Psychosocial assessment and repetition of self-harm: The significance of single and multiple repeat episode analyses

Abstract

Background

Self-harm is a common reason for presentation to the Emergency Department. An important question is whether psychosocial assessment reduces risk of repeated self-harm. Repetition has been investigated with survival analysis using various models, though many are not appropriate for recurrent events.

Methods

Survival analysis was used to investigate associations between psychosocial assessment following an episode of self-harm and subsequent repetition, including (i) one repeat, and (ii) recurrent repetition (≤ 5 repeats) using (a) an independent episodes model, and (b) a stratified episodes model based on a conditional risk set. Data were from the Multicentre Study on Self-harm in England, 2000 to 2007.

Results

Psychosocial assessment following an index episode of self-harm was associated with a 51% (95% CI 42%–58%) decreased risk of a repeat episode in persons with no psychiatric treatment history, and 26% (95% CI 8%–34%) decreased risk in those with a treatment history. For recurrent repetition, assessment was associated with a 57% (95% CI 51%–63%) decreased risk of repetition assuming independent episodes, and 13% (95% CI 1%–24%) decreased risk accounting for ordering and correlation of episodes by the same person (stratified episodes model). All models controlled for age, gender, method, history of self-harm, and centre differences.

Limitations

Some missing data on psychiatric treatment for non-assessed patients.

Conclusions

Psychosocial assessment appeared to be beneficial in reducing the risk of repetition, especially in the short-term. Findings for recurrent repetition were highly dependent on model assumptions. Analyses should fully account for ordering and correlation of episodes by the same person.

Introduction

Repetition of non-fatal self-harm is common. Approximately 15%–25% of people who self-harm will repeat an episode within one year, and 20–25% over the next few years (Owens et al., 2002, Platt et al., 1988). Subsequent risk of suicide is elevated in people who self-harm relative to the general population (Cooper et al., 2005, Hawton et al., 2003b, Owens et al., 2002), and greater for those with repeated self-harm relative to single-episode self-harm (Haw et al., 2007, Zahl & Hawton, 2004). The costs in terms of public health and individual well-being are great. Psychosocial assessment following self-harm, as outlined in the National Institute for Clinical Excellence (NICE) guidance for England and Wales (National Collaborating Centre for Mental Health, 2004), is a necessary starting point for preventive interventions (e.g. Guthrie et al., 2003, Hawton et al., 1998). However, despite the advice that all patients presenting to the Emergency Department (ED) with self-harm receive a specialist psychosocial assessment (National Collaborating Centre for Mental Health, 2004), many do not (Hughes & Kosky, 2007, Jones & Avies-Jones, 2007, Kapur et al., 2008), especially those who self-harm repeatedly (Hickey et al., 2001).

Most studies have focused on risk factors for repetition (e.g. Chen et al., 2010, Reith et al., 2003). Some studies suggest that repetition rates are lower for people who are assessed; however, these were based on small samples (Crawford and Wessely, 1999), used follow-up to one repeat only (Johnston et al., 2006, Kapur et al., 2006), or assumed that all repeat episodes were equivalent (Kapur et al, 2008). There is little information of other aspects of repetition such as timing between episodes. No studies to our knowledge have investigated differential risk of repetition by episode number. This needs further investigation given the clinical importance of the question of whether psychosocial assessment can reduce the repetition of self-harm.

Repetition of self-harm may be considered as a series of naturally ordered, multiple-failure events (Wei and Glidden, 1997), also called recurrent events if the repeat events are of the same type (Kelly and Lim, 2000). Analysis of repetition using a dataset with information on time to occurrence of self-harm episodes may be done in a number of ways depending on the specific research question (Cleves, 1999, Lim et al., 2007, Wei & Glidden, 1997). In a simple person-based analysis, the first episode for each person in a study period may be taken as the index episode, and the time to repeat episode measured using survival methods such as Cox proportional hazards regression. This approach may be appropriate for investigation of the short-term effect of a clinical intervention at the index episode. However, this method utilises only a small proportion of all available data, and for this reason many investigators (e.g. Kapur et al., 2008, Lilley et al., 2008, Reith et al., 2003) have preferred to include subsequent repeat episodes in the study period, rather than just the first.

This second type of analysis in which all repeat episodes are included often follows the Andersen and Gill (AG) model (Andersen and Gill, 1982), which relies on the assumption that all episodes are independent and equal, i.e. that the risk of a repeat episode is unaffected by the patient's earlier episodes (Wei et al, 1997). Elsewhere we have argued that this approach is appropriate for investigations conducted from the perspective of clinical services where all patients are to be offered an assessment regardless of their history (Kapur et al, 2008). It may also be appropriate where the frequency of repeat episodes is very large and hence the risk of repetition tends to vary less between repeat episodes (Lim et al., 2007). However, this approach assumes that the whole sample is at risk for each episode. It ignores the fact that repeat episodes of self-harm are ordered events, and that the third event cannot occur before the second event, for instance. It also ignores the correlation of failure times (time between repeat episodes) for each person, and violates the requirement in Cox regression that failure times are independent.

A third type of analysis which addresses these issues (Kelly & Lim, 2000, Lim et al., 2007) is the conditional risk set approach of Prentice et al. (1981). In this approach the sample of people at risk of a kth episode is restricted to those who have had a k − 1th episode. The model takes account of the ordering of events and correlation of failure times by stratifying by the number of episodes, thus using a different baseline hazard for each episode. This latter method is appropriate for investigation of longer-term associations between various clinical factors and recurrent repetition (Wei et al, 1997).

Other analytical approaches such as multinomial regression (e.g. Haw et al. (2007)) and hurdle models (Bethell et al., 2010) are not considered here.

We used the three statistical approaches described above to explore associations between clinical factors and repetition of self-harm, and to highlight the importance of choosing the appropriate model for the research question of interest. Specifically, we investigated the relationship between psychosocial assessment following an episode of self-harm and subsequent repetition of self-harm, (i) in the short-term, and (ii) recurrent repetition in the longer term using two different methods of statistical analysis. We posed the following questions of clinical relevance:

• Short-term repetition: does psychosocial assessment at the first episode of self-harm in the study period decrease the risk of a repeat episode, taking into account the patient's method of self-harm, psychiatric treatment history, and history of self-harm?

• Recurrent repetition: is survival time between each episode in the first six episodes related to psychosocial assessment at the previous episode, taking into account the patient's method of self-harm, psychiatric treatment history, and history of self-harm, when (a) all episodes are assumed to be independent (referred to as the independent episodes model), and (b) fully accounting for the ordering and correlation of episodes (referred to as the stratified episodes model).