classes of antidepressant drugs, while controlling for patient characteristics and other factors. Adverse event rates were also estimated enabling us to estimate and compare the cost per adverse event averted for different antidepressants.

Data extraction

The analysis used data from the cohort of patients described above. In addition to data on antidepressant prescriptions, adverse outcomes and patient characteristics, some additional health economic data were extracted. These were the number of practice nurse, community nurse and general practitioner (GP) visits for each patient during the follow-up period from diagnosis with depression up to the earliest of the study end date (31 December 2008), the date of death (if applicable), the date the patient left the study practice (if applicable) or the date of the latest download of data. The use of secondary care is not routinely recorded within the QResearch database, so we were unable to include this information in the analyses.

Analyses

Where technologies have an impact on costs and outcomes over a patient’s lifetime, conducting analysis over a lifetime horizon is appropriate. 59 Within this study, data were available, however, for only up to 13 years, therefore we sought to estimate the costs of antidepressant prescriptions and practice nurse, community nurse and GP visits that would be expected to fall upon the health service within the first 5 years post diagnosis for depression in patients aged ≥ 65 years, and compare these with patients prescribed different antidepressant drugs. In line with the statistical analysis, a 5-year perspective was chosen in the base-case analysis, as the mean length of follow-up was 5.02 years for patients in this study; a large proportion of patients would, thereby, have complete data over this period. We first identified patients for whom at least 5 years of data were available (we therefore excluded those who were diagnosed after 1 January 2004 as the study end date was 31 December 2008). Patients who died before the 5-year follow-up was complete were included in the analysis, but those who moved practices within 5 years (i.e. had incomplete data) were not. Patients who died were included, as resource-use data were available up to the date of death and this constituted the burden upon the NHS over the 5-year period. Patients who moved practices before the 5-year follow-up was complete were not included, as to include only the data up to the point at which they left the GP practice could result in an underestimation of the health-care resources that would be used by such patients over a 5-year period. However, with a view to assess whether those who switched practices had different costs from those who remained with the same practice for the whole 5-year period we compared the 1-year costs (for those patients for whom this was available) for those who were diagnosed on or before 1 January 2004, but switched practices within the 5-year period with the costs for those who were diagnosed within the same period and remained with the same practice for a 5-year period. Costs were estimated in pounds sterling (£) at 2007–8 cost-year levels.

Prescription costs

Unit costs for each drug were estimated using data from the Prescription Cost Analysis database (for the financial year 2007–8),60 which is based on the September 2007 version of the BNF. 52 Within the database, unit costs are estimated at the level of the individual chemical after calculating the weighted average (mean) cost, based on the unit cost and number of prescriptions prescribed for each individual preparation. In order to check that the mean prescription cost for each individual chemical, as calculated within the Prescription Cost Analysis database,60 was applicable to those aged ≥ 65 years, we compared the average dose for each prescription (at the level of the chemical) for patients in our study with that for all prescriptions in the Prescription Cost Analysis database. To do this we extracted the dose for each prescription (including tablets and solutions) for each patient. By taking account of the number of prescriptions at each dose, the weighted average dose was estimated for each type of chemical. The same method was used to estimate the weighted average dose for the Prescription Cost Analysis data. These values were then compared.

To estimate the expected mean cost over a 5-year period post diagnosis for each patient, we assigned the unit cost for each chemical to each prescription, where the unit cost was assumed to be equivalent to the weighted average from the Prescription Cost Analysis database,60 assuming that the mean prescription dosage for each individual chemical, as calculated within the Prescription Cost Analysis database, was applicable to those aged ≥ 65 years, where costs in future years (≥ 1 year post diagnosis) were discounted at 3.5% per annum. 61 Furthermore, to compare costs for patients who were prescribed different antidepressant drugs, we identified patients who were first prescribed each type of antidepressant drug within the first 12 months of diagnosis of depression (of those prescribed an antidepressant within 12 months of diagnosis, 88.3% had a prescription on the date of diagnosis and 93.5% had a prescription within 30 days). Subsequently, for each of the 11 most commonly prescribed antidepressant drugs within our data set, we estimated the mean total prescription cost over the 5-year period post diagnosis. Patients were categorised according to the drug they received first (those initially prescribed more than one antidepressant were excluded from the analyses).

