A multicentre programme of clinical and public health research in support of the National Suicide Prevention Strategy for England

Suicide statistics

Good-quality suicide statistics are critical to inform suicide prevention priorities and to assess the impact of prevention strategies. However, whether a self-inflicted death is recorded as a suicide, an open verdict, a misadventure or an accidental death by the Office for National Statistics (ONS) depends on the outcome of a coroner's investigation, and coroners differ in their threshold for labelling deaths as suicide. 9 Such differences and changes in coroners' categorisation of possible suicides over time may compromise an assessment of the incidence of suicide and so undermine the evaluation of specific preventative activities using routine data. 10

Medicines commonly taken in fatal overdoses

Suicide may occur in the context of serious mental illness or result from an impulsive act in a moment of crisis. Whether an episode of self-harm results in death depends on the ease of availability of lethal suicide methods, and some of the best evidence concerning suicide prevention concerns restricting access to lethal methods. 5,11 Drugs associated with high case fatality when taken in overdose are therefore a common target for preventative intervention. Overdoses of paracetamol (acetaminophen), co-proxamol and dosulepin account for 10% of suicides in England, and the Medicines and Healthcare products Regulatory Agency (MHRA) has implemented a number of initiatives to reduce the incidence of suicides using these and other medicines. One of the most recent initiatives has been the phased withdrawal of co-proxamol since 2005. The impact of the withdrawal of co-proxamol requires evaluation to inform both international prescribing policies and future interventions of this sort, such as restricting the use of dosulepin. Furthermore, continued deaths from paracetamol overdose since regulatory action in 199812 suggest the need to review existing sales regulations and evaluate the long-term effects of the legislation.

Assessment and management of people who self-harm

Whether someone who attempts suicide and survives later successfully repeats his or her act may depend on the NHS's response to his or her initial attempt. 13 Yet there is uncertainty concerning the most appropriate service structures to assess and manage self-harm patients. This uncertainty was demonstrated in our previous studies showing wide variations between hospitals in their management of self-harm. 14,15 It is unclear whether or not these variations have any implications for patient outcomes; although person-based studies suggest that some aspects of service provision may affect outcome,16–18 these findings may be confounded both because patient characteristics may influence the services they receive and because high-risk patients may be most likely to self-discharge without receiving care. It is crucial to assess whether or not differences in hospital management influence outcomes in this key area.

Trials of interventions to reduce self-harm in high-risk clinical populations

Half of all people who die by suicide have been in contact with health services within a month of their death. 19 The two highest-risk groups are patients who have recently been discharged from psychiatric inpatient care and those who have self-harmed. One-quarter of all people who die by suicide have been under the recent care of psychiatric services,20 and around half have been in contact with services following a previous episode of self-harm. 6 Thus, interventions to reduce suicide risk in these groups are a priority; yet, Cochrane reviews21 and National Institute for Health and Care Excellence (NICE) guidance13 have highlighted the lack of good-quality evidence concerning the management of self-harm. Trials of contact by telephone or letter, or the provision of ‘crisis cards’ facilitating access to urgent care following self-harm have had mixed results. 22 Likewise, only one randomised controlled trial of an intervention to reduce the risk of suicide following psychiatric inpatient discharge has been carried out,23,24 and, although this trial showed promising results, it has not been replicated.

Objectives and research questions

Our research programme consisted of four interlinked studies, each of which related to key aspects of the 2002 National Suicide Prevention Strategy for England. 3 Our objectives were to:

1. conduct a review of the inquest records of 12 coroners, spanning a period of 15 years, to quantify the extent to which suicide rates, secular trends in suicide, and suicides from co-proxamol, paracetamol and tricyclic antidepressant (TCA) overdose may be over- or underestimated when they are based on deaths given suicide and open verdicts by coroners

2. carry out a series of pharmacoepidemiological studies using national mortality data together with information from self-harm registers in Oxford and Manchester and coroners' records to investigate:

   1. (a) associations between blood paracetamol levels and liver damage (note: because of challenges obtaining relevant data, this was subsequently modified to a comparison of the impact of differing parcetamol regulations in the UK and Ireland)

   2. (b) whether or not the current sales limit for paracetamol is appropriate and being adhered to, determined by interviewing a sample of people who have taken a paracetamol overdose about their episode

   3. (c) whether or not any compensatory increased use of non-steroidal anti-inflammatory drugs (NSAIDs) has led to increased rates of gastrointestinal (GI) haemorrhage

   4. (d) the impact on suicide of the recent withdrawal of co-proxamol

   5. (e) whether or not regulatory action is warranted concerning dosulepin (a TCA)

3. assess whether or not between-hospital variations in four key aspects of self-harm services (psychosocial assessment, general hospital admission, psychiatric admission and follow-up) influence patient outcome, as indexed by repeat self-harm

4. develop, pilot and evaluate contact-based interventions aimed at reducing:

   1. (a) repeat self-harm

   2. (b) self-harm following specialist inpatient psychiatric treatment.

Examination of coroners’ inquest records

To investigate the impact on trends in national suicide rates of possible changes in coroners' use of different verdicts for likely suicides, we examined the inquest records of a sample of 12 of the 107 coroners' jurisdictions in England in 2005. Our sample comprised the three coroners whose jurisdictions covered the three collaborating research centres (the cities of Bristol, Oxford and Manchester) and a random sample of nine further jurisdictions within 90 minutes' travel time of each of these three centres. Four of the study jurisidictions were urban, three were rural and five were mixed urban/rural. We have previously conducted collaborative studies with these 12 coroners. Ministry of Justice data38 indicate that the relative proportions of suicide, accident/misadventure, open and other (mainly narrative) verdicts given by our sample of 12 coroners in 2005 were broadly similar to those in England as a whole in that year (participating vs. non-participating jurisdictions: suicide 14.3% vs. 19.3%; accident/misadventure 62.7% vs. 53.5%; open 13.5% vs. 15.1%; other 9.6% vs. 12.1%).

Analysis of Ministry of Justice data on narrative verdicts

To investigate the impact on suicide rates of the growth in use of narrative verdicts, we obtained data on the number and types of coroners' verdicts for the 113 jurisdictions in England and Wales for 2008–9 from annual statistics published by the Ministry of Justice. 40 We excluded data for the Isles of Scilly and the Queen's Household because of the very small numbers of deaths in these jurisdictions.

To investigate geographical variation in the use of narrative verdicts we used a proxy measure: the number of verdicts classified as ‘other’ verdicts by the Ministry of Justice. ONS data suggest that approximately 80% of ‘other’ verdicts in 2008 and 2009 were narrative verdicts. 29 We were unable to obtain data on the specific numbers of narrative verdicts by coroner jurisdiction as the Ministry of Justice receives data on the breakdown of ‘other’ verdicts from only around three-quarters of coroners and these data are of varying quality, making the estimation of narrative verdicts at this level difficult (Mark Edwardes, Ministry of Justice, 1 July 2011, personal communication).

To investigate the possible impact of the use of narrative verdicts on apparent trends in suicide we compared changes in suicide rates between 2001–2 and 2008–9 in the English local authorities (LAs) served by the 10 coroners who gave the highest proportion of ‘other’ verdicts and the 10 LAs whose coroners gave the lowest proportion of such verdicts. We combined suicide data for these two 2-year periods to reduce the impact of small-area variations in suicide rates based on small numbers of events. We obtained the suicide data [ICD-10 codes X60–X84 (suicide – ‘Intentional self harm’) and Y10–Y34 (open verdicts – ‘Event of undetermined intent’)] for the LA served by each coroner from the National Compendium of Clinical and Health Indicators. 41 As this source includes data for England only, this part of our analysis was restricted to English LAs. Because many coroners serve more than one LA, we present rates for all LAs within relevant coroners' jurisdictions and pooled LA rates for the top and bottom 10 coroners' jurisdictions.

Examination of coroners’ records

Search of the coroners' records for 1990–1, 1998 and 2005 led to the retrieval of 1296 coroner-defined suicides and 790 cases of possible suicide. Amongst the 790 cases of possible suicide, 518 (65.6%) had been assigned an open verdict, 240 (30.4%) a verdict of accident or misadventure and 32 (4.1%) a narrative verdict.

The 790 possible suicides were reviewed independently by DG, KH and NK. They agreed on the inclusion/exclusion of 632 (80.0%) of the cases without the need for discussion. The remaining 158 (20.0%) cases were discussed face to face or by teleconference. Following this review, more than three-quarters of the possible suicides assigned an open (415/518, 80.1%) or narrative verdict (25/32, 78.1%) by the coroner were rated as suicide, as were about half (131/240, 54.6%) of those with an accident/misadventure verdict. Altogether, 571 (72.3%) of 790 possible suicides with a non-suicide verdict were rated as suicide (Table 1).

Temporal change in verdicts assigned by coroners to researcher-defined suicide cases

In keeping with national suicide statistics,42 the number of suicide cases that we identified from our sample of 12 coroners rose between 1990 and 1998 before falling between 1998 and 2005. In 1990–1, 72.0% [95% confidence interval (CI) 68.3% to 75.5%] of researcher-defined suicide cases in our sample had been assigned a verdict of suicide by the coroner. By 2005 this had decreased somewhat to 65.4% (95% CI 61.4% to 69.2%) (Figure 1). There was also a slight decrease in researcher-defined suicide cases assigned an open verdict between 1990–1 and 2005. In contrast, the proportion of researcher-defined suicide cases with a coroner's verdict of accident/misadventure nearly doubled between 1990 and 2005, from 4.6% (95% CI 3.1% to 6.6%) to 9.1% [95% CI 6.9% to 11.7%; p(trend) = 0.02]. Most of this rise was due to an increase in the number of cases of researcher-defined suicide by self-poisoning. In 1990–1, 12.3% (20/163) of cases of researcher-defined suicide by self-poisoning had been assigned an accidental verdict by the coroner; this increased to 22.2% (35/158) in 2005. In logistic regression models controlling for age, sex and method of suicide, statistical evidence for the trend towards increasing use of accidental verdicts for probable suicides increased [p(trend) = 0.001]. This indicates that neither changes in the age/sex distribution of suicides nor changes in the methods used account for the observed increase in use of accidental verdicts for probable suicides.

FIGURE 1.

Number (%) of researcher-defined suicide cases (all methods) assigned each verdict for deaths in 1990–1, 1998 and 2005. a, No narrative verdicts were recorded in 1990–1 or 1998. Source: Gunnell D, Bennewith O, Simkin S, Cooper J, Klineberg E, Rodway C, et al. Time trends in coroners' use of different verdicts for possible suicides and their impact on officially reported incidence of suicide in England: 1990–2005 [published online ahead of print 1 November 2012]. Psychol Med 2012. http://dx.doi.org/10.1017/S0033291712002401. Reprinted with the permission of Cambridge University Press.

Analysis of Ministry of Justice data on verdicts given by coroners and local authority suicide data

In 2008 and 2009, coroners delivered 58,777 inquest verdicts: 28,996 in 2008 and 29,781 in 2009. The mean annual number of verdicts given per coroner was 262 (range 17–979). Of these, the most common verdict (figures for 2009) was accident or misadventure (n = 8673, 29%), closely followed by natural causes (n = 8281, 28%); there were 3797 (13%) ‘other’ (mainly narrative) verdicts, 3330 (11%) suicide verdicts and 2240 (8%) open verdicts.

There is considerable variation between jurisdictions in the proportion of inquest outcomes recorded as ‘other’ (mainly narrative) verdicts: these ranged from 0% to 50% (median 9%) of all inquests. Table 6 shows the number (%) of verdicts given by the coroners in the 10 jurisdictions with the highest proportionate use of ‘other’ verdicts in 2008 and 2009 and by the coroners in the 10 jurisdictions with the lowest proportionate use of ‘other’ verdicts in 2008 and 2009. Jurisdictions ranked in the top 10 gave ‘other’ verdicts in > 20% of inquests whereas jurisdictions in the bottom 10 gave ‘other’ verdicts in < 2% of cases. In Birmingham and Solihull an ‘other’ verdict was recorded in 985 (50.3%) of the 1958 inquests. The proportions of suicide verdicts were higher in the bottom 10 jurisdictions, with a mean proportion of 15.2% compared with 9.3% in the top 10 jurisdictions. Of note, the jurisdictions with the greatest proportion of ‘other’ verdicts seem to also deal with a higher number of inquests overall, with an average of 669 inquests in 2008 and 2009 compared with an average of 450 across England and Wales. Conversely, jurisdictions with a lower proportion of ‘other’ verdicts dealt with a lower average number of inquests over the same period (mean 218 per year).

Analysis based on all 113 jurisdictions confirmed that the use of ‘other’ (mainly narrative) verdicts was inversely related to the recording of suicide verdicts (r = − 0.41, p < 0.001), although there was only a weak association with the proportion of open verdicts (r = − 0.16, p = 0.09) (Table 7). There was no association between the proportion of natural death verdicts given by coroners and their use of ‘other’ verdict categories (r = 0.01), although the proportion of accidental death verdicts was also inversely associated with ‘other’ verdicts (r = − 0.50, p < 0.001). Surprisingly, there was no association between the proportions of suicide verdicts and open verdicts given by coroners (r = 0.02, p = 0.8).

The weighted average change in the rate of suicide between 2001–2 and 2008–9 was − 16% (95% CI − 27% to 5%) in the 30 LA areas served by the 10 English coroners' jurisdictions giving the highest proportions of ‘other’ (mainly narrative) verdicts (Table 8); in the 30 LA areas served by the 10 English coroners' jurisdictions giving the lowest proportions of ‘other’ verdicts the rate did not change (0%, 95% CI − 15% to 6%) (Table 9). There was weak statistical evidence that these two changes in rates differed: difference in rates 1.51 (95% CI − 0.14 to 3.36) per 100,000 (p = 0.09). These findings were based on suicide data for England and so we were unable to include the four Welsh LAs listed in Table 6 in this analysis.

