Recent self-harm and psychological measures in the emergency department

Sample

Patient enrolment occurred within the EDs of the Royal Alexandra Hospital and the University of Alberta Hospital in Edmonton, Canada. Study enrolment began in August 2009 and was completed in May 2010. These sites are the two largest EDs and teaching hospitals in the Edmonton region and are staffed by full-time ED physicians. Both EDs teach rotating interns and residents and have fulltime day ED psychiatric staff and inpatient mental health services.

Patients presenting to these EDs with self-harm or ideation were enrolled while they were being assessed in the ED. Only permanent residents of Alberta were enrolled (to increase follow up rates for the prospective portion of the study) and enrolment was limited to the adult population (age >17). We excluded patients who were violent, did not have the capacity to provide informed consent to the study (e.g., dementia, cognitive impairment), were suffering from an acute medical condition that would prevent them from participating (e.g., intentional drug overdose, coma, etc.), and/or those who were unable to understand or communicate in English. Eligibility for the study was determined by the most responsible physician; normally this was an ED physician, a psychiatrist or a psychiatry resident.

Assessments

Patients who agreed to study enrolment were administered a verbal interview questionnaire that collected relevant demographic (e.g., age, gender, education level, living situation) and medical (e.g., history of psychiatric disorders and self-harm) information. The patients then completed several written questionnaires.

The Brief Symptom Inventory (BSI; Derogatis & Melisaratos, 1983) is a 53-item questionnaire that assesses a variety of psychiatric symptoms present in the past week using a five point rating scale. Subscales of this questionnaire measure anxiety, depression, hostility, interpersonal sensitivity, obsessive compulsivity, paranoid ideation, phobic anxiety, psychoticism and somatisation. The Barratt Impulsiveness Scale (BI; Patton, Stanford & Barratt, 1995) is a 30-item questionnaire designed to measure a person’s level of impulsiveness using a four point rating scale. The questionnaire provides several subscales that attempt to measure specific aspects of impulsiveness including attentional, attentional impulsiveness, cognitive complexity, cognitive instability, motor, motor impulsiveness, non-planning impulsiveness, perseverance and self-control. The Beck Hopelessness Scale (BHS; Beck et al., 1974) is a 20-item true or false questionnaire that attempts to measure the level of hopelessness that a patient has about their future. The CAGE (Ewing, 1984) questionnaire is a 4-question alcohol abuse screen with a cut off score of 3 or greater while the Drug Abuse Screening Test-10 (DAST; Skinner, 1982) is a 10-item questionnaire that measures the use of drugs. A Manchester Self-harm rule (Cooper et al., 2006) risk assessment was calculated by the researcher at the time of the interview based on information gathered from the patient and medical staff. The Manchester Self-harm rule contains four screening questions where a positive response to any question entails a positive result.

Analysis

Failure to answer one or more of the questionnaire items was adjusted for during analysis. The method of scoring for the BSI adjusted for missing answers by averaging the scores of the responses (Derogatis, 2006). The BHS and BI do not have explicitly stated methods to deal with missing values; however, the same method used for the BSI was applied to questionnaires that had 90% or more of the questions answered. For the BI this method was not possible due to the scoring method of the measurement so it was decided that mean replacement would be used for missing questions. As with the BHS questionnaire, those BI questionnaires that were less than 90% completed were counted as missing. The pattern of missing values was analyzed for the questionnaires and there were no indication that the missing values would affect the results. Continuous variables were assessed for suitability for analysis as continuous variables and data which were not suitable for analysis as a continuous variable were grouped.

Patients were grouped by two dichotomous variables: presentation with recent self-harm, and history of self-harm. For the presentation variable, it was determined whether the patient presented with recent self-harm (self-harm occurring shortly before ED presentation and requiring treatment and/or instigating their presentation), or the patient presented without recent self-harm. Secondly, for the history of self-harm variable, it was determined whether the patient had a history of prior self-harm, not including self-harm treated during current presentation, or no prior history. First, ANOVA or regression analysis was performed using a categorical variable with all 4 possible subgroup combinations of the two variables of interest (history of self-harm, presentation with recent self-harm). ANOVA was used to analyze the questionnaire data. Logistic regression analysis was used for binary variables (e.g., history of mood disorder) and multinomial logistic regression was used for categorical variables (e.g., age category). If a significant p-value was obtained for the overall model, ANOVA or logistic regression was used to determine which of the two variables were individually significant.

