Introduction

Cases involving deliberate self-harm, including intentional self-poisoning or self-injury, are common psychiatric situations at emergency departments (EDs) (Skegg 2005). Deliberate self-poisoning (DSP) is the deliberate ingestion of an overdose of a medication or substance to effect some change that the person desires through the expected consequences of the act (Hawton et al. 2007; Perry et al. 2012). DSP is the most common form of deliberate self-harm and is closely associated with suicide (Hawton et al. 2007; Perry et al. 2012). The risk of suicide in patients with a history of DSP is higher than that of the general population (Hawton et al. 2003). Therefore, identifying DSP patients and implementing adequate prevention strategies thereafter are critical for reducing the risk of suicide.

Previous studies of DSP have focused primarily on presentations, behaviors, and substance categories (Hawton et al. 2007; Perry et al. 2012; Watve et al. 2012; Prescott et al. 2009; Fliege et al. 2009; Zakiullah et al. 2008; Rhodes et al. 2008; Fathelrahman et al. 2008; Cook et al. 2008). Risk factors for DSP include a history of psychiatric illness, substance abuse, and socioeconomic disadvantage (Skegg 2005; Madsen et al. 2012). However, the relationship between DSP and prescription medication is not clear. Benzodiazepines (BZDs) are frequently prescribed by psychiatrists as anxiolytics (Prescott et al. 2009; Longo and Johnson 2000; Berger et al. 2012). Reports from one National Poison Control Center indicated that drugs with CNS effects were encountered the most frequently, and the majority of these drugs were benzodiazepines (Lin et al. 2003). BZDs are also frequently used for DSP in other countries (Cook et al. 2008; Cox et al. 2011). Therefore, we evaluated the relationship between BZD treatment in psychiatric outpatients and DSP cases at EDs using data from the Taiwan National Health Insurance Research Database (NHIRD).

Methods

Data source

A retrospective nested case–control study was conducted using data from the Taiwan NHIRD. The NHIRD is a nationwide database that was established in 1996. It contains reimbursement claims data from the Taiwan National Health Insurance system, which has provided coverage for approximately 99 % of the population since 1998. The National Health Research Institutes (NHRI) manages the annual claims data in the NHIRD and established the Longitudinal Health Insurance Database (LHID) for use in medical research. Demographic data, medications, treatments (including operations), and disease diagnoses (coded based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)) for patients attending health-care facilities due to health issues were recorded in the NHIRD. Health facilities that are enrolled in the Taiwan National Health Insurance (NHI) include local clinics, community hospitals, regional hospitals, and medical centers. The Taiwan NHI covers almost all primary, secondary, tertiary, and quaternary health-care facilities in Taiwan, with the exception of some local clinics. The LHID is composed of historical claims data for one million patients who are randomly selected from the NHIRD. The NHRI encrypts patients' personal information for the protection of privacy and provides researchers with anonymous identification numbers that connect to the relevant claim information, including the patient's sex, date of birth, registry of medical services, and medication prescriptions. Patient consent was not needed for the NHID or LHID. Our study was also approved by the Ethics Review Board of China Medical University (CMU-REC-101-012).

Study design

The selection process used in our study is demonstrated in the flow chart in Fig. 1. The psychiatric patient cohort included patients who were diagnosed with a psychiatric illness (ICD-9-CM 291-293, 295-298, 300-301, 303-305, 308, and 311) between 2000 and 2008. Subjects prescribed with BZD for at least 2 months during the study period were identified and defined as the BZD cohort. The initial BZD treatment date was defined as the reference date. Patients younger than 20 years old at the time of the psychiatric diagnosis were excluded from our study. Participants who subsequently presented to an ED and were diagnosed as DSP (ICD-9-CM 960-979 and E930-E949) before December 31, 2009 were included in the DSP study group, and the date of the DSP diagnosis was used as the index date. The control study group contained ten participants from the remaining members of the psychiatric patient cohort for each DSP group participant. Control participants were randomly frequency matched to DSP participants based on age category, sex, and the year of enrollment in the psychiatric patient cohort. The follow-up period for all participants ended on December 31, 2009.