Limitations

In our assessment of the likelihood of suicide among individuals given open, narrative or accidental verdicts by the 12 coroners in our study, there was an initial lack of consensus across the three coders for about one-fifth of the cases examined, illustrating the difficulties involved in deciding whether some cases were suicide or not. Although the coroners' jurisdictions studied comprised 10% of all jurisdictions in England and Wales, the number of jurisdictions and the variability in their size meant that a difference in practice (verdicts assigned) by a single coroner with a large jurisdiction could bias the assessment of a temporal increase in the use of the accident/misadventure verdict. Furthermore, because of resource limitations, we studied only three time periods. An analysis of data for additional years would have increased confidence in our assessment of time trends.

The main limitation of our analysis of Ministry of Justice data is that we used ‘other’ verdicts as a proxy for narrative verdicts. Nevertheless, recent ONS analysis29 indicates that this is a reasonable assumption, with over three-quarters of ‘other’ verdicts in 2008–9 being narratives. It remains possible that there are regional variations in the proportion of narratives amongst ‘other’ verdicts. The Ministry of Justice does not receive consistent nor reliable data from individual coroners on their ‘other’ verdicts, which prevents the accurate reporting of narrative verdict use in each jurisdiction.

Conclusions

National suicide statistics are crucial to public health surveillance and this research has a number of policy and public health implications. First, small area (primary care trust/LA)-specific suicide rates and trends in suicide rates should be treated with caution in those areas where local coroners make high use of narrative verdicts. Second, approaches to ensure the future reliability of national suicide statistics should be taken – these might include asking coroners to both record the short-form verdict and, when appropriate, accompany this with a longer narrative account of the death. LA medical examiners and the post of Chief Coroner may lead to improvements in reporting practices; what is clear is that approaches are needed to ensure consistency in reporting cause of death between coroners. Third, the ONS might consider including in its suicide statistics deaths from medicine poisoning given verdicts of accident/misadventure by coroners. In particular, the assessment of the overall burden of suicide from co-proxamol, paracetamol and TCA poisoning is best achieved by combining deaths from these medicines receiving suicide, open and accidental verdicts.

Data sources

Statistical analyses

Analyses were conducted using Stata version 10.0. We used interrupted time-series analysis80 to estimate changes in levels and trends following the 1998 legislation. We compared the mean quarterly numbers of deaths and liver unit registrations and transplantations that might have occurred in the post-intervention period without the legislation with the number that occurred with the legislation. 81 The end of the third quarter of 1998 was chosen as the point of intervention. For more details of this method see Appendix 3.

In addition to the basic regression model for the analysis of paracetamol-related deaths, we included adjustment for potentially confounding trends in ‘all drug-poisoning suicide deaths’ by inclusion of ‘all drug suicide deaths excluding paracetamol' as a covariate. We also calculated a conservative estimate of the absolute effect, which assumed no increase in the number of deaths in the absence of the legislation. The absolute effect of the legislation was determined as the difference between the outcome expected at the last point of the pre-intervention period and the outcome expected at the mid-point of the post-intervention period. For analysis of liver unit registrations and transplantations we also used the basic regression model and the conservative estimate analysis.

We also conducted a sensitivity analysis to determine whether our results changed when January 1998 was used as the intervention point – this corresponds to the date when the packaging changes began to occur (9 months before the legislation).

Deaths

The numbers of deaths in England and Wales between 1993 and 2009 from poisoning with all drugs and from paracetamol specifically that received suicide, open and accidental verdicts are shown in Table 10. Paracetamol poisoning deaths constituted between approximately 9% and 10% of drug-poisoning suicide deaths before the legislation and between approximately 7% and 9% after the legislation.

Regression analysis of quarterly data indicated a significant decrease corresponding to the September 1998 legislation in both level (i.e. step change) and trend in deaths involving paracetamol in England and Wales that received a suicide or open verdict (Figure 2 and Table 11).

FIGURE 2.

Number of suicide and open verdict deaths involving paracetamol alone in England and Wales, 1993–2009, by quarter years, and best fit regression lines related to the 1998 legislation.

The estimated average decrease in number of deaths was 17 per quarter (95% CI − 25 to − 9 deaths per quarter) in the post-intervention period compared with the expected number based on trends in the pre-intervention period (Table 12). This change equated to an overall decrease in number of deaths of about 43% in the 11.25 years post legislation, or 765 fewer deaths than would have been predicted based on trends during 1993–September 1998. An overall decrease of 36% was found when a conservative method of analysis was used (see Table 12).

There was also a downwards trend in all drug-poisoning (excluding paracetamol) deaths receiving a suicide or open verdict during the post-legislation period, although this was smaller in magnitude (25%) and was not associated with the step change seen for paracetamol following the introduction of the 1998 legislation. When the change in paracetamol deaths was adjusted to take account of the fall in poisoning deaths involving other drugs, the decline in paracetamol deaths changed very little (see Table 12). Similar results were found when accidental poisoning deaths involving paracetamol were included with suicides and open verdicts (see Table 12). The reduction in deaths in the post-legislation period when accidents were included equated to 990 fewer deaths than expected. Although suicides (including open verdicts) involving any method showed a significant downwards trend during the post-legislation period, there was no step change associated with the legislation (see Table 11).

When the intervention point was moved back 9 months to the beginning of 1998 to take account of earlier introduction of packaging changes, there remained a significant downwards trend in suicide and open verdict deaths involving paracetamol during 1998–2009 but no step change (see Appendix 3, Table 45).

There was no major change in number of deaths involving paracetamol compounds following the 1998 legislation (see Tables 11 and 12). These deaths represented a relatively small proportion of overall paracetamol-related deaths (see Table 10).

Liver unit activity

Registration for liver transplantation

There was a decrease in level and trend of the number of registrations for liver transplantation related to paracetamol-induced hepatotoxicity in England and Wales following the 1998 legislation (Figure 3). The mean quarterly change compared with the expected number of registrations based on trends in the pre-intervention period was − 10.7 (95% CI − 20 to − 1), equating to a 61% reduction (mean 6.9 individuals per quarter vs. mean 17.6 in the pre-legislation period) (Table 13). This was equivalent to 482 fewer registrations over the 11.25-year period following the legislation. We obtained a similar result when we used the beginning of 1998 as the legislation point. A conservative estimate of the absolute change in number of registrations for liver transplants was smaller, and was not significant.

FIGURE 3.

Paracetamol-related registrations for liver transplantation waiting list in England and Wales, 1995–2009, by quarter years, and best fit regression lines related to the 1998 legislation.

TABLE 13 - Changes in number of paracetamol-related registrations for liver transplantation and liver transplants in England and Wales, 1995–2009, associated with 1998 legislation to reduce pack size

Using interrupted time-series segmented regression analysis80 in which the intervention point is taken as the third quarter of 1998.

Estimated for the mid-point quarter of October 1998–2009.

Absolute difference between estimated number with legislation and estimated number without legislation, taken at the mid-point of the post-intervention period.

95% CI calculated according to Zhang et al. 82

	Estimation of the absolute effect of the legislation October 1998–2009a			Conservative estimation of the absolute effect of the legislation October 1998–2009a
Liver unit registrations and transplants	Mean quarterly estimated no. without legislationb	Mean quarterly estimated no. with legislationb	Mean quarterly change October 1998–2009c (95% CId)	Mean quarterly estimated no. without legislationb	Mean quarterly estimated no. with legislationb	Mean quarterly change October 1998–2009c (95% CId)
Paracetamol-related
Registrations	17.6	6.9	− 10.7 (− 20 to − 1)	11.1	6.9	− 4.3 (− 14 to 5)
Transplants	8.2	4.8	− 3.4 (− 12 to 6)	6.8	4.8	− 2.0 (− 11 to 7)
All liver transplants	132.3	148.8	16.5 (− 30 to 63)	138.9	148.8	9.9 (− 37 to 57)

Liver transplantation

There was a non-significant reduction in number of liver transplantations for paracetamol-induced hepatotoxicity in England and Wales in the period October 1998–2009 following the legislation compared with the estimated level based on the 1995–September 1998 trend (Figure 4). The estimated mean quarterly change was − 3.4 (95% CI − 12 to 6) transplantations (see Table 13). The number of liver transplantations for all causes in the UK showed a small but non-significant increase during the post-legislation period (from a mean estimated quarterly number of 132.3 before the legislation to 148.8 after).

FIGURE 4.

Paracetamol-related liver transplantations in England and Wales, 1995–2009, by quarter years, and best fit regression lines related to the 1998 legislation.

Non-fatal self-poisoning with paracetamol

There was no decline in the number of hospital presentations in Oxford between 1993 and 2009 for non-fatal overdoses involving paracetamol (pure or compound form) (1993–8 mean = 567.0 per year vs. 1999–2009 mean = 601.6 per year).

Strengths and limitations

One limitation of this study is that we used only data on deaths from poisoning with paracetamol (with or without alcohol) in pure or compound form, not deaths in which paracetamol was consumed with other drugs. This approach, however, ensured that the findings of the study were restricted primarily to paracetamol and not substantially affected by the toxicity of other drugs or compounds. One strength of the study is that it was based on national data for both deaths and liver unit activity.

We have not been able to estimate the number of possible substitutions of paracetamol overdoses with other methods of poisoning or self-harm, but the trend in total suicides (all methods) during the post-1998 legislation period was downwards, as was the trend in suicides by ingestion. This downwards trend in all suicides probably reflects other factors that have favourably influenced suicide rates and, hence, may have contributed to the findings for paracetamol poisoning deaths. A further limitation was that data on non-fatal overdoses of paracetamol were restricted to one hospital, but this included all presentations, not just admissions.

Conclusions

The legislation introduced in 1998 to restrict pack sizes of paracetamol was associated with long-term benefits in terms of reduced numbers of deaths from paracetamol poisoning and registrations for liver transplantation. The number of deaths annually from paracetamol poisoning suggests, however, that further methods of prevention should be sought.

Patient characteristics

The study sample included 35 females [mean age 29.9 years, standard deviation (SD) 13.2 years, range 16–65 years] and 25 males (mean age 33.3 years, SD 14.9 years, range 19–65 years) (Table 14). For 24 patients (40.0%) this was their first overdose, but 10 patients (16.7%) were multiple repeaters, having taken at least five previous overdoses (see Table 14). Over half (n = 32, 53.3%) had taken a previous overdose of paracetamol. The SIS was completed for 59 patients. High (13–20) or very high (21+) scores were recorded for over half (n = 33, 56.0%) of these patients. At the time of the research interview, 43 patients (72.9%) reached case status for anxiety (score > 9) on the HADS and 37 (62.7%) reached case status for depression (score > 9).

Nature of the overdose

Over one-third of the patients had consumed alcohol at the time of the overdose (Table 15). Presentation to hospital was delayed by > 6 hours after taking the tablets in 22 cases (36.7%). The majority (n = 50, 83.3%) were treated with the antidote N-acetylcysteine. Seventeen patients took one or more other drugs in their overdose.

The number of paracetamol tablets taken ranged from 18 to 224. Over three-quarters of the patients took fewer than 50 tablets, but over half (n = 35, 58.3%) took all of the tablets available. Half of the patients had taken the overdose impulsively, within an hour of first thinking about it (see Table 15). One-quarter said that they had thought about the act for < 15 minutes before taking the tablets. However, seven people had considered the overdose for more than a week.

Source of the paracetamol

In many cases (n = 32, 53.3%) the tablets were already in the home, mostly for general pain relief. Three people had boxes of 100 tablets, which had been prescribed after operations or for back pain. Over half (n = 32, 58.3%) bought tablets specifically for the overdose (nine of these also had supplies at home). Most purchases were made from a supermarket or pharmacy (see Table 15). Other outlets included four garages and an online pharmacy site.

Ten people tried to buy > 32 tablets in one transaction; four succeeded. Of these, one was able to buy in bulk from a cash and carry outlet, one purchased 14 packs of 16 tablets from an online pharmacy, and one obtained multiple packs by telling pharmacy staff that he was going away on holiday and needed supplies to take with him. The fourth bought three packs of 16 tablets from a supermarket. Nine patients commented that they were aware of the limit on sales and had therefore bought supplies in more than one transaction.

Anticipated effects of the overdose

When asked what they had thought the physical effects of the overdose would be if untreated, the majority (n = 53, 88.3%) said that they knew that a paracetamol overdose could cause death or permanent damage (see Table 15). Over three-quarters (n = 48, 80.0%) knew that excess paracetamol could harm the liver. However, 42 (70.0%) thought that they would lose consciousness after the overdose. Over half of these (n = 23, 54.8%) said that they would not have taken paracetamol if they had known that they would not lose consciousness and that the effects would not be immediate.

The patients were asked how many tablets they thought would cause death. Twelve (20.0%) did not know. Of the rest, the number chosen ranged from 7 to 200. Five people (8.3%) thought that ≤ 16 tablets could kill, and 17 (28.3%) thought that 17–32 tablets would be lethal. Nine (15.0%) thought that ≥ 100 tablets would be necessary to cause death. Some people's opinion was influenced by the pack sizes: ‘I thought that over 32 tablets would be toxic, as that is the limit you can buy over the counter'; ‘Ten tablets could kill, because it says on the pack not to take over eight’.

Paracetamol packs include a warning about the dangers of overdose. Just under half of the patients (n = 29, 48.3%) said that they had noticed this, but in most of these cases (23/29, 79.3%) this had not affected their decision to take the tablets. Six people reported that the warning had strengthened their decision, for example, ‘The wording [i.e. go straight to doctor if you take more than eight] implies that it would be good to overdose on . . . The warning made it seem that an overdose would be effective.'