Mean costs were also estimated for the antidepressant classes (TCAs, SSRIs and other antidepressants), based on the weighted average prescription costs for all of the TCA, SSRI and other antidepressant prescriptions. We also identified patients who were not prescribed any antidepressant drugs within the 5-year period. In addition to estimating the mean total prescription cost for each antidepressant drug in question, we estimated the mean total prescription cost for all antidepressant drugs prescribed to each patient in the time period. Again these values were compared for patients who were prescribed the 11 most commonly prescribed antidepressant drugs within our data set and for all TCAs, SSRIs and other antidepressants. Finally, the mean prescription costs (for both the antidepressant drug in question and all antidepressant drugs) associated with each of the 11 antidepressant drugs were ranked from lowest to highest cost.

Costs associated with practice nurse, community nurse and general practitioner visits

We estimated the costs associated with practice nurse, community nurse and GP visits using unit costs extracted from Curtis,62 where all contacts with a practice nurse and GP were assumed to take place at the practice, and all contacts with a community nurse were assumed to be home visits. These visit costs were estimated over the same 5-year period as above, where future costs were discounted and the mean visit cost for each type of health professional was estimated for the 11 most commonly prescribed antidepressant drugs and the antidepressant classes (TCAs, SSRIs, and other antidepressants), as well as for those who received no antidepressant prescriptions within the 5-year period. Additionally, we compared the summation of the total visit cost and the total prescription cost for all antidepressant drugs (referred hereafter as the overall visit plus prescription cost) for the same groups of individuals. The mean overall visit plus prescription costs were ranked from lowest to highest.

Adverse events and their associated costs

Data were extracted on the 13 specific adverse events as detailed above. We used the absolute risks for each of these adverse events for each of the different antidepressant drugs as described previously, along with those for periods of non-use of antidepressants, to estimate the number of additional adverse events per 10,000 patients treated that one would expect to occur in a 5-year period, for each of the different antidepressant drugs compared with no treatment. We used these figures to estimate the mean number of first events that one would estimate each patient to have while receiving a prescription for each of the antidepressants in question. First, we needed to align the time period for the number of events with that for the cost data. This was a period of 5 years, except when a patient died before the end of the time period. Thus, the mean period over which costs were estimated was less than 5 years, if any of the patients prescribed a particular prescription died within the 5-year period. Second, after estimating the mean follow-up time for patients prescribed each type of prescription, we estimated what proportion of the 5-year period this constituted and accordingly estimated the mean expected number of events (per patient) that would be estimated to occur in that period (events in future years were discounted at a rate of 3.5% per annum). These same calculations were performed in order to estimate the number of adverse events (per patient) that would be expected for those prescribed the 11 most commonly prescribed antidepressant drugs, and for the antidepressant classes in the first 5 years post diagnosis. Subsequently, the incremental number of adverse events was estimated by taking the difference between the estimated number of adverse events for patients prescribed each of these antidepressants and the estimated number for patients prescribed no antidepressants. These analyses were preformed for each of the 13 adverse events.

After calculating the mean expected number of events (per person) it was necessary to consider the issue of assigning costs to each of the different types of adverse events. Though we knew whether an event had occurred, the QResearch database does not routinely contain non-primary care resource-use data and, thus, it was difficult to estimate the actual resource use associated with each event. We considered whether we could make an assumption about the mean cost associated with each adverse event, informed by wider literature (where applicable). Though we might, for example, be able to assume the secondary-care costs associated with each fall admission to be equivalent to that for the weighted average of the three fall descriptions contained within the National Schedule of Reference Costs,63 not all falls reported in primary care result in an admission. Alternatively, it could be that secondary-care costs represent an underestimate, as this would exclude additional rehabilitation and community follow-up costs that might occur. Thus, we considered assigning a cost to each of these types of adverse events to be highly speculative. Moreover, to use precise figures might portray an element of robustness that was not appropriate. As a consequence we did not include adverse event costs within our analyses.