Post-hoc analysis

A measure of distress was created using the questionnaire results that were significantly different between the self-harm and ideation groups. This measure was derived by combining the results of the significant questionnaire results with each scale having equal weight. This measure of distress was plotted against the length of time that the patients had been in the ED before they were assessed by the research team. A lowess smoother (a scatterplot with a line used to estimate the data trend) was used to determine the trend for the self-harm and ideation groups. Multiple imputation was used to estimate the missing values for the distress variable and then regression analysis was performed to assess whether demographic or medical history variables collected were influencing the relationship between a patients distress score and their presentation status (i.e., self-harm versus ideation only). The demographic variables entered into the regression analysis were education level, age, marital states, living arrangements (e.g., living alone, living with family etc.), and gender. The medical history variables were lifetime history of self-harm as well as a history of mood disorders, personality disorders, psychotic disorders, bipolar disorder and another disorder variable containing any previous diagnosis that did not fit into one of the previously mentioned diagnostic groups. The purpose of these analyses was to assess possible explanations of the results of the regression analysis.

Ethics

This study received approval from the Human Research Ethics Board of the University of Alberta (HERO #: Pro00005188). Informed written consent was obtained from each participant.

Sample

A total of 181 patients agreed to participate in the study, while 89 patients refused to participate. Complete information was obtained for a total of 157 patients. Participants in the study did not differ from those who refused based on gender, age, occurrence of self-harm at presentation, psychiatry consultation, and disposition (population characteristics described in Table 1). There was no relationship between failure to complete the questionnaires and the demographic characteristics measured.

Questionnaire results

Those who presented with recent self-harm were compared to those without recent self-harm to determine if any significant differences existed between those two groups. Those with recent self-harm scored significantly lower on several questionnaire scores compared to those without recent self-harm (Table 2). For the BSI, the Global Severity Index (GSI; p = 0.020), Positive Symptom Distress Index (PSDI; p = 0.023), Positive Symptom Total (PST) (p = 0.026), Anxiety (p = 0.020), Depression (p = 0.014), Hostility (p = 0.003), Interpersonal Sensitivity (p = 0.036), Obsessive Compulsive (p = 0.005) and Phobic Anxiety (p = 0.032) were significantly different between these two groups (self-harm patients endorsed less symptoms than those without self-harm). The BI found the Attentional Impulsiveness (p = 0.026), Cognitive Instability (p = 0.015), were significantly lower in those presenting with recent self-harm. The two groups also differed in their Beck Hopelessness Scale scores (p = 0.012). The DAST-10 scale and CAGE questionnaire were not significant.

The questionnaire scores of those with a history of previous self-harm were also compared to those without a history in Table 3. The mean differences were examined and many of the scales were found to have significant differences. For the BSI the Global Severity Index (GSI; p < 0.001), Positive Symptom Total (PST; p = 0.003), Anxiety (p = 0.006), Depression (p < 0.001), Hostility (p < 0.001), Interpersonal Sensitivity (p = 0.006), Obsessive Compulsive (p = 0.037), Paranoid Ideation (p = 0.003), Phobic Anxiety (p < 0.001), Psychoticism (p < 0.001) and Somatization (p = 0.024) were significantly higher in the group with a history of self-harm. Analysis of the BI found that only the cognitive complexity subscale failed to achieve statistical significance. All of the remaining subscales were significantly higher for the group with a history of self-harm. Of the remaining three questionnaires, the DAST-10, Beck Hopelessness Scale and CAGE questionnaire, no significant differences were identified.

Post-hoc analysis

The post-hoc analysis revealed a decrease in distress over the first hours in the ED for those that had engaged in self-harm (Fig. 1). However this trend was not significant. There was no apparent trend in the level of distress for patients that presented with ideation only. The regression analysis found no evidence that the result is due to confounding from extraneous variables. Entering the other potentially relevant variables not only failed to decrease the strength of the noted association between the questionnaires and recent self-harm but instead increased the strength of the relationship slightly. A similar pattern occurred when the individual scales were regressed instead of the aggregate measure of distress.

Limitations

As this is a cross sectional study, caution should be taken with attempts to determine any causal patterns from the data provided as directionality of any established relationship cannot be ascertained with a high degree of certainty. Due to the large number of comparisons there is a risk of Type I error for the differences detected in this paper. For differences such as those detected in the questionnaire mean scores, however, the chance that these scores are spurious is reduced by the high agreement between multiple questionnaires and scales.

This study also represents a convenience sample of patients. Attempts were made to determine if these patients were different than those that were not included in the study and these differences appear to be minimal; however, it is still possible that the enrolled patients may differ from those that refused to join the study or missed patients in significant ways. This study was also not capable of enrolling patients who presented with high risk self-harm attempts as these patients were often admitted to the ICU or were otherwise ineligible for study participation. It is therefore uncertain if those who present with high risk self-harm incidents were different from those who were described in this study. Another noteworthy aspect of this study was that our measurement of self-harm did not distinguish between suicidal and non-suicidal self-injury. Additionally, there were some missing values in the questionnaires results but analysis of the results did not show any apparent bias that would result from them.