Most of those who had not noticed the warning said that they would still have taken the overdose if they had seen it (24/30, 80.0%). Only five people thought that seeing the warning would have deterred them from taking the tablets. Almost all (52/58, 89.7%) said that they would still have taken a paracetamol overdose even if the packs had contained fewer tablets.

Motivation

When asked to indicate which of the motives shown on cards best explained why they had taken their overdose (they could choose as many as they liked), over three-quarters said that they had wanted to die (Table 16). Other common reasons were ‘to get relief from a terrible state of mind’ and ‘to escape from an unbearable situation’.

Influences and reasons for taking paracetamol

Over one-third of the patients (n = 23, 38.3%) knew someone else who had taken an overdose of paracetamol. Twenty-one patients (35.0%) had seen or read about paracetamol overdoses in the media; nine (42.9%) of these thought that this had influenced their decision to take an overdose. For example, two people had read newspaper articles that specified the number of tablets taken in fatal overdoses and then took the same number of tablets in their own overdose. Another person saw a local news item about someone who had taken paracetamol together with alcohol and other medication; she went out the next day to buy paracetamol to take with her own prescribed medication and alcohol. Eight people had used the internet to obtain information on suicide methods. Five of them said that this had influenced their decision to take paracetamol and had provided them with information on toxic amounts.

When asked why they had chosen to take paracetamol in their overdose, over one-third (n = 23, 38.3%) said that the tablets were already in the household. Twelve (20.0%) bought paracetamol because it was cheap and readily available. Sixteen people (26.7%) had chosen paracetamol because they knew that it was effective: ‘. . . catastrophic effect on the liver’.

Although patients were aware of the unpleasant effects of paracetamol, this did not deter them: ‘I heard it was a slow and painful death’; ‘I don't deserve any better than a horrible death’. Five people (8.3%) said that they had taken paracetamol because they had done so in the past: ‘I've always done paracetamol – I know it's a nasty killer, it takes a long time to do it’; ‘I've always done so in the past and always managed to get through it’.

Limitations

The study was based in a single large general hospital and so it is uncertain if the findings are generalisable to patients presenting to other hospitals. However, the characteristics of people presenting to the hospital in Oxford with paracetamol overdoses are similar to those seen elsewhere. 46 We relied on patient report of the number of tablets that they had taken in overdose, and this may be unreliable. 42,46,92

Conclusions

The characteristics of the patients in our study who took paracetamol overdoses were remarkably similar to those in the previous Oxford study64 over 15 years earlier. The impulsive nature of many of the overdoses and the fact that a substantial proportion of patients used household supplies of paracetamol for self-poisoning mean that special attention needs to continue to be paid to the availability of paracetamol, especially in quantities that can cause hepatotoxicity. There appears to be reasonable adherence to the sales guidance. Media influences on use of paracetamol for self-poisoning, including through the internet, are important in some cases.

Data sources

Ethical approval

The monitoring systems in Oxford, Manchester and Derby all have approval from local ethics committees to collect data on self-harm for local and multicentre projects. The National Registry of Deliberate Self Harm in Ireland has ethical approval from the National Research Ethics Committee of the Faculty of Public Health Medicine, and from the relevant hospitals and Health Service Executive ethics committees.

Data analysis

We used SPSS for Windows version 14 to compare overdoses of paracetamol in England and Ireland. Analyses comparing the numbers of tablets consumed, and numbers of packs used, in overdoses of paracetamol in England and Ireland were conducted using the chi-squared test, Mann–Whitney U-test and Kruskal–Wallis test. Analyses were conducted on combined data and separately by sex, age group and alcohol involvement with the overdose (data from the English study on alcohol consumed during the 6 hours before self-poisoning and/or as part of the overdose were combined for this analysis). The number of packs used was calculated as a multiple of the non-pharmacy maximum pack size in each country, from one pack to nine or more packs (e.g. in England an overdose involving up to one pack was 4–16 tablets, two packs was 17–32 tablets, three packs was 33–48 tablets, and so on; in Ireland, up to one pack was 4–12 tablets, two packs was 13–24 tablets, three packs was 25–36 tablets, and so on).

Study samples

During the 6-year study period (2002–7) there were 31,107 hospital presentations for self-poisoning (alone) in the three English centres (six hospitals) and 42,877 in Ireland (up to 40 hospitals: 34 in 2002, 37 in 2003, 38 in 2004–5 and 40 in 2006–7). Of these, paracetamol was involved in 10,208 (32.8%) episodes in the English centres and in 9057 (21.1%) episodes in Ireland. Paracetamol alone (with or without alcohol) was involved in 5444 episodes in the English centres and in 3886 episodes in Ireland.

Of the episodes involving paracetamol alone, data on number of tablets taken in overdose were missing for 559 (10.3%) episodes in the English centres and 358 (9.2%) in Ireland. We also excluded episodes in which fewer than four tablets were taken, data on sex or age were missing or patients were < 10 years old [totalling 27 (0.5%) in the English centres and 19 (0.5%) in Ireland]. Thus, the samples for inclusion in the study consisted of 4858 episodes in the English centres and 3509 in Ireland.

The female to male ratio for episodes was somewhat larger among the Irish sample (2.1  :  1) than among the English sample (1.8  : 1) (Table 17). The age distributions of the two samples showed a greater proportion of younger individuals among the Irish sample and a greater proportion of older people among the English sample. Alcohol use at the time of overdose occurred with similar frequency in the English and the Irish samples.

Number of tablets taken in overdoses

The distribution of the number of tablets taken in paracetamol overdose in the English centres and in Ireland is shown in Figure 6. There were clear peaks in each of the countries corresponding to the pack size limits for non-pharmacy and pharmacy sales. Thus, in the English sample there were peaks at 16 and 32 tablets and in the Irish sample there were peaks at 12 and 24 tablets. There were also peaks at multiples of these pack sizes. There were, in addition, peaks in both samples at 10 and multiples of 10 tablets.

FIGURE 6.

Distribution of the number of tablets taken in paracetamol overdose by those aged ≥ 10 years in England (n = 4858) and Ireland (n = 3509), 2002–7, shown as percentages of the total number of overdoses of paracetamol in each country.

The median number of tablets consumed in paracetamol overdoses did not differ significantly between the English (median 22 tablets) and the Irish (median 24 tablets) samples (Table 18). When the samples were divided into three age groups and the two countries compared, a small but statistically significant difference in the median number of paracetamol tablets consumed between the English (median = 24) and the Irish (median = 23) samples was found among women in the ≥ 35 years age group (see Table 18).

In both England and Ireland, more paracetamol tablets tended to be taken by men but not women when alcohol was involved in overdoses (England: Z = − 3.141, p = 0.002; Ireland: Z = − 2.105, p = 0.035) (Table 19). There was a non-significant trend for smaller overdoses in males in Ireland who had not consumed alcohol than in those in England who had not consumed alcohol.

Number of packs used in overdoses

When the numbers of tablets consumed in overdoses by individuals in each country were categorised into numbers of pack equivalents used, the mean number of packs used was greater in Ireland (2.63, 95% CI 2.57 to 2.69 packs) than in England (2.07, 95% CI 2.03 to 2.10 packs). For both sexes combined, the largest proportion of overdoses in England (39.0%) involved one pack only, but the largest proportion of overdoses in Ireland (37.9%) involved three or more packs (Table 20). Females and the younger age group (15–24 years) in England were more likely to use one pack, but two packs were more likely in Ireland in these groups. Males and older age groups (25–34 years and 35+ years) were far more likely to use three or more packs in Ireland than in England.

Strengths and limitations

The study involved large numbers of patients in both countries: the Irish sample included all overdose presentations to general hospitals in Ireland, and the English sample included presentations to six general hospitals in three centres (although these may not be fully representative of all general hospitals in England). The recording of the number of tablets taken in overdoses relied largely on patient self-report, which is known to be subject to inaccuracy. 42,54,92

It is not known whether or not there was any change in the sizes of overdoses in Ireland following the introduction of the 2001 legislation on pack sizes there. 77 However, the size of overdose for which calls were made to the National Poisons Centre decreased in the first 2 years after the legislation was introduced. 96

We have not been able to examine the impact on actual suicides of the differing legislation in the two countries because data for suicide deaths according to specific drugs ingested are not currently available in Ireland.

Conclusions

No major difference was found between England and Ireland in the size of paracetamol overdose, in spite of differences in maximum pack sizes between the two countries. Although this may suggest that further reductions in pack sizes in the UK would not be effective in further reducing self-poisoning with paracetamol, the finding may reflect differences in adherence to sales guidance in Ireland compared with England, and other factors such as patient characteristics and ease of access to clinical care.

Sources of data

Statistical analysis

Gastrointestinal admission and prescription rates (number of admissions and prescriptions per 100,000 population) for each age group were calculated on a quarterly basis using the population data. The impact of the legislation on numbers of GI admissions and prescriptions and trends in GI admission and prescription rates was analysed using interrupted time-series analysis. This method controls for baseline rate and trend when estimating expected changes in the number of GI admissions (or prescriptions) due to the intervention. The third quarter of 1998 was chosen as the point of intervention. All analysis was carried out using Stata version 11.2.

Prescriptions for non-steroidal anti-inflammatory drugs and other analgesics

There was a significant increase in the rate of prescribing of NSAIDs following the 1998 legislation regarding pack sizes of paracetamol of 31.5 per 100,000 per quarter (p = 0.009). There was also an increase in prescribing of other non-narcotic analgesics (p = 0.05) (Table 21 and Figure 7).

FIGURE 7.

Rates of prescribing of NSAIDs, other non-narcotic analgesics, antiulcerants and other GI drugs, 1994–2004 (second quarter), from a UK primary care database. Source: IMS Health.

Hospital admissions for gastrointestinal bleeds

There was an immediate increase in the hospital admission rate for GI bleeds of 3.66 per 100,000 per quarter after the 1998 legislation (p = 0.012). However, the rate of change of GI admissions during the whole post-legislation period declined by 0.4 per 100,000 per quarter (p = 0.012) compared with beforehand (Table 22 and Figure 8).

FIGURE 8.

Rates of hospital admissions in England for GI haemorrhage, 1994–2004 (second quarter).

During the study period, the rate of admission for GI bleeds was highest in the 65+ years age group and lowest in the 15–34 years age group. Trends in admission rates showed a decline after the legislation in some age groups (Figure 9), including in those aged 55–64 years (estimated mean change in the trend from before to after the legislation = − 0.63, 95% CI − 1.12 to − 0.14, p = 0.014) and 65+ years (mean change in trend = − 1.34, 95% CI − 2.06 to − 0.62, p = 0.001).

FIGURE 9.

Rates of hospital admissions in England for GI haemorrhage, 1994–2004 (second quarter), by age group.

Prescriptions for gastrointestinal drugs

There was a steady overall increase in prescriptions for antiulcerants following the legislation (p < 0.001) (see Table 21 and Figure 7). The increase was significant in the younger age groups [mean change in trend (95% CI) for those aged 15–34 years, 35–54 years and 55–64 years = 9.41 (7.03 to 11.79), 38.25 (17.09 to 59.42) and 90.37 (69.80 to 110.94), respectively (all p ≤ 0.001)], but not in those aged 65+ years [mean change in trend (95% CI) = 22.21 (− 130.76 to 175.17), p = 0.77].

There was no evidence of an increase in prescribing of other GI drugs following the legislation (p = 0.85) (see Table 21). This applied to all age groups except those aged 15–34 years, in whom there was a gradual rise in prescribing over time [mean change in trend (95% CI) = 9.55 (3.32 to 15.77), p = 0.004].

Limitations

We were unable to examine the impact of the legislation on OTC sales of analgesics, as this information was not available. This is important as any impact might be expected to occur primarily in relation to OTC sales. However, use of the data on admissions for GI bleeds and drugs used to treat GI disturbances allowed us to examine any negative effects of possible changes in use of NSAIDs on gastric health.

Conclusions

Prescribing of NSAIDs increased following the 1998 legislation on pack sizes of paracetamol and other analgesics. However, there was no clear evidence that the legislation had a specific impact on prescribing of NSAIDs as the gradual increase in prescribing of these drugs over time was also seen for other analgesics. There was also no evidence that GI bleeds increased as a result of any altered prescribing (or changes in OTC sales) associated with the legislation. However, a gradual increase in prescribing of antiulcerants could have offset changes in the incidence of GI symptoms.

Data types

Statistical analyses

Analyses of trends in prescribing and deaths were conducted using Stata version 10.0. We used interrupted time-series analysis to estimate changes in levels and trends in prescribing and deaths following the CSM announcement of the withdrawal of co-proxamol. This method controls for baseline level and trend when estimating expected changes in the number of prescriptions (or deaths) due to the intervention. 81

Specifically, segmented regression analysis80 was used to estimate the mean quarterly numbers of prescriptions and deaths that might have occurred in the post-intervention period without the CSM announcement, and the numbers of prescriptions and deaths that actually occurred following the CSM announcement. The latter were obtained from best-fitted data lines from the regressions and are better estimates than taking the average of the actual values. The beginning of 2005 was chosen as the point of intervention. Thus, our data comprised 28 quarters in the pre-intervention segment and 12 in the post-intervention segment for the evaluation of the initial impact of the withdrawal (2005–7) and 24 for the evaluation of the longer-term impact (2005–10). Slope and level regression coefficients were used to estimate the mean quarterly absolute differences in prescriptions and deaths (first, at the mid-point of the withdrawal phase, midway between quarter two and quarter three of 2006; and, second, at the mid-point of the post-intervention period, midway between quarter four of 2007 and quarter one of 2008).