Base-case analysis

Of the aforementioned analyses, which included different elements of NHS costs, the mean total prescription cost for all antidepressant drugs was considered to constitute the base case. Visit costs were not included within the base-case analysis as visits would have been made for a number of reasons, many of which might have been unrelated to the particular antidepressant(s) prescribed, the patients’ depression and/or associated comorbidities. That said, within such a large sample it could be argued that non-depression-related costs would be approximately equivalent across different types of antidepressants and that by calculating the incremental cost, between different antidepressants, non-depression-related costs would be excluded.

Before comparing costs for different antidepressant drugs it was necessary to adjust for any patient characteristics and clinical factors that might differ between patients prescribed different antidepressant drugs. Consequently, we used linear regression analyses to estimate the difference between the mean total prescription cost (for all antidepressant drugs) for patients prescribed each of the 11 most commonly prescribed antidepressant drugs, compared with patients not prescribed antidepressant drugs, controlling for the following variables: gender, age at diagnosis, calendar year at diagnosis, depression severity, depression before age 65 years, smoking status, Townsend score, baseline comorbidities (CHD, stroke, diabetes, hypertension, cancer, dementia, epilepsy/seizures, falls, attempted suicide, Parkinson’s disease, hypothyroidism, obsessive–compulsive disorder) and previous use of certain drugs at baseline (statins, NSAIDs, antipsychotics, lithium, aspirin, antihypertensive drugs, anticonvulsant drugs, hypnotic/anxiolytic drugs).

Levels of cost-effectiveness

After adjusting for the aforementioned factors it was possible to estimate the mean incremental total prescription cost (for all antidepressant drugs) for each of the 11 most commonly prescribed antidepressant drugs compared with patients who received no prescriptions. We have described above the calculation of the estimated incremental number of adverse events for each of the 11 most commonly prescribed antidepressant drugs. These cost and event data were used to calculate the incremental cost (mean incremental prescription cost for all antidepressant drugs) per adverse event averted, for different antidepressant drugs, via the efficiency frontier64,65 [the efficiency frontier connects the potentially cost-effective (i.e. non-dominated) options]. The efficiency frontier can be calculated by first identifying the antidepressant drug with the lowest mean prescription cost (for all antidepressant drugs), hereafter referred to as lowest cost. Other antidepressant drugs that are dominated66 by another antidepressant drug (i.e. have a higher mean cost and are estimated to be associated with a greater number of adverse events) can then be excluded, as can antidepressant drugs that are subject to extended dominance66 (i.e. where combinations of other drugs have equivalent or lower mean cost and fewer adverse events). Extended dominance would be apparent if an option was less effective and had a higher incremental cost-effectiveness ratio (ICER) in terms of adverse events than an alternative option. 67 The remaining antidepressant drugs will be located on the efficiency frontier, where one can calculate the incremental cost per averted event (mean incremental cost/expected incremental number of averted events) (ICER) for each antidepressant drug located on the efficiency frontier. Previous studies65,67 provide further details of how ICERs are calculated when evaluating multiple options. These methods were used to calculate the mean incremental cost per averted event (and associated efficiency frontier) for each of the 13 adverse events, comparing the 11 most commonly prescribed antidepressant drugs to each other and the different antidepressant classes (TCAs, SSRIs and other antidepressants).

There is no previously defined threshold (in terms of willingness to pay) against which to compare levels of incremental cost per case averted (in order to assess whether the expected costs would be considered to be worthwhile, i.e. constitute value for money). However, given that the study group patients are aged ≥ 65 years it is unlikely that averting a particular adverse event would, on average, result in a gain of 20 quality-adjusted life-years (QALYs) (after discounting). For example, were one to extend life by 20 years and increase health-related quality of life (HRQoL) by 0.5 years (e.g. assuming HRQoL was initially 0.5, giving a resulting HRQoL of 1.0), then this would equate to a QALY gain of < 20 (after discounting) and it seems unlikely that avoiding an adverse event would (on average) be associated with such a large QALY gain. On that basis, given that NICE has stated that interventions that cost > £30,000 per QALY are unlikely to be deemed cost-effective,59 if the incremental cost per adverse event averted were > £600,000, for a particular option, then, assuming it would not result in a QALY gain of > 20, it would be unlikely to be deemed cost-effective. In the light of this, we assumed that all options which had a cost per adverse event averted of > £600,000 would not form part of the efficiency frontier.