Preliminary analyses indicated some autocorrelation in the data; therefore, the Cochrane–Orcutt autoregression procedure was used (rather than ordinary linear regression) to correct for first-order serially correlated errors. The Durbin–Watson statistic of all final models was close to the preferred value of 2, indicating that no serious autocorrelation remained.

Impact of initial withdrawal phase (2005–7)

Because the impact of the initial withdrawal phase is included in our longer-term evaluation to the end of 2010 (see below), we will only summarise the results of our investigation of this phase.

Prescriptions

There was a 59% reduction in prescribing of co-proxamol in the initial withdrawal period (Table 23 and Figure 10), with increased prescribing of some other analgesics but also decreased prescribing of NSAIDs and dihydrocodeine.

TABLE 23 - Changes in the numbers of prescriptions and poisoning deaths involving co-proxamol, other analgesics and all drugs in England and Wales, 1998–2007, associated with the CSM announcement

Using interrupted time-series segmented regression analysis80 in which the intervention point is taken as the end of 2004 (the CSM announcement on the withdrawal of co-proxamol, January 2005).

Estimated for the mid-point quarter of 2005–7.

Absolute difference between the estimated number with the CSM announcement and the estimated number without the CSM announcement, taken at the mid-point of the post-intervention period.

95% CIs taken from Stata results or calculated according to Zhang et al. 82

Other analgesics include co-codamol, codeine, co-dydramol, dihydrocodeine, NSAIDs, paracetamol and tramadol.

Data type	Estimation of the absolute effect of the CSM announcement, 2005–7a
	Mean quarterly estimated no. without announcementb	Mean quarterly estimated no. with announcementb	Mean quarterly change, 2005–7c (95% CId)
Prescriptions (×1000)
Co-proxamol	1465.1	605.7	− 859 (− 1065 to − 653)
Co-codamol	2524.7	3024.6	500 (459 to 540)
Codeine	534.6	578.0	43 (31 to 55)
Co-dydramol	1018.2	1140.0	122 (99 to 145)
Dihydrocodeine	634.6	600.0	− 35 (− 68 to − 2)
NSAIDs	5633.8	4581.0	− 1053 (− 1186 to − 920)
Paracetamol	2947.0	3330.0	383 (268 to 497)
Tramadol	1130.1	1193.9	64 (− 5 to 133)
Deaths: suicide and open
Co-proxamol	39	20	− 19 (− 29 to − 8)
Other analgesicse	40	45	6 (− 4 to 16)
All drugs (except co-proxamol and other analgesics)	204	186	− 18 (− 39 to 3)
All drugs	283	252	− 31 (− 66 to 3)
All causes	1152	1130	− 22 (− 89 to 45)
Deaths: suicide, open and accidental
Co-proxamol	47	25	− 22 (− 32 to − 13)
Other analgesicse	57	62	5 (− 9 to 19)
All drugs (except co-proxamol and other analgesics)	350	379	29 (− 18 to 76)
All drugs	452	466	14 (− 46 to 74)

FIGURE 10.

Trends in prescriptions dispensed for co-proxamol and seven other analgesics in England and Wales, 1998–2010.

Impact of longer-term withdrawal (2005–10)

Deaths involving co-proxamol and other analgesics

Table 24 shows the numbers of deaths reported as suicide (including open verdicts) or accidental poisoning between 1998 and 2010 involving co-proxamol alone and those involving the seven other analgesics alone (both with and without alcohol), together with all drug-poisoning deaths and all suicides. Between 2005 and 2010 there was a marked reduction in the number and proportion of all poisoning deaths recorded as suicide that involved co-proxamol. Although the number of deaths involving other analgesics receiving a suicide or an open verdict was lower in 2007–10 than in earlier years, their proportionate contribution to total drug-poisoning deaths increased somewhat (presumably because the overall number of deaths declined in line with the reduction in co-proxamol deaths). There was a similar marked decrease in number of deaths involving co-proxamol when accidental deaths were included, but without an increase in percentage of poisoning deaths involving other analgesics (although the number of accidental poisonings with other analgesics increased slightly in 2008–10). Exclusion of NSAIDs made little difference to the results of these analyses.

Figure 11 shows graphically the impact of the initial withdrawal phase (2005–7) and the full withdrawal (2008–10) on both deaths involving co-proxamol and deaths involving other analgesics. There was some reduction in number of deaths involving co-proxamol in the years preceding the beginning of the initial withdrawal phase (1998–2005), then a steady, marked reduction during the initial withdrawal phase (2005–7), with a further, smaller reduction during the first 3 years of the full withdrawal (2008–10). The number of deaths involving other analgesics did not appear to change markedly during these periods.

The overall apparent beneficial changes following the withdrawal of co-proxamol were confirmed when the data were subjected to interrupted time-series regression analyses (Table 27). During the 2005–10 period, there was a highly significant mean quarterly reduction in number of deaths involving co-proxamol alone that received a suicide (including open) verdict (− 21, 95% CI − 34 to − 8 deaths), and a similar reduction when accidental deaths were included (− 25, 95% CI − 38 to − 12 deaths). There was no significant change in number of deaths involving the seven other analgesics combined. When we examined these drugs separately (see Table 25), none showed a significant step or trend change associated with the withdrawal of co-proxamol. There were increases in numbers of deaths involving co-codamol and codeine during the overall study period, but these began well before the withdrawal of co-proxamol. All drug-poisoning deaths (excluding those involving co-proxamol and other analgesics) receiving a suicide or open verdict were reduced following the withdrawal of co-proxamol, although when accidental deaths were included there was a significant increase of 84 deaths per quarter (95% CI 28 to 141 deaths per quarter) (see Table 27).

FIGURE 11.

Deaths in England and Wales from poisoning with co-proxamol and other analgesics, 1998–2010 (suicide and open verdicts, substances taken alone, with or without alcohol).

TABLE 27 - Changes in numbers of poisoning deaths involving co-proxamol, other analgesics and all drugs in England and Wales, 1998–2010, associated with the CSM announcement in January 2005

Using interrupted time-series segmented regression analysis in which the intervention point is taken as the end of 2004 (the CSM announcement on the withdrawal of co-proxamol, January 2005).

Estimated for the mid-point quarter of 2005–10.

Absolute difference between the estimated number with the CSM announcement and the estimated number without the CSM announcement, taken at the mid-point of the post-intervention period.

95% CIs taken from Stata results or calculated according to Zhang et al. 82

Other analgesics include co-codamol, codeine, co-dydramol, dihydrocodeine, NSAIDs, paracetamol and tramadol.

Cause of death	Estimation of the absolute effect of the CSM announcement during 2005–10a
	Mean quarterly estimated no. without announcementb	Mean quarterly estimated no. with announcementb	Mean quarterly change 2005–10c (95% CId)
Suicide and open
Co-proxamol	34	13	− 21 (− 34 to − 8)
Other analgesicse	38	42	4 (− 8 to 16)
Other analgesicse (except NSAIDs)	38	40	2 (− 10 to 13)
All drugs (except co-proxamol and other analgesics)	201	181	− 19 (− 49 to 10)
All drugs	275	236	− 39 (− 87 to 9)
All causes	1129	1140	11 (− 81 to 104)
Suicide, open and accidental
Co-proxamol	40	15	− 25 (− 38 to − 12)
Other analgesicse	55	63	8 (− 7 to 24)
Other analgesicse (except NSAIDs)	54	61	7 (− 7 to 21)
All drugs (except co-proxamol and other analgesics)	336	420	84 (28 to 141)
All drugs	430	499	69 (− 7 to 144)

The estimated reduction in number of deaths involving co-proxamol alone between 1998–2004 and 2005–10 was 61% for suicide and open verdict deaths and 62% when accidental deaths were included. This reduction equated to approximately 500 fewer deaths from suicide between 2005 and 2010 than would be expected without the withdrawal, and 600 fewer deaths when accidental poisonings were included. As can be seen in Table 24, during the full withdrawal phase (2008–10) there was an average of just 19 deaths per year (23 per year including accidents). This figure is in contrast to the 228 deaths per year during 1998–2004 (271 per year including accidents).

We have not presented data on deaths involving morphine because the ONS cannot distinguish between deaths resulting from oral administration and deaths resulting from intravenous administration (and between those due to morphine and those due to heroin). Deaths involving oxycodone alone and receiving suicide, open or accidental verdicts increased during the study period from a mean of 2.3 per year between 2001 and 2004 to a mean of 8.2 per year between 2005 and 2010 (including 15 in 2010).

Strengths and limitations

One strength of this study is that it is based on national data and includes reasonably long pre-intervention (7 years) and post-intervention (6 years) periods. Also, the method of statistical analysis (interrupted time-series regression) controls for baseline trends when estimating effects of the intervention on prescribing and deaths, which is preferable to methods such as comparison of changes in proportions before and after the intervention. However, estimates of overall effects do involve extrapolation and hence are subject to some degree of uncertainty. Also, percentage changes during the post-intervention period compared with the pre-intervention period are based on mean quarterly change estimates, which have associated uncertainty, and therefore should be interpreted with caution.

To ensure that we were examining deaths due to specific drugs, we based our mortality calculations solely on deaths involving single analgesics alone (with or without alcohol). There are also considerable numbers of poisoning deaths involving multiple drugs. 110 When co-proxamol is one of the drugs, it is very likely to have been the lethal agent, given its high toxicity. It is possible, therefore, that our findings underestimate the full effects of the withdrawal of co-proxamol. The increasing number of narrative verdicts recorded by coroners in England and Wales29 could have influenced the findings through a small reduction in suicide verdicts, although this would probably have equally affected poisonings with co-proxamol and the other analgesics, and any impact would have been included in our analysis of suicide, open and accidental deaths combined.

We have presented prescription data for the analgesics that we have investigated. For some analgesics, notably paracetamol and co-codamol, sales will mainly be OTC rather than through prescription. Comprehensive OTC sales data are unfortunately not available; this would not, of course, affect the validity of the mortality data.

As noted above, we were unable to investigate the impact of the withdrawal of co-proxamol on deaths involving morphine; however, most of such deaths result from illicit drug use. 110

Finally, we have not been able to assess possible compensatory increases in numbers of deaths involving methods other than poisoning that may have occurred following the withdrawal of co-proxamol. This is because of the range of other potential methods of suicide and also the likely temporal effects of other influences, such as the economic recession. However, as individuals tend to show a preference for particular methods of suicide,108 it is unlikely that any such effect would have been substantial.

Prescriptions

Quarterly estimates for the years 2000–6 for the UK for the numbers of antidepressant prescriptions dispensed in the community were obtained from the Medical Data Index and supplied by IMS Health. 125 Mid-year population estimates for the UK for the years 2000–6 were obtained from the ONS. 126 Rates of prescribing were calculated per 100,000 population for people of all ages for the years 2000–6 combined. We were unable to take account of average numbers of prescriptions per course of treatment.

Deaths

Information on drug-poisoning deaths receiving a verdict of suicide or death of undetermined intent that involved the antidepressants under investigation was provided by the ONS based on death registrations during 2000–6 in England and Wales. In England and Wales it has been customary to assume that the majority of injuries and poisonings of undetermined intent are cases in which the harm was self-inflicted but there was insufficient evidence to prove that the deceased deliberately intended to kill him- or herself. 32,109 In our investigation of coroners' practices (see Chapter 2) we found that 80% of possible suicide deaths given an open verdict were judged by the research team to be probable suicides. We have restricted our analyses to deaths involving single drugs, or single drugs and alcohol. Data were obtained for males and females separately. Mortality rates were calculated as the number of deaths per 100,000 population in England and Wales for those aged ≥ 10 years.

Self-poisoning

Self-poisoning data came from three centres currently involved in the Multicentre Study of Self-harm in England. 1 Data in the present study were for all patients who presented to emergency departments at general hospitals in Oxford (one hospital), Manchester (three hospitals) and Derby (two hospitals) during the study period with an overdose of antidepressants, including those involving other drugs (with or without alcohol).

Episodes of self-poisoning involving the antidepressants in those aged ≥ 10 years for the defined population areas of Oxford City, Manchester City and Derby Unitary Area were included in the study. Mid-year population estimates for these areas for the years 2000–6 were obtained from the ONS. 126 Rates of self-poisoning per 100,000 population were calculated for these areas combined.

Statistical analysis

Fatal toxicity index (mortality to prescriptions rate ratios)

There was significant heterogeneity in fatal toxicity within the TCA group [χ² = 365.02, degrees of freedom (df) = 6, p < 0.001]. The mortality to prescriptions rate ratio was considerably higher for dosulepin and doxepin than for amitriptyline (Table 31). This is reflected in the toxicity ratios relative to amitriptyline for these drugs (dosulepin 3.2, doxepin 2.5). The fatal toxicity ratios were generally far higher in males than in females but the relative toxicity indices were mostly similar in males and females.

The relative toxicity index for venlafaxine (0.46) was approximately half that for amitriptyline but over five times higher than that for the SSRIs (0.08). The relative toxicity index for mirtazapine (0.32) was slightly lower than that of venlafaxine.

There was significant heterogeneity in fatal toxicity within the SSRI group (χ² = 28.76, df = 4, p < 0.001). The fatal toxicity index for citalopram was three times higher than that for the other SSRIs (1.7, 95% CI 1.3 to 2.3, vs. 0.6, 95% CI 0.4 to 0.8).

Case fatality index (mortality to self-poisonings rate ratios)

The findings based on mortality to self-poisonings rate ratios were essentially the same as for fatal toxicity, including relatively high case fatality of dosulepin and doxepin within the tricyclics, similar case fatality of venlafaxine and higher case fatality of citalopram than the other SSRIs (Table 32).

Comparison of fatal toxicity and case fatality

There was a very high correlation between the rankings of the results of the fatal toxicity and case fatality approaches to estimating relative toxicity of the specific antidepressants (Spearman's ρ = 0.99, p < 0.001).

Study sites

We intended to carry out the study in the stratified random sample of 32 hospitals in England identified in our earlier investigation. 15 Only hospitals with an emergency department were included in the original sampling process. The sample was stratified so that four hospitals were selected within each of the eight former health regions in England and to ensure that small and large hospitals were included, as well as those with high and low self-harm readmission rates. In total, 31 of the original 32 hospitals agreed to take part in the current study. The one hospital that declined to participate was replaced by an alternative, randomly selected hospital from within the appropriate stratum.

Local approvals

Approval to carry out the study was sought through the research and development departments at each NHS trust through completion of a Site Specific Investigation Form [via the Integrated Research Approval System (IRAS)]. As part of this process we approached potential local collaborators at each trust, commonly through the assistance of appropriate local and national networks [e.g. the Mental Health Research Network (MHRN) and Comprehensive Local Research Network].

Service interviews

Once approval had been obtained, the local collaborators identified the key mental health and emergency deparatment staff involved in the provision of self-harm services. These personnel were interviewed by telephone or in person about current service structures (see Appendix 5). Using the staff reponses, hospitals were then rated on a measure of service quality developed as part of the previous study15 and based on the Royal College of Psychiatrists guidelines for the general hospital management of self-harm132 (Box 1). Each item was scored as ‘1’ or ‘0’ depending on the presence or absence of a particular aspect of the service. For two items (6 and 10), for which such a strict categorisation was not possible, scores of ‘0’, ‘0.5’ or ‘1’ were given in consultation with the research team.

Ethical approval

This study received ethical approval from Tameside and Glossop NHS Research Ethics Committee in August 2009.

Potential total scores ranged from 0 to 21, with higher scores indicating better-quality services. Information on the hospital catchment area (estimated size; urban, rural or mixed urban/rural) was collected during interviews with staff or from hospital records.

Service audits

With the assistance of our local collaborators and with subsequent approval of the Information Governance departments allowing data sharing between relevant trusts, we facilitated the set-up of audits of self-harm in each hospital through NHS trust personnel or their equivalent (e.g. clinical studies officers who were employed by the trusts in some centres). For each audit, data were collected on all episodes of self-harm in those aged ≥ 18 years occuring during a 3-month period. Service configurations for young people were likely to be markedly different from adult services and therefore those aged < 18 years were excluded from this study. The exact time period varied between centres but all audits took place between May 2010 and June 2011.

Characteristics of the sample

This section is based on each individual's first (index) episode during the 3-month audit period. A total of 6442 individuals presented with 7689 episodes of self-harm at the 32 hospitals during the 3-month audit period. Overall, 56% (n = 3583) of individuals were female and 51% (n = 3274) were aged < 35 years [age range 18–94 years, median age 34 years, interquartile range (IQR) 24–45 years]. Information on ethnicity was not widely available for seven of the hospitals. Data were 85% complete in the remaining hospitals (n = 4333): 93% (n = 4017) of the individuals were white, 3% (n = 124) were South Asian, 2% (n = 78) were black and 3% (n = 114) were from other ethnic groups.

The main method of self-harm was known in 99.7% (n = 6424) of index episodes: self-poisoning with drugs in 79% (n = 5073), self-poisoning (other) in 2% (n = 102), self-cutting in 14% (n = 890) and other methods of harm (including burning, attempted hanging and jumping) in 6% (n = 359). Use of alcohol/drugs with the act was known for 90% (n = 5828) of individuals. Alcohol was taken within 6 hours of the self-harm act in 53% (n = 3111) of cases and recreational drugs in 7% (n = 385). Data were available for previous self-harm episodes in 97% (n = 6237) of individuals and reported as present in 51% (n = 3173) of these. Data were available on whether or not patients were receiving psychiatric treatment at the time of their index self-harm episode for 96% (n = 6181) of individuals, and 32% (n = 1982) of these were known to be in current psychiatric treatment. Data on psychiatric inpatient care were 97% (n = 6269) complete, and 10% (n = 636) of individuals had been an inpatient in a psychiatric ward in the 12 months before the self-harm presentation.

FIGURE 12.

Time of presentation: all episodes during the 3-month audit (n = 7622).

Time of presentation was known in 99% of episodes (n = 7622). There was a steady increase in presentations during the day with a peak time of presentation between 2200 and 2300 (Figure 12). The majority (57%, n = 4362) of presentations occurred out of hours (we considered ‘out of hours’ to be 2000 to 0800 and daytime/evening to be 0800 to 2000, based on information of routine service availability from hospital interviews).

Management in hospital

A psychosocial assessment by a mental health specialist took place in 58% of index episodes (Table 33). A total of 66% (n = 2651) of these were carried out by a mental health nurse (including mental health liaison nurses and those from specialist self-harm teams and crisis teams), 20% (n = 683) by a psychiatrist (any grade) and 4% (n = 142) by another mental health professional (such as a social worker). Out-of-hours presentations were slightly less likely to receive psychosocial assessment than presentations during working hours (56% vs. 60%, p = 0.001). Daytime and evening presentations were more likely to be assessed by a psychiatrist than out-of-hours presentations (22% vs. 18%, p = 0.012).

A total of 2155 out of 2660 individuals (81%) were assessed within 24 hours of their presentation. The median number of hours between time of hospital presentation and time of assessment was 10 hours (IQR 4–20 hours). For those not admitted to a medical bed, the median time to assessment was 5 hours (IQR 3–8 hours) and this did not vary by time of presentation (working hours vs. out of hours). A total of 42% (335/798) of assessments occurred within 4 hours of presentation in this group. Amongst all assessed individuals, those using alcohol at the time of self-harm had a median of 11 hours (IQR 6–20 hours) wait to be assessed compared with 9 hours (IQR 5–20 hours) when alcohol was not used. Over half of the index episodes resulted in admission to a medical bed. Amongst these, 63% (n = 2245) were admitted to an emergency department observation/assessment ward or equivalent, 32% (n = 1145) to a general ward and 5% (n = 194) to other types of medical beds including intensive care units. A total of 7% of presentations resulted in admission to a psychiatric bed and in 30% (n = 1828) of presentations patients were referred to specialist mental health outpatient services (including psychiatric care, crisis teams, community mental health teams and statutory drug and alcohol services). In 18% of episodes patients were referred to other, non-statutory services, most commonly drug/alcohol services (25%, n = 255), counselling/psychotherapy (16%, n = 167), social services (13%, n = 128) and prison/police custody (12%, n = 123).

In total, 20% (n = 1313) of individuals reattended with a repeat episode of self-harm within 6 months of the index episode. This ranged from 6–27% (IQR 18–24%) in the 32 hospitals.

Aggregate-level outcome analysis

This section presents results at the hospital level. Analyses relating to the 21-item service scale and admission to general hospital were conducted using data from 32 hospitals. Those relating to specialist mental health management were carried out on data from the 31 hospitals in which mental health records were accessed.

The Spearman's rank correlation coefficient between the proportion of individuals receiving psychosocial assessment and the rate of 6-month repetition was 0.16 (p = 0.38) (Figure 13). There was no apparent association between 6-month repetition and the rate of hospital medical admission (Spearman's ρ = 0.15, p = 0.413), psychiatric admission (Spearman's ρ = 0.077, p = 0.679) or specialist outpatient referral (Spearman's ρ = 0.20, p = 0.28). The rate of repetition did not vary according to total service scale score (Spearman's ρ = − 0.065, p = 0.73) (Figure 14) or individual aspects of the scale.

FIGURE 13.

Association between level of psychosocial assessment and level of repetition within 6 months.

FIGURE 14.

Association between service scale score and level of repetition within 6 months.

Individual-level outcome analysis

After adjusting for baseline characteristics, we found that psychosocial assessment was associated with a decreased risk of repeat self-harm, although this was non-significant at the p = 0.05 level (Table 35). In a post hoc analysis we compared the association between repetition and main speciality of assessor. The adjusted hazard ratio (HR) for repeat self-harm in individuals assessed by a psychiatrist (n = 629; based on complete data) compared with those not assessed was 0.78 (95% CI 0.64 to 0.94, p = 0.01). Psychosocial assessment by a mental health nurse (n = 2620; based on complete data) was not associated with a significantly decreased risk of repetition (HR 0.89, 95% CI 0.77 to 1.04, p = 0.13).

Specialist mental health follow-up was associated with a significantly increased HR, reflecting the increased risk of this group. Adjusting for characteristics associated with risk reduced the HR somewhat. There was a similar process for psychiatric inpatient admission.

Comparison between 2001–2 and 2010–11 data

Our earlier study15 in 2001–2 was based on episodes of self-harm, not individuals; therefore, all comparisons here are episode based. Comparisons are based on the 31 hospitals that were included in both studies.

Main findings in relation to objectives

We collected data on over 6400 individuals who had presented with self-harm to 32 sites across England in a 3-month period. The characteristics of our sample were broadly consistent with those in other hospital-based studies in the UK,1 with the majority of episodes related to self-poisoning, and self-harm being more common in younger age groups and women. Alcohol was involved in over half of the episodes, and over half of the individuals had a previous history of self-harm. There was marked variability in service provision with a 3.5-fold difference between hospitals in the proportion of individuals who received a specialist assessment, a fourfold difference in medical admission and a fivefold difference in the proportion of individuals receiving specialist follow-up. Overall, 4 out of 10 individuals left hospital without having had an assessment with a mental health specialist.

We found little evidence that aspects of care were related to outcome in terms of repetition of self-harm in our aggregate analysis. Findings based on area-level analyses should not be used to necessarily infer individual-level associations (the ecological fallacy); however, aggregate-level associations are really the only means of evaluating services as a whole. It could be that higher levels of assessment and admission are partly markers of good-quality care. This may have the effect of increasing engagement and making individuals more likely to present to hospital when they self-harm. There was limited evidence from the individual-level analyses that psychosocial assessment might be associated with a 12% reduced risk of repetition but, again, these findings are based on observational data and should be interpreted cautiously. We attempted to adjust for individual characteristics at baseline but our findings could simply reflect unmeasured confounding factors. The possible protective effect of assessment is, however, consistent with results from previous studies. 136 In a post hoc analysis there was an interesting possible difference in the effect of assessment according to the main speciality of the assessor. However, these results should be interpreted cautiously as psychiatrists and nursing staff are likely to have assessed distinct groups of patients.

Considering changes over time, we found that the overall number of episodes of self-harm in the study centres increased by around one-quarter. However, our data suggested that this might partly reflect service consolidation rather than a true increase in incidence. Disappointingly, given clinical guidelines and policy emphasis, variations in service provision were as wide as 10 years previously. The proportion of individuals receiving assessment from specialist services had remained static and the level of specialist follow-up had decreased, perhaps because of a greater involvement of voluntary organisations in services (nearly one-fifth of individuals received follow-up from non-statutory services in this study, but these data were not available for the earlier time period). The increased proportion of episodes resulting in admission may well have reflected the greater use of emergency department observation and assessment wards rather than an increase in acute admissions to general medical beds. This may partly have been driven by the policy emphasis on reduced waiting times in emergency departments. 137 A measure of service quality developed as part of the previous study did show an improvement in 23 of the 31 hospitals, with a 26% improvement in the median service quality score. The individual service items that showed an improvement in the greatest number of hospitals related to the availability of supervision, social services input and joint service planning meetings between mental health and acute care services as recommended by NICE guidelines. 13

Permissions and data sharing

One major unanticipated challenge that we encountered was securing local research governance permissions. As this study involved acute and mental health services, this meant making a total of 62 separate applications for local approval (not 64 trusts as some participating mental health trusts provided services over multiple sites). Processes and the time to approval varied between organisations. Even when local approvals had been given, additional permissions were required in some instances (e.g. from IT departments). Because of the nature of the study, acute and mental health services needed to share data. Despite both services providing care to the same group of patients – those who had harmed themselves – data-sharing agreements, when required, were sometimes difficult (and in one case impossible) to achieve within the study period. Identifying local staff who could collect data was also a major challenge but was helped considerably by clinical studies officers (from local and topic-specific research networks).

Strengths and limitations

This was a large study of hospital attendances following self-harm using recent data at individual level and episode level from multiple sites randomly selected from across England. Different electronic systems in the study hospitals required individual methodologies to identify and capture data. However, the robustness of the data was affirmed at each study site by the individual data collectors, and pilot data collection was carried out in each centre.

Nevertheless, our findings should be interpreted in the context of a number of methodological limitations. We collected data on self-harm attendances to hospital only and did not record outcomes or repeat episodes that did not come to the medical attention of the participating hospitals. As a country-wide audit, the data sources were based on clinical records rather than in-depth interviews. If there was no information in the notes of an item of interest then this was coded as absent. As a consequence, some of our findings may underestimate the true prevalence of particular characteristics or associated factors.

Another potential weakness was that we did not collect data on aspects of care offered after discharge from hospital, such as a patient receiving an assessment by the mental health team in the community. In addition, the mental health follow-up variable described referral for such follow-up, but we were unable to determine whether or not patients actually attended.

Although data completeness was high overall and we were able to include 31 of the original 32 sites, there were some difficulties relating to individual hospitals. For example, in one acute trust, patients received mental health care from a variety of mental health providers, so mental health data were unavailable for a small proportion of presentations. Within another site, data-sharing agreements between acute and mental health trusts could not be achieved, which again resulted in missing mental health data.