Sensitivity analysis

Sensitivity analysis66 was undertaken in order to estimate the robustness of our results, where the incremental cost of different antidepressant drugs and the incremental cost per adverse event avoided were recalculated using different assumptions. Results were first recalculated using the summation of the total visit cost and the total prescription cost for all antidepressant drugs. One might expect people who have been prescribed different antidepressant drugs to have different consultation rates in primary care, for example consultation rates might be higher for patients prescribed certain antidepressant drugs, as some are more prone to dose changes (this may be more applicable to TCAs than SSRIs) or consultation rates may be higher due to side effects. These results are presented in Appendix 2 and summarised in the main text.

A further sensitivity analysis was to estimate costs and adverse events rates over a 1-year period, enabling data to be used from a greater number of patients, as estimation of 5-year costs resulted in the exclusion of those who were diagnosed post 1 January 2004. Thus, all previously defined costs and adverse events were recalculated for all patients in the study database for a 1-year period (this resulted in the inclusion of all patients, except those who left their practice within 1 year of diagnosis or were initially prescribed more than one antidepressant). The same methods as described previously were used, in which costs and benefits were not discounted as they occurred in the first year of care.

Selection of study cohort

A total of 88,701 patients in the QResearch database were diagnosed with depression at age 65 years or over between 1 January 1996 and 31 December 2007. After consecutively excluding 22 patients aged 100 years and over at diagnosis, 3178 patients with schizophrenia, bipolar disorder or other psychoses, 15,690 who had joined the practice in the previous year and 9065 with a diagnosis of depression or a prescription for an antidepressant in the previous year, there were 60,746 eligible patients remaining who formed the study cohort. Figure 2 shows the selection of patients for the study cohort.

FIGURE 2.

Selection of patients for the study cohort.

The 60,746 patients included in the study were from 570 QResearch practices in the UK. These practices included 543 in England, 14 in Wales, 4 in Scotland and 9 in Northern Ireland. The practices in England were spread throughout the regions, with 32 in the North-East, 61 in the North-West, 59 in Yorkshire and the Humber, 85 in the East Midlands, 45 in the West Midlands, 43 in the East of England, 70 in London, 81 in the South-East and 67 in the South-West. The total number of patients registered with eligible practices during the study period was 9,583,082.

Incidence of diagnosed depression

Table 1 shows the incidence rates of diagnosed depression in people aged 65 year and over, by gender and age group. Rates were higher in women than in men, although the difference was less marked with increasing age.

Study cohort

Baseline characteristics of the study cohort are shown in Table 2. There were 20,230 (33.3%) men and 40,516 (66.7%) women. There were 31,341 patients who were aged 65–74 years at baseline (51.6%), with 7908 (13.0%) aged 85 years and over. Nearly 20% of patients had a diagnosis of CHD at baseline and 38.9% had hypertension. Substantial proportions were taking prescribed medications at baseline, including antihypertensive drugs (50.0%), aspirin (29.4%), hypnotic/anxiolytic drugs (23.7%) and NSAIDs (57.0%).

Patterns of antidepressant treatment

A total of 1,398,359 prescriptions for antidepressants were received during the study follow-up period. The duration of each prescription could be calculated for 1,244,296 (89.0%) of these, based on the quantity prescribed and dosing directions. The median prescription length was 28 days [interquartile range (IQR) 28 days to 30 days] and more than one-half of these prescriptions were for 28 days’ duration (641,811 prescriptions, 51.6%). For 154,063 prescriptions (11.0% of 1,398,359) there was insufficient information on quantity or dosing directions to enable direct calculation of duration, but values were estimated (as described in Chapter 2) based on the quantity prescribed, which was available for 147,165 of these prescriptions (95.5%), and where this was missing a value of 28 days was assumed.