Implications for future research

One of the main difficulties that we encountered in this study was securing local permissions. Future multicentre work that is national in scope would benefit from a simplified system of centralised approval. Our understanding is that the NHS Health Research Authority (www.hra.nhs.uk/) is currently considering how this might be achieved. Our study demonstrates the invaluable contribution that clinical studies officers can make to the conduct of research when appropriate access permissions have been granted. Future work on this scale would not be possible without such posts.

Research in this area should continue to monitor the variability of self-harm services. In this study we found a discrepancy between process measures (e.g. proportion of people receiving psychosocial assessments), which did not improve, and quality scores, which did. An area of future research might be to determine what aspects of quality are important determinants of outcome and how they should be measured. The impact of management on outcomes is an important but methodologically challenging area. We need to understand which aspects of treatment are beneficial in routine practice and why. We also need to understand in which groups of individuals treatments might have the most impact. The role of psychosocial assessment warrants particular attention. 136 Randomised trials of national service-level interventions are not always possible or practical, and alternative observational designs (e.g. pre–post studies138) may be worthwhile.

Service user interviews

Service users who had recently attended the emergency departments of three hospitals in a city in the north-west of England following self-harm were given information packs inviting them to participate in the study. ‘Non-fatal self-harm’ was defined as intentional self-poisoning or self-injury, irrespective of motivation. 25 We employed purposive sampling149 for service user interviews to select male and female adult patients of varying age and self-harm history and directed the distribution of information packs accordingly. People fulfilling these criteria were likely to have different needs and preferences regarding support following a self-harm episode. Semistructured interviews were conducted using a topic guide (see Appendix 7) with prompts that allowed the researcher flexibility to follow up issues that emerged. During the interviews, the researcher introduced different contact-based interventions and asked the participant to comment on them.

Staff focus group

A purposive sample of key informants from clinical staff in three relevant service areas (emergency department, liaison psychiatry and primary care) and workers from the voluntary sector were recruited. A focus group was conducted with most of the staff participants (n = 8) using a topic guide (see Appendix 8). The focus group was introduced as an open discussion with the aim of informing the development of a pilot intervention for people who had self-harmed. Additional individual interviews were conducted for those unable to attend the focus group, and their contributions within the thematic coding frame identified from the group discussion analysis were included in the results.

Data analysis

We conducted thematic analyses150 of the transcribed interviews with service users (mean time of service user interview was 57 minutes) and staff and the focus group discussion (70 minutes' duration). Analytical themes were identified to summarise the data and were coded using computer-assisted qualitative data analysis software (NVivo version 7; QSR International, Southport, UK), to facilitate comparison and develop descriptive accounts. Quotations were selected to illustrate particular themes.

Researcher orientations

The main investigator (JC) conducted all of the interviews and facilitated the focus group. She is a trained psychiatric nurse, with experience of working with those who self-harm. The other investigators included an academic psychiatrist (NK), an experienced qualitative methodologist (JD) and a doctoral student in qualitative methodology (CH). Two researchers (including the main investigator) (JC, CH) analysed the interview transcripts, with another blind coding the initial interviews (AOS). We acknowledge that our professional backgrounds may have influenced the themes derived from the data, although effort was made through the methods of constant comparison and discussion within the research group to keep our themes grounded in the data.

Participants

Of the 17 service users who expressed interest in the study, 11 (SU1–SU11; six females, five males), aged 18–53 years, participated. The predominant method of harm was self-poisoning (10/11), although four of the participants described mixed methods.

Of 16 clinical staff from relevant service areas (emergency department, liaison psychiatry, primary care and the voluntary sector) invited to participate, eight (SP1–SP8) agreed to take part in the focus group and two (SP9, SP10) in individual staff interviews.

Thematic analysis

A number of themes emerged from the data (Table 39 provides illustrative quotations). These included the need for support and encouragement and for some form of intervention shortly after hospital attendance for self-harm. Staff agreed that the immediate period after discharge was a vulnerable time, and both service users and staff expressed a need for contact to be proactive, as service users might find it difficult to initiate obtaining help.

An important aspect of care mentioned by both groups was the ‘genuineness of the service’, although there was a subtle difference in interpretation. Staff were more concerned with the style of the service, whereas service users were more concerned with the manner of the clinicians' responses. Staff felt that, for the intervention to be worthwhile and perceived as genuine, it should be integrated into current services, and there should be some consistency of staff delivering the intervention.

Potential mechanisms emerged from the transcripts that provided insight into how a contact-based intervention might be of benefit. Letters expressing concern were perceived as a gesture of caring that was ultimately reassuring for the recipient as he or she experienced being looked after (‘contained’). For some, this would prevent further self-harm, as contact with other people would counter against loneliness and fear of abandonment, and acknowledging their pain might reduce their sense of isolation or suffering alone.

Promoting engagement with services by providing people with relevant support information was viewed by staff as an important function, particularly for those who did not engage with statutory treatment. Raising awareness of other options was a particular benefit highlighted by the service users.

Practical issues and problems relating to the proposed interventions included a preference by service users for a mode of contact that was immediate (at least initially), perhaps by telephone; however, it was stated that letters might be acceptable later. The letters were viewed as potentially counteracting feelings of loneliness, and the proposed wording was considered caring, although there was no consensus on the preferred frequency of contact. There was agreement between service users and staff that the intervention should be delivered by mental health specialists, as they had the necessary skills and experience of dealing with people in crisis; however, they suggested that the impact on the staff of dealing with these situations should be considered. Issues of privacy, confidentiality and access were raised in the context of using the internet as a means of communication, similar to concerns relating to text messaging in the Bristol-led pilot study (see Chapter 8). As in the Bristol study, letters and the telephone were seen as the preferred mode of contact, although illiteracy, no fixed address, not having a telephone and not having any telephone credit were raised as potential barriers to a contact-based intervention.

Strengths and limitations

As with many qualitative studies, the small sample size limits the generalisability of the findings and our results need therefore to be interpreted with caution. 151 Those who self-harm are a difficult group to engage in research,152,153 and it may be that the views presented here differ from those of potential participants who did not respond to the invitation to participate in the study.

Informing the design of the proposed intervention

Preferences around timing, content, duration, method and delivery of contact were highlighted in the service user accounts. We identified some of the aspects of contact interventions that people valued, in particular the need for encouragement and support soon after discharge. Telephone calls were mentioned as providing immediacy of contact and a means of proactively engaging the service users. Service users also expressed positive views of simple messages of concern delivered by post. They considered that any contact provided should be genuine and caring, with, for example, letters being worded appropriately. Awareness of other service options was also considered useful.

The findings suggested that an appropriate design for an intervention might be the provision of an information leaflet, telephone calls (soon after discharge) and then letters (offering continuity of contact). In the next phase of the work we piloted an intervention informed by these findings.

Design

The design for this pilot study was a parallel-group randomised controlled trial. Individuals were randomised to the intervention and treatment as usual or to treatment as usual alone.

Inclusion criteria

Potentially eligible participants were adults aged > 18 years who presented to the emergency departments of two study hospitals with self-harm during a 6-month period and were resident in Manchester. ‘Self-harm’ was defined as ‘an act of intentional self-injury or poisoning irrespective of the apparent purpose of the act’. This included acts of self-harm regardless of the associated degree of suicidal intent or medical seriousness.

Exclusion criteria

The following groups of patients were excluded from the study: psychiatric inpatients, patients with a general hospital admission of > 7 days, those of no fixed abode or with no telephone access and those unable to give informed consent during the first telephone call. Patients who were not able to understand English were also excluded from the study as the intervention involved telephone calls and would not be suitable for this group.

Sample size

We aimed to achieve a maximum of 50 patients in each of the intervention and usual treatment groups. This was a pilot study designed to inform a larger multicentre trial and provide preliminary data that would enable a more precise calculation of sample size in a main trial. We estimated that recruitment would be required for a minimum of 3 months.

Recruitment and consent procedure

Once clinicians working in the clinical teams at the two hospitals had ascertained patients' eligibility, patients were sent by post an invitation letter, an information sheet about the study and a copy of the consent sheet. Patients were contacted by a clinical researcher by telephone at least 3 working days later, who then invited them to take part in the study and audio recorded their verbal consent. Contact details available on hospital medical records systems were verified with the patients' GPs if required. The patients' GPs and/or care co-ordinators were contacted with the consents to inform them of their patients' participation in the study. Those patients randomised to the intervention arm were sent an advice leaflet (see Appendix 9) and a notification of change of address form. Participants in both groups were advised that their routine treatment would not be affected by their participation in the study.

Randomisation procedures

The randomisation process was conducted by a remote internet-based randomisation service (www.sealedenvelope.com). The allocation sequence used a computer random number generator to select randomised permuted blocks with randomly varying block sizes. Separate allocation sequences were generated for each of the two collaborating hospitals/centres, and patients had an equal probability of assignment to each group. Patients were then informed of their allocation status. An e-mail confirmation of the identification number, the allocated treatment and the date and time of randomisation was sent to the clinical researcher. The randomisation list was kept separately from the collection of outcomes.

Blinding

Because of the nature of the intervention, all patients were unblinded and fully aware of their allocation to the intervention group or the control group. A researcher who was also unblinded administered the intervention, including the initial telephone call and the mailing of the letters. However, a researcher blind to allocation status collected outcome data using self-harm databases.

Intervention

Patients in the intervention group received an information leaflet (see Appendix 9) listing local sources of help, two telephone calls and then a series of letters. The leaflet provided information on national and local sources of help and was reviewed by the local service provider (Manchester Mental Health and Social Care Trust). It was posted to participants as soon as possible after consent, and another copy was included in the first letter. The telephone calls were made by clinical researchers using a semistructured format. The purpose of the telephone calls was to make contact and facilitate access to appropriate treatment. The first telephone call was attempted at least 3 days after the invitation letter was sent. The second telephone call was made a week later. The letters were sent to the participants over a 12-month period (schedule: 1, 2, 4, 6, 8 and 12 months), which is similar to the methodology used in other studies. The format of the letters followed a standard pattern and included a general statement of concern (see Appendix 10 for an example of a typical letter), but they were modified when indicated to personalise them in response to individual circumstances (e.g. ‘we are sorry to hear that you have been in hospital recently’). The first letter included further details of how to opt out of the study and a request that participants notify us of any change of address.

Usual treatment

Both the intervention and control groups received usual treatment, which was similar at the two participating sites. A mental health liaison nursing team was in operation 7 days a week between 0900 and 2100, with out-of-hours care provided by the duty psychiatrist. One-quarter of patients in the centres are referred for mental health follow-up after a specialist assessment, with one-fifth referred to social services or to the voluntary sector. 135

Study outcomes

We assessed the proportion of patients who accepted the invitation to join the study and gave informed consent. We recorded the proportion of patients with at least one repeat episode of self-harm resulting in attendance at a study hospital within 12 months, identified from hospital information systems. In addition, we noted the number of repeat episodes during the same time period. For all participants we also investigated use of health services to pilot collection of such data for the future economic evaluation of a full trial. Repetition and resource use data were obtained from routine information systems and hospital databases. In addition, patients receiving the intervention were asked in their final letter if they could be approached for their views on the intervention, and a stamped addressed envelope was enclosed for their reply. Within this self-selected group of patients who responded, qualitative interviews (see topic guide in Appendix 11) were conducted after the study period ended to assess the acceptability of the intervention and to help refine it.

Statistical analysis plan

In accordance with the CONSORT (Consolidated Standards of Reporting Trials) principles,154 we have reported all participant flow in the pilot study. All primary analysis was based on intention to treat. Because this was a feasibility study and was not powered to test for a difference between the intervention and usual treatment groups, our analyses were primarily descriptive in nature. We examined the incidence of any repeat self-harm as well as the number of repeat episodes.

Data analysis for the follow-up qualitative interviews

Data analysis for the qualitative interviews (see topic guide in Appendix 11) used methods of constant comparison with repeated scrutinising of transcripts to determine the themes emerging from the data.

Ethical approval

The study was ethically approved by Greater Manchester West North West 9 Research Ethics Committee and the International Standard Randomised Controlled Trial Number is ISRCTN65171515.

Recruitment statistics

In total, there were 250 adult patients resident in the Manchester area who presented to the two study hospitals from 1 November 2010 to 17 May 2011 and who were identified as having harmed themselves. Figure 15 summarises the recruitment of participants to the study.

FIGURE 15.

Flow chart of recruitment to the study. Reproduced with permission from the British Journal of Psychiatry (URL: http://bjp.rcpsych.org/content/203/1/73/suppl/DC1).

Comparison of study participants with eligible non-participants

Four clinical researchers recruited 66 participants from two emergency departments (hospitals A and B) during the study period. The median time between presentation to the emergency department and the information pack being sent was 2 days (IQR 1–6.25 days, range 0–64 days) and between presentation to the emergency department and recruitment to the study was 20 days (IQR 14–29 days, range 5–89 days).

When we compared study participants (n = 66) and eligible non-participants (i.e. those who we were unable to contact and those who declined, n = 84), we found that the two groups were similar in terms of sex (consented participants: males 34.8%, females 65.2%; eligible non-participants: males 42.9%, females 57.1%; χ² = 0.993, p = 0.40) and age [consented participants: mean (SD) age 33.3 (10.2) years; eligible non-participants: mean (SD) age 35.9 (13.4) years].

Comparisons between the intervention group and the treatment as usual group

The two groups were similar with respect to sociodemographic characteristics (Table 40). However, participants in the intervention group were slightly more likely to be white, be unemployed, use methods other than poisoning to self-harm and have a past history of psychiatric treatment or previous self-harm; however, none of these differences reached a conventional level of statistical significance.