Details of the first antidepressant drug prescribed and the total number of prescriptions received during follow-up are shown in Table 3. Of the 60,746 patients in the cohort 6708 (11.0%) received no prescriptions for an antidepressant during follow-up and the remaining 54,038 (89.0%) received at least one prescription during follow-up. For nearly half (49.0%) of the patients in the cohort the first antidepressant prescribed was an SSRI, whereas for just over one-third (34.6%) it was a TCA.

The 54,038 patents prescribed antidepressant drugs during follow-up received a median of 12 prescriptions, with a range of 1 to 727. A total of 6484 patients (10.7% of 60,746) had only a single prescription and around one-third (20,697, 34.1%) received three prescriptions or fewer during follow-up. The median total duration of treatment with antidepressants during follow-up was 364 days (IQR 91 days to 1029 days), and during the first year of follow-up it was 140 days (IQR 56 days to 308 days).

Table 3 also shows the number of episodes of antidepressant treatment in patients who received at least one prescription for an antidepressant drug during follow-up, where a new treatment episode was defined as one that occurred after a gap of at least 90 days after the end of the previous prescription. Nearly half of the treated patients (47.6%) had only one treatment episode during follow-up, around one-quarter (26.6%) had two treatment episodes and 25.9% had three or more. The median duration of antidepressant treatment per treatment episode was 179 days and the median duration of treatment as a percentage of total follow-up time was 31.3%.

Table 4 shows the total number of prescriptions received during follow-up for each antidepressant class and also for each specific drug, as well as the numbers of patients with one or more prescriptions for each drug. SSRIs were the most commonly prescribed drug class with more than three-quarters of treated patients being prescribed an SSRI during follow-up and 54.7% of the total antidepressant prescriptions were for this class. The most commonly prescribed SSRI drugs were citalopram hydrobromide (23.0% of all prescriptions) and fluoxetine (14.0%). There were 442,192 prescriptions for TCAs, constituting 31.6% of all antidepressant prescriptions. The most commonly prescribed TCAs were amitriptyline (13.5% of all prescriptions) and dosulepin (10.3%). The group of other antidepressants contributed 13.5% of the total prescriptions. The most commonly prescribed drugs within this group were venlafaxine (6.3% of all prescriptions) and mirtazapine (5.9%). MAOI drugs were the least commonly prescribed class, constituting only 0.16% of the total number of prescriptions issued. The 10 most commonly prescribed antidepressant drugs constituted 93.6% of all prescriptions: these were citalopram hydrobromide, fluoxetine hydrochloride, amitriptyline hydrochloride, dosulepin hydrochloride, paroxetine hydrochloride, venlafaxine hydrochloride, sertraline hydrochloride, mirtazapine, lofepramine and escitalopram. As there were only slightly fewer prescriptions for trazodone hydrochloride, the 11 most commonly prescribed drugs were considered separately in some analyses; these constituted 96.0% of all prescriptions.

Figure 3 shows the total number of prescriptions during follow-up for the 11 most commonly prescribed antidepressant drugs issued over the study period.

FIGURE 3.

Total number of prescriptions during follow-up for the 11 commonly prescribed antidepressant drugs.

Figure 4 shows the number of prescriptions for each antidepressant class, by year of prescription. There was a steep increase in the proportion of prescriptions which were for an SSRI over time, with a corresponding reduction for TCAs. There was also an increase for the group of other antidepressants over time. In terms of the first antidepressant prescribed, the proportion of patients for whom the first antidepressant prescribed was a TCA fell from 65.7% in 1996 to 18.7% in 2007, whereas the proportion for whom it was an SSRI increased from 29.9% in 1996 to 75.0% in 2007. The proportion of patients in whom the first antidepressant prescribed was in the group of other antidepressants increased from 4.0% in 1996 to 8.0% in 2004 and then decreased to 5.9% in 2007.