Repetition of self-harm within 12 months

The repeat rate for individuals in the intervention group was 34% (95% CI 20% to 52%) whereas the repeat rate for individuals in the usual treatment group was 12.5% (95% CI 4% to 29%). The total number of episodes of repeat self-harm was also higher in the intervention group than in the usual treatment group (41 vs. 7). These differences were statistically significant when tested using logistic regression and negative binomial regression models respectively. The distribution of repeat self-harm episodes in any sample is highly skewed with a small number of individuals accounting for a large number of repeats. 1 In this study, in the analysis of number of self-harm episodes, there was one extreme outlier in the intervention group who alone accounted for nearly half of the total number of repeat episodes. Excluding this individual, the total number of repeat self-harm episodes was 21 compared with 7 (IRR = 3.0, 95% CI 0.84 to 10.68).

In analyses adjusting for baseline clinical factors (Table 41), the OR for repetition and the IRR for number of repeat episodes remained elevated; however, in the case of the OR, this was no longer statistically significant.

One person in the intervention group died following a drug overdose but was not included in the repeat self-harm group in the main analyses, as they did not re-present to the study hospitals with self-harm during the study period. As they had incomplete follow-up they were excluded from the main analyses (see Figure 15). Including this person in the repeat group increased the OR and IRR slightly (adjusted OR 4.97, 95% CI 1.10 to 22.50, p = 0.037; adjusted IRR 7.34, 95% CI 1.61 to 33.45, p = 0.01).

The death was reported to the study sponsor, the ethics committee and the research and development department of the NHS trust. The research team was given access to medical records and the hospital's own serious untoward incident investigation and spoke to the consultant psychiatrist providing care for the patient. There was no indication that the study procedures resulted in the death.

Intervention telephone calls

Most of the telephone calls by clinical researchers to the participants in the intervention arm of the trial (29/33) were rated as ‘good’ in terms of engagement with the patient.

The first intervention call in most cases focused on identifying and reinforcing available support and encouraging self-support, for example discussing triggers to self-harm, anticipating problems and related coping mechanisms and helpful activities, and discussing support that had been useful in the past. The clinical researcher would often signpost relevant support services listed in the information leaflet. Some patients expressed that they were receiving adequate support and were already aware of the information provided in the leaflet; others thought that they might use the information on the leaflet in the future if they needed help. Some participants found the telephone call supportive whereas others thought the face-to-face contact that they were already receiving from mental health staff was adequate.

The second telephone call focused on encouraging engagement with treatment (attending appointments, following up missed appointments, etc.), sometimes clarifying referral status from medical records (in one case because the patient felt unwell at the time of assessment and was unable to assimilate information, and in another case because the mental health worker called at the wrong address and subsequently closed the case as a ‘did not attend’).

Resource use

From hospital records and databases we collected resource use data retrospectively for both the intervention group and the treatment as usual group, from the index episode date to 12 months following the recruitment date. We assessed the feasibility of accessing records and extracting these data, and described the health service activity in both groups. From the emergency department records we recorded the number and types of presentations and disposal from the emergency department (estimated time taken to retrieve information = 5 minutes per individual record). From the mental health records we recorded the number of face-to-face contacts with each patient and the discipline of the person contacting (approximately 10 minutes per individual record). These data are summarised in Table 42. It was interesting to note that, despite a higher number of repeat episodes of self-harm in the intervention group, the overall number of emergency department presentations in the two groups was similar. The non-self-harm presentations in the usual treatment group resulted from a variety of physical complaints as well as specific mental health problems. The intervention group had a greater number of contacts with specialist mental health services [median (IQR) = 2 (0–9) vs. 0 (0–2), p = 0.053].

Evaluation of the intervention by participants

In the final letter (at 12 months after recruitment), participants receiving the intervention were invited to contact the research team to provide their opinion of the intervention and to suggest ways that it could be improved. Out of 33 participants, four responded (all male) but only three were interviewed as one respondent was unable to be contacted subsequently by telephone or letter. Of the three respondents, one repeated self-harm on multiple occasions, one repeated on two occasions and one did not repeat. Five themes emerged from these interviews (see Appendix 12 for illustrations of themes): (1) support received from other services, which was valued but, in the case of a participant receiving care from the voluntary sector only, not always enough; (2) participant needs were too great and the intervention did not meet, or was irrelevant to, needs; (3) useful for signposting/promoting help-seeking behaviour (telephone contact); (4) effects of letters/telephone contact (mitigates against loneliness, calming, prevents further self-harm and provides extra support); and (5) practical issues around the design of the intervention (easy to read, good intervals between letters, wanted letters to continue for a longer time in some cases).

Feasibility

Overall, our proposed methodology proved feasible. Approximately 60% of those who were screened were found to be eligible for participation in the trial. Of those who were eligible, 30% could not be contacted and just over one-quarter (28%) declined. This refusal rate was very similar to refusal rates in previous studies in Western health-care settings140,142,146 but higher than that in an Iranian study. 143 Only one individual dropped out of the trial.

The process of online randomisation worked well and permitted real-time allocation of intervention status during the recruitment telephone call. The trial database allowed efficient tracking of participants and the scheduling and mailing of letters was not associated with any difficulties. There were no adverse incidents associated with trial processes, for example we recorded no instances of letters being sent to incorrect addresses or therapeutic telephone calls to individuals who were not the participants. Comparatively novel aspects of our methodology such as digital recording of consent and opting opt out by SMS were feasible. We restricted ourselves to basic resource use data within secondary care as these resources make up the majority of the costs of self-harm,155 and such data were relatively straightforward to collect.

Recruitment to the study was more challenging than we had anticipated. This may have been in part because we restricted ourselves to adult patients resident in Manchester (who on the basis of previous data make up approximately 60–70% of presentations to the study hospitals135). We did this because of differences in service provision for young people in the study centres, because of lack of evidence of efficacy of contact-type interventions in this group and because the information leaflet contained contact details for Manchester-specific services and helping agencies. Including individuals from a wider catchment area would have necessitated a number of different leaflets and would have added to the complexity of this pilot trial.

Contact information obtained from emergency departments was not always valid. Accessing mental health and GP medical records, when required, improved the accuracy of this information. The recruitment pathway involved multiple staff members involved in a multistage recruitment process. Clinical staff entered names of potentially eligible participants into the database. Designated members of the team screened participants for eligibility and then sent out recruitment packs. If individuals did not opt out they were contacted by one of four clinical researchers. The process from entry by clinicians to determine eligibility to the sending out of information packs created delays in recruitment.

Acceptability

Just over one-quarter of individuals who were eligible for the study refused to take part, but there was only one dropout from the study. Data from the post-intervention telephone interviews were limited but seemed to suggest that individuals found the intervention acceptable and helpful to some extent.

Differences in outcome between groups

We found that a higher proportion of individuals repeated self-harm in the intervention group than in the usual treatment group (33% vs. 13%). We also found that there was a higher number of repeat episodes in the intervention group than in the usual treatment group (41 vs. 7) and this reached conventional levels of statistical significance (p < 0.05). This difference was less marked when one potential outlier was excluded (21 vs. 7). We found an increased number of contacts from specialist mental health staff in the intervention group after recruitment to the trial compared with the treatment as usual group, but interestingly there was no difference in emergency department contacts. In addition, there was one death in the intervention group from an overdose of drugs. Baseline markers of increased risk (previous self-harm, psychiatric history, method of harm other than self-poisoning) appeared to be more common in the intervention group than in the usual treatment group. However, the differences between the groups persisted when we carried out a limited adjustment for these confounders.

Limitations

Our findings need to be interpreted in the context of the methodological limitations of the study. It was a small pilot study in just two centres and we were able to randomise just under half of the eligible participants. Although participants and non-participants were similar in terms of age and sex, it is possible that the two groups differed on other important variables, which could have meant that those included in our study were not typical of others who had harmed themselves. We included only adults resident in Manchester, an area with high rates of deprivation, self-harm and suicide, factors that could feasibly have limited the impact of the intervention. There may also have been problems with our main outcome measure as we considered only repeat self-harm behaviour that presented to the study hospitals. We did not record self-reported or community episodes of self-harm nor consider wider treatment outcomes156 because this would have been beyond the scope of the project. The lower than expected number of participants in the post-intervention interview also meant that we were unable to draw meaningful conclusions from this part of the study.

Implications for future work

In this trial, baseline risk factors appeared to be somewhat unequally distributed between the groups. We do not think this represented a problem with our randomisation (which was carried out using a commercially available online package), but rather was more likely to be a chance occurrence made more likely by our small sample size. The importance of stratifying for baseline variables, particularly previous self-harm, has been highlighted in previous work. 142,157

With respect to recruitment, a simpler and more flexible process could lead to higher recruitment rates. Perhaps the most productive measure would be to embed recruiting staff within clinical teams or to give existing staff dedicated time to carry out recruitment. Ensuring that administrative staff in emergency departments check patient details at reception is likely to be important, but was apparently routine practice in the centres in this study. Individuals who have harmed themselves may be a hard-to-reach population and one possibility would be to recruit patients in the emergency department before they leave (assuming of course that they have the capacity to consent). For those who are not approached in this way, recruitment by telephone, SMS or mail could also be tried. It should be noted that recruitment of sufficient numbers of individuals might be less challenging in hospitals with geographically defined catchments than it was in the current study. Researchers involved in recruitment need to be experienced clinical staff. At times they will be required to manage distressed patients and they also need to engage participants who have complex needs and who may be regarded by some staff as hard to help.

Some flexibility in the intervention might also improve delivery. For example, individuals who prove difficult to reach by telephone might receive only the information leaflet and letter components of the intervention.

This study considered repeat presentations to hospital for self-harm at 12 months as a main outcome. However, this may not be the most important measure of a treatment effect. 144 It does not take into account repeat episodes not resulting in presentation to hospital. Equally, it could be that increased hospital presentation for self-harm is a positive treatment outcome for some people, representing better engagement with services. A broader range of outcomes might be collected in future work (e.g. self-reported repetition, depression, hopelessness, suicidal ideation), but of course this will increase the complexity and cost of such studies.

Intervention development

We carried out a questionnaire survey of current and recent psychiatric inpatients and conducted some qualitative interviews to inform the design of the contact-based intervention. The aim of these studies was to assess (1) patients' preferred means of contact (letter/e-mail/text/telephone); (2) whether or not they might find such letters useful; (3) the wording/contents of the letters; (4) the best frequency of contact; (5) who the letters should be sent by; and (6) any other issues that patients felt were relevant.

Exploratory interviews were undertaken with 10 recently discharged service users sampled from five general adult wards in two acute psychiatric hospitals in Bristol. Initially, consecutive patients were approached on the wards prior to discharge to obtain their consent for a post-discharge interview. As the interviews progressed we then used purposive sampling to identify patient groups who had not been interviewed, including people who lived alone, who had a history of self-harm or depression/affective disorder or who abused alcohol. Ward managers used their professional judgement with regard to which service users should be approached, and were guided by the research team as to which characteristics were of particular interest (see above). The aim of the interviews was to explore participants' attitudes to discharge, their experiences since discharge and their views about the proposed contact-based intervention.

A questionnaire was administered to 71 patients on four general adult psychiatric inpatient wards in Bristol. The questionnaires asked about patients' access to and use of different forms of communication technology and their preferred method of post-discharge contact.

To allow further development of the intervention and to ensure that its implementation was feasible and appropriate within existing discharge arrangements for psychiatric inpatients, we then presented the findings from these studies to groups of ward and community staff associated with the psychiatric inpatient units where the intervention would be piloted.

Pilot study

Development studies

Pilot study

Acceptability and usefulness of the intervention

Nearly two-thirds (102/162, 63.0%) of patients discharged from wards A and B over the recruitment period received at least one letter (Figure 16). The main reason for non-recruitment was residence out of the ward's catchment area. Of those who received at least one study letter, only 45 (44.1%) received all of the study letters (see Figure 16). These represent 27.8% (45/162) of all discharges over the relevant recruitment periods.

An opt-out slip and stamped addressed envelope were enclosed with each mailing so that, if they wished, participants could opt out of receiving the letters. The main reasons for cessation of mailings were patient opt-out (24/102, 23.5%) and readmission (26/102, 25.5%). Although a lower proportion of patients opted out from ward C (18.2%), this was most likely because of the shorter period over which letters were mailed from that ward.

Some people who opted out or who were excluded for other reasons were also subsequently readmitted but were not then recorded as excluded for that reason.

FIGURE 16.

Patient flow for recruitment and mailings.

Service contacts during the 12 months post index discharge

To pilot data collection for an economic evaluation, we collected detailed data on post-discharge contacts made by mental health-care staff with 12 individuals. On average they made 12 contacts (either face-to-face or by telephone) during the first month after discharge (Figure 17). Eleven of the 12 individuals had at least two follow-up contacts in the first week following discharge. The one individual who had no contact during the first week had three face-to-face contacts during the second week. When we stratified our sample to investigate whether or not the distribution of contact timings was affected by readmissions to hospital we found no evidence that such readmissions contributed to the observed rise in contacts in months 6 and 7.

FIGURE 17.

Mean number of face-to-face and telephone contacts by month over the 12 months following the index inpatient discharge (n = 12 patients).

Information from the qualitative interviews (see below) provides evidence on whether the participants felt that the intervention added anything to the perceived level of support provided by community mental health services.

During the course of the study, the mental health trust in one of the areas in which the pilot took place instituted a policy of two home visits to patients in the first week following hospital discharge.

Service users’ views of the intervention

Our qualitative interviews with 13 service users who received the letter, as well as the responses to the questionnaire survey, indicated that, although most patients could see the potential benefits of the letters and were positive about receiving them, they generally felt that they would be of more benefit to people who were new to the mental health system. Some patients did use the information in the leaflet, but this was unusual.