For the 11 most commonly prescribed antidepressant drugs, the proportions of total prescriptions that were for the TCAs amitriptyline, dosulepin and lofepramine all decreased over time, while they increased for the SSRIs citalopram hydrobromide, fluoxetine hydrochloride and escitalopram (data not shown). The proportion of prescriptions that was for the SSRI paroxetine hydrochloride decreased from 13.3% in 1996 to 4.0% in 2008, and stayed fairly constant throughout this period for the SSRI sertraline hydrochloride (at around 6.3%) and the TCA trazodone hydrochloride (at around 2.4%). In the group of other antidepressants the proportion prescribed increased for mirtazapine from 1996 to 2008 and increased for venlafaxine hydrochloride from 1.2% in 1996 to 9.1% in 2004, after which it declined to 5.6% in 2008.

FIGURE 4.

Percentages of prescriptions for each antidepressant class by year of prescription.

Antidepressant treatment by baseline characteristics

The baseline characteristics according to the class of antidepressant first prescribed, excluding the combined group, are shown in Table 5. There were significant differences between the groups (excluding MAOIs owing to small numbers) for all baseline characteristics except for obsessive–compulsive disorder and anticonvulsant treatment, although absolute differences were generally small. The most marked differences were that, compared with the treated groups, the untreated group had a higher proportion of men, a higher proportion of patients aged 85 years and over, higher proportions of patients with CHD, diabetes, dementia and epilepsy/seizures at baseline, lower proportions treated with antipsychotics and hypnotics/anxiolytics and a higher proportion treated with statins. Comparing treated groups directly there was a higher proportion of men in the SSRI group than in the other groups and a lower proportion in the TCA group. There were fewer people aged 85 years and over in the TCA and MAOI groups. Patients in the TCA group tended to be less likely to have comorbidities than patients in the SSRI group. For example, 17.4% of patients in the TCA group had CHD compared with 20.6% in the SSRI group; they were also less likely to be treated with antihypertensive drugs, aspirin or statins than patients in the SSRI group.

Antidepressant dose

Table 6 shows the doses prescribed in terms of DDDs by antidepressant class. Dose could not be calculated for 160,170 (11.5%) of the 1,398,359 prescriptions issued during the study period either because dosing directions were not recorded or were unclear or for certain drugs a DDD value was not available. Prescribed doses tended to be lowest for TCAs, with 70.0% of prescriptions being for ≤ 0.5 DDD, compared with 13.8% for SSRIs. Doses prescribed were highest for MAOIs.

Table 7 summarises doses prescribed for the 11 most commonly prescribed drugs. The median doses were below the DDD values for all of the TCAs, except lofepramine.

Total number and duration of prescriptions for each antidepressant class

Table 8 shows the number of prescriptions received by patients for each antidepressant class and the median total duration of treatment, both for the whole follow-up period and for the first year of treatment. Of the 29,085 patients who had one or more prescriptions for a TCA during follow-up, one-quarter had only one TCA prescription; this proportion was the same among the 10,485 patients with prescriptions for other antidepressants, whereas 20.2% of the 42,575 patients who had one or more prescriptions for an SSRI had only one SSRI prescription during follow-up. Among the patients with one or more TCA prescriptions, the median duration of use during follow-up was 127 days; it was 117 days for MAOIs, 206 days for SSRIs and 172 days for the group of other antidepressants. In the first year of treatment the median duration of use was lowest for TCAs and highest for MAOIs and the group of other antidepressants.

Changes between antidepressant classes

Table 9 shows whether patients had prescriptions from only one class, or changed to another class during follow-up, according to the class of the first antidepressant prescribed. For example, in patients whose first antidepressant was a TCA, then 41.9% had prescriptions only for TCAs during follow-up and 58.1% also had prescriptions from other classes of antidepressants. In patients whose first prescription was for an SSRI, then 67.7% had prescriptions only for SSRIs during follow-up, and 32.3% had prescriptions for other antidepressants. Among those patients who changed from a TCA, the majority of patients changed to an SSRI (10,797, 88.3%). Among patients who changed from an SSRI, 6521 (67.8%) changed to a TCA and 3080 (32.0%) changed to a drug from the group of other antidepressants, and among those patientswho changed from the group of other antidepressants, 541 (33.7%) changed to a TCA and 1062 (66.2%) changed to an SSRI.

These differences may reflect, in part, differing amounts of follow-up between the classes, as TCAs were more likely to be prescribed as a first antidepressant earlier in the study period. An additional analysis was therefore carried out looking at changes within 1 year restricted to patients who had at least 1 year’s follow-up (Table 10). For patients who were prescribed a TCA as their first antidepressant, 20.1% had prescriptions for other classes of antidepressants within 1 year of their first TCA prescription. Among patients whose first prescription was an SSRI, 15.7% had prescriptions for other classes of antidepressants within 1 year of their first prescription. The highest rate of switching occurred in the group receiving other antidepressants, of whom 30.4% had prescriptions for other antidepressant classes during their first year of treatment. Among those patients who changed from a TCA within 1 year, the majority changed to an SSRI (3446, 87.4%). Among those patients who changed from an SSRI, 2717 (66.3%) changed to a TCA and 1371 (33.5%) changed to a drug from the group of other antidepressants. Among those patients who changed from the group of other antidepressants, 280 (34.5%) changed to a TCA and 532 (65.5%) to an SSRI.

Table 11 distinguishes patients who did not switch class, but had only one prescription in their first year of treatment. Among patients who were prescribed a TCA as their first antidepressant, 27.6% had only one prescription in the year and 20.1% had prescriptions for other classes of antidepressants during the year. This compares with 20.8% and 15.7%, respectively, for SSRIs and 19.7% and 30.4%, respectively, for other antidepressants. Among all patients who switched classes within the first year, 52% switched after only one prescription. The proportions who switched classes within the first year of treatment were similar in male and female patients, and by age group (data not shown).

Table 12 shows whether patients switched or only had one prescription within their first year of treatment by individual drug. The proportion of patients who did not switch, but only had one prescription in the first year of treatment was the highest for amitriptyline hydrochloride (31.3%) and the lowest for mirtazapine (16.8%). The proportion of patients who switched from the first drug they were prescribed within a year was the highest for trazodone hydrochloride (34.7%) and lofepramine (32.9%) and the lowest for citalopram hydrobromide (21.7%).

Practice variation in antidepressant prescribing

Table 13 shows variation in practice prescribing by antidepressant class and for the 11 most commonly prescribed drugs across the 570 practices included in the study. The median number of study patients in each practice was 77 (IQR 40 to 131) with a range of 1 to 436. Across practices the median proportion of TCA prescriptions out of all prescriptions for antidepressant drugs was 30.3%, but this ranged from 0% to 100% (IQR 22.6% to 39.4%). The median percentage of SSRI prescriptions was 54.9%, but this also ranged from 0% to 100% (IQR 47.2% to 62.9%). There was considerable variation between practices for all of the individual drugs.

Severity of depression by gender and age band

Table 14 shows the level of severity of the initial diagnosis of depression according to age group and gender. The distribution of the severity of depression was similar in all age bands, and in men and women.

Follow-up details

Table 15 gives details of person-years of follow-up for the 60,746 patients in the study cohort. The total number of person-years of follow-up was 305,188, with a mean per patient of 5.0 years [standard deviation (SD) 3.3 years] and a median of 4.6 years (IQR 2.2 years to 7.4 years).

Table 16 shows the number of patients who had the outcomes of interest during follow-up and the numbers who had these outcomes at baseline. The most common outcome during follow-up was death (29.4% of the cohort), followed by falls (20.2%), fractures (10.1%) and stroke/TIA (9.9%). A total of 43 people committed suicide. Only four patients had antidepressant poisoning recorded during follow-up, so this outcome was excluded from further analysis.

Results of analyses for all-cause mortality

Results of analyses for sudden cardiac death

Results of analyses for suicide

Results of analyses for attempted suicide/self-harm

Results of analyses for myocardial infarction outcome

Results of analyses for stroke/transient ischaemic attack

Results of analyses for falls

Results of analyses for fractures