There was concern about the repetitive nature of the letters and that they might appear impersonal. Issues such as how the salutation on the letter was worded, for example first name only or first name and family name (as standard practice in the trusts), could make a difference, but it was clear that different approaches would appeal to different people. The sense of impersonality may have been compounded by the need for us to send an opt-out sheet and reply envelope with each letter. There is clearly a tension between the need for some degree of automation and the inevitable loss of personalisation.

Another indication of the (limited) acceptability of the intervention was the level of opt-out: 23.5% of patients opted out over the intervention period, some after just one or two letters.

Feasibility of a national study of the letter-based intervention

We were able to run the intervention on three different psychiatric inpatient wards across two different mental health trusts. We estimated that it took approximately 48 hours to establish the intervention on a ward, a maximum of 7 hours per week of NHS staff time to deliver the intervention and up to 1 hour of research assistant time per week to respond to issues arising from a ward.

No adverse events arising as a result of us sending the letters were reported, although, as reported in the qualitative interviews, one patient was upset by what he perceived to be the impersonal nature of the salutation used in the letters.

A key issue that became clear to us during the piloting was the high level of readmission (25.5%) in the 12 months following discharge – a figure in keeping with national data indicating that, between April 2004 and March 2005, 24.7% of people discharged from an inpatient psychiatric unit in England were readmitted within 12 months. 164 This raises issues concerning the potential repetitiveness of the cycle of letters in this patient group – if the letters were restarted after the second, third or any subsequent discharge, they may seem less personal. Furthermore, when we compared the overall readmission rates of two of the intervention wards with those for other non-intervention wards in the same hospital we saw no evidence of an intervention-related reduction. In keeping with this, the recent pilot study of a similar letter-/e-mail-based intervention in US military hospitals indicated that those in receipt of the intervention had higher levels of readmission than non-participants (13.6% vs. 7.4%). 166

Eight (10.0%, 95% CI 3.0% to 17.0%) of the participants discharged from the Bristol wards who attended an emergency department following an episode of self-harm after discharge were admitted to a hospital bed. These figures are in keeping with national data: between 1 April 2004 and 31 March 2005, 6.5% of people were admitted to a general hospital bed following a self-harm episode during the 12 months following inpatient psychiatric discharge. 164

The high levels of readmission (25.5%) and opt out (23.5%) meant that only half of those patients who received the first letter received them all. A further 24.7% of discharges were not eligible to receive the letters, mainly because they lived outside the wards' catchment areas. This means that the intervention might be acceptable and deliverable to less than half of all ward discharges.

During the course of the piloting in Bristol, the local mental health trust introduced a policy of visiting patients discharged from inpatient care twice in the week after discharge, thereby potentially diluting any effect of the letter-based intervention.

Based on our analysis of HES data164 indicating that 6% of patients are admitted to hospital after self-harm in the year after discharge from psychiatric inpatient care and an estimated 12% (n = 10,000) present to emergency departments with self-harm, the sample size required to detect a 15% reduction in episodes of non-fatal self-harm in the year after discharge would be about 9800 patients. The sample size would need to be larger (n = 20,800) if only inpatient admissions for self-harm were used as the outcome measure. With MHRN support, a UK trial of this intervention might be feasible, although challenging, to mount.

Feasibility of collecting data to identify the costs of the intervention (an economic evaluation) and outcome data

Collection of data relevant to an economic evaluation of the intervention and outcome data (readmissions/emergency department attendance for self-harm) was straightforward using routine sources.

Paracetamol

The 1998 UK legislation57 to restrict pack sizes of paracetamol appears to have had significant beneficial effects on both deaths and registrations for liver transplantation. In spite of these apparent benefits, there continues to be a considerable number of deaths each year from paracetamol poisoning, averaging 121 per year (for suicide, open and accidental verdicts) between 2000 and 2009 for paracetamol alone with or without alcohol, excluding compounds. The benefits should therefore not lead to complacency. Further measures may be required to limit this death toll. These might include stronger enforcement of the legislation, although we found few breaches of the sales guidance. Another possibility is for pack sizes to be reduced further. Lower pack sizes have been introduced in some other countries (e.g. Ireland77 and Germany44). Although the findings from our Anglo-Irish study did not suggest that smaller pack sizes of paracetamol in Ireland have resulted in smaller overdoses than in England, there may be several reasons for this. For example, in Ireland, more often than in England, it appears that the numbers of tablets taken are equivalent to multiples of packs. This raises the question of whether or not this practice reflects differences in patients' characteristics and access to care and greater ease of purchasing of multiple packs in Ireland than in England. The findings therefore do not provide definitive information on whether or not a further reduction in the maximum pack sizes for paracetamol sold over the counter in the UK would have further beneficial effects on overdose size (and hence on deaths from paracetamol overdose). The current pharmacy upper limit pack size of 32 tablets includes sufficient paracetamol that if all of it is ingested in an overdose the risk of hepatotoxicity is rather high. 54

Our patient interview study identified media influences (including the internet) in some cases. Media producers should be encouraged to comply with guidelines concerning detailed descriptions of suicide methods. 90 Internet influences on self-harm and suicide are increasingly recognised, including proliferation of sites giving details of drugs that may be used for fatal self-poisoning. 189,190 This requires careful monitoring and, when feasible, possible restriction of those that are considered dangerous.

Another possible measure to limit the dangers of paracetamol overdose would be to reduce the amount of paracetamol in each tablet (e.g. to 325 mg, as has recently been instigated by the US FDA for paracetamol in compound preparations191), especially as lower-dose and 500-mg tablet preparations have been shown to have equivalent pain-relieving efficacy. 192

Fortunately, although we have shown that prescribing of analgesics has risen since the 1998 legislation, increased prescribing of NSAIDs does not appear to have had deleterious effects in terms of GI disturbances, although greater prescribing of antiulcerants could have offset any negative effects on GI function. This study, however, illustrates the importance of widening the evaluation of measures aimed at limiting the availability of specific medications so that any deleterious effects that result from changes in the availability of alternative medications can be detected.

Co-proxamol

In 2003, a report highlighted that co-proxamol was used in 20% of poisoning suicides in England and Wales and that co-proxamol was far more toxic than TCAs in overdose. 101 Consequently, the MHRA reviewed its relative benefit/safety profile and in January 2005 recommended to the CSM that co-proxamol be withdrawn, with a 3-year initial partial withdrawal period. 106 We have shown that during both this period and the subsequent 3 years, as prescribing of co-proxamol declined and there was an increase in prescribing of some other analgesics, there was a net reduction in analgesic poisoning deaths. 79,175 By the end of the 6-year period following the withdrawal of co-proxamol, there had been an estimated 500 fewer suicides involving co-proxamol alone, and 600 when accidental deaths were included, with no evidence of significant substitution by death involving other analgesics. 175

Our findings were communicated to the MHRA both after the evaluation of the first 3 years following its decision and at the end of the 6 years following withdrawal of co-proxamol. They were also presented to the European Medicines Agency and influenced its decision to recommend to the European Union that prescribing of dextropropoxyphene (the more dangerous component of co-proxamol) cease. 114 Prescribing of dextropropoxyphene has now been banned in the USA and Canada.

The UK initiative appears to have been a very successful suicide prevention measure. Our findings are consistent with other evidence which indicates that reducing the availability of a common and particularly dangerous method of suicide can reduce the number of suicides by that method and does not necessarily lead to substitution by an increase in numbers of suicides by similar methods. 11 However, such initiatives require careful long-term evaluation, especially as new methods of suicide become more popular and available.

Antidepressants

Another way of restricting access to dangerous means of self-poisoning is through prescribing less toxic drugs to people at risk. Depression is the most common psychiatric disorder in people dying by suicide, with approximately 60% of those who die suffering from affective disorder. 118 Affective disorders are also common in patients who engage in non-fatal self-harm, with 72% having such disorders in one UK study. 193 People with depression who engage in suicidal acts often take antidepressants in overdose and therefore considerable attention has been paid to their toxicity, especially that of the TCAs. 45 It is well known that people often take their prescribed medication in overdose. 194 Choice of antidepressant in someone at risk should be determined by effectiveness, side effects and toxicity.

The usual approach to assessing fatal toxicity of a prescribed drug is to compare number of deaths involving that drug with the extent to which it is prescribed. 146 We have employed both this and another less used measure, case fatality, in which the death rate is compared with the rate of non-fatal overdoses,45,159 to assess the relative toxicity of all of the antidepressants in common current use. Importantly, we showed that the results obtained with the relative toxicity and case fatality approaches were practically identical.

Although we have confirmed the well-known higher toxicity of TCAs relative to SSRIs, we have highlighted important differences within these categories. Thus, among the TCAs, both dosulepin and doxepin were found to be considerably more toxic than amitriptyline. The elevated toxicity of dosulepin has previously been recognised,143 but our results, which we communicated to the MHRA, highlight the need for stronger advice to be given to clinicians, especially as dosulepin is still prescribed widely in some parts of the UK. 159 Although doxepin is less often prescribed than dosulepin, clinicians should be aware of its relatively high level of toxicity.

Selective serotonin reuptake inhibitors are well recognised as being safer than the TCAs, but we have shown that citalopram is approximately three times more toxic than the other SSRIs. It is important that clinicians are also aware of this finding, which is supported by other research,129,130 as it is again relevant to prescribing decisions, especially in patients at risk of self-poisoning.

Lastly, although our finding that the toxicity of venlafaxine is intermediate between that of the TCAs and that of the SSRIs has been recognised before and resulted in advice from the MHRA about restricting pack sizes and initial supplies for patients,128 the similar toxicity of mirtazapine is a new finding. This result has also been communicated to the MHRA.

Implications for practice and research

1. Small-area (primary care trusts/LAs) suicide rates and changes in these rates over time since 2000 should be interpreted with caution in those areas where coroners make high use of narrative verdicts.

2. Further increases in the use of narrative verdicts will compromise the quality of national suicide statistics.

3. Coroners could be required to provide both the short-form verdict and a longer narrative account of the death (when appropriate).

4. The ONS might consider including in its suicide statistics deaths from medicine poisoning given a verdict of accident/misadventure by coroners.

5. The Department of Health might consider carrying out surveillance to enable the early identification of increases in the use of high-lethality, easily accessible suicide methods, to enable a timely response.

6. Estimates of the numbers of suicides by co-proxamol, paracetamol and TCA poisoning would be more reliable if they included accidental poisonings from these drugs as well as deaths given suicide and open verdicts by coroners.

7. There should be an in-depth analysis of the proportion of suicides in which the internet may have played a contributory role, assessed alongside evidence of the beneficial effects on mental health and suicide risk.

Implications for clinical practice

1. Services should ensure optimal treatment for those who self-harm, in particular, prioritising the provision of psychosocial assessment as emphasised in national guidelines.

2. When prescribing antidepressants, clinicians should take account of the risk of overdose, especially in patients at risk of self-poisoning, as well as their relative efficacy, acceptability and possible interactions with other medication and alcohol, and patients' concurrent physical morbidity.

3. To prevent ongoing deaths involving paracetamol, further measures might be aimed at reducing breaches of sales guidelines and at encouraging media and internet site producers to follow guidelines on the reporting of suicide.

4. National, multicentre research would benefit from a simplified system of centralised approval for local research governance permissions.

5. Despite their low cost and apparent simplicity, contact-type interventions following psychiatric hospital discharge or self-harm cannot be recommended for widespread introduction.

Recommendations for future research

1. Variability in self-harm services should continue to be monitored to gain a greater understanding of aspects of treatment that are beneficial for preventing repeat self-harm.

2. Further work is needed to elucidate the active components of therapeutic contact following self-harm and to understand in which groups treatments might have the most impact.

3. Trends in the use of narrative verdicts and their impact on national and small-area suicide rates should be reassessed following recent ONS-led interventions to improve the accuracy of suicide reporting.

4. An assessment should be carried out of the feasibility and costs of developing a surveillance system to identify as quickly as possible rises in the use of novel methods of suicide to enable rapid interventions to restrict ease of availability to at-risk individuals.

5. The relative toxicity of other drugs commonly used for intentional self-poisoning should be evaluated to assist clinicians in making prescribing decisions and for informing regulatory agencies.

6. Future changes in availability of medications that are used for self-poisoning should be evaluated in terms of both impacts on self-harm and suicide and indirect consequences because of altered availability of other drugs.

7. The effect on the quality of services and patient outcomes of new guidance and future policies on management of self-harm (such as the November 2011 NICE guidelines139 on the longer-term management of self-harm) requires careful evaluation.

8. Assessment of the relative toxicity of antidepressants should continue as new antidepressants are marketed, and international comparisons are warranted in view of the differences in prescribing practices between countries.

Long-term evaluation

Dr Helen Bergen, Dr Sue Dodd, Dr Phil Pocock, Dr William Bernal and Claudia Wells.

Interview study

Fiona Brand, Sheryl Denman Taylor, Karen Lascelles, Jacklyn Patzsanda, John Ryall, Dr Kate Saunders and Linda Whitehead for their help in identifying patients for the study and conducting some of the interviews; Dorothy Rutherford for assistance in extracting data from the Oxford Monitoring System for Attempted Suicide; and Dr Helen Bergen and Deborah Casey for support and advice on analyses.

Anglo-Irish comparison

Dr Helen Bergen, Dr Ella Arensman, Dr Paul Corcoran and Keith Waters.

Non-steroidal anti-inflammatory drug prescribing and gastrointestinal consequences

Ly-Mee Yu, Omar Omar, Phillip Gichuru, Peter Stephens of IMS Health for organising the provision of prescription data and Dr Ben Wheeler for providing data on trends in hospital admissions for gastrointestinal haemorrhage in England.