Fig. 1
figure 1

Flow diagram for patients presenting with deliberate self-poisoning (DSP) who were enrolled in this study

Full size image

The following BZDs were selected for analysis: clonazepam (ATC code: N03AE01), diazepam (ATC code: N05BA01), oxazepam (ATC code: N05BA04), lorazepam (ATC code: N05BA06), bromazepam (ATC code: N05BA08), alprazolam (ATC code: N05BA12), nordazepam (ATC code: N05BA16), fludiazepam (ATC code: N05BA17), cloxazolam (ATC code: N05BA22), lorazepam combinations (ATC code: N05BA56), flurazepam (ATC code: N05CD01), nitrazepam (ATC code: N05CD02), flunitrazepam (ATC code: N05CD03), estazolam (ATC code: N05CD04), triazolam (ATC code: N05CD05), lormetazepam (ATC code: N05CD06), and midazolam (ATC code: N05CD08).

The average BZD prescribed dose was calculated as the total cumulative BZD dose in grams divided by the duration of BZD use (from prescription to the index date in years). The average benzodiazepine dose was separated into three groups by dose tertiles (33rd and 66th percentiles). Because of the potential confounding effects, the disease history was also collected for sleep disorders (ICD-9-CM 780.5 and 307.4), anxiety disorders (ICD-9-CM 300.0, 300.2, 300.3, 308.3, and 309.81), schizophrenia (ICD-9-CM 295), depression (ICD-9-CM 296.2-296.3, 300.4, and 311), and bipolar disorder (ICD-9-CM 296.4-296.6, 296.80).

Statistical analysis

The distribution of the study population based on the demographic characteristics and the disease history data were analyzed. To evaluate differences among the study groups, the chi-squared test was used for categorical variables, and Student’s t test was used for continuous variables. The odds ratio (OR) and 95 % confidence interval (95 % CI) were measured for each comparison to estimate the associations between BZD use and DSP using logistic regression. Adjusted odds ratios (AORs) were also determined after adjusting for potential confounders. To evaluate the dose response of the association between average BZD dose and DSP, the average BZD dose was treated as a continuous variable across the range of average doses to evaluate trends in DSP diagnosis using logistic regression. All data management and the statistical analyses were performed using SAS 9.1.3 software (SAS Institute, Cary, NC, USA). All statistical tests were two-sided, and associations with a p value less than 0.05 were considered statistically significant.

Results

A total of 184,509 patient files comprised the psychiatric patient cohort (Fig. 1). The DSP study group contained 629 participants, and the control group contained 6,290 participants.

The demographic characteristics of the study population are presented in Table 1. The DSP and control groups were similar in age (42.4 ± 17.1 vs. 42.3 ± 17.1 years, p = 0.9668) and had similar sex ratios (p = 1.0000). Approximately 83.5 % of the patients had previously received BZD treatment. Psychiatric patients with a DSP history were more likely to present with repeat episodes of DSP (6.4 vs. 0.5 %, p < 0.0001). Psychiatric patients who presented with DSP had a higher proportion of BZD use (83.5 vs. 53.1 %, p < 0.0001), sleep disorders (44.8 vs. 14.4 %, p < 0.0001), and anxiety disorders (39.6 vs. 14.2 %, p < 0.0001).

Table 1 The demographic characteristic and psychiatric disease history for DSP and control group during 2000–2009
Full size table

Psychiatric patients under BZD treatment and those with a history of DSP were significantly more associated with DSP diagnosis in EDs (OR = 4.46, 95 % CI = 3.59–5.53 and OR = 13.27, 95 % CI = 8.28–21.29, respectively, both p < 0.0001) compared with psychiatric patients who did not receive BZD treatment, and patients with other psychiatric diseases were more likely to be diagnosed with DSP. Sleep disorder (OR = 5.04, 95 % CI = 4.24–5.98, p < 0.0001), anxiety disorder (OR = 3.95, 95 % CI = 3.32–4.70, p < 0.0001), schizophrenia (OR = 7.80, 95 % CI = 5.28–11.52, p < 0.0001), depression (OR = 15.20, 95 % CI = 12.22–18.91, p < 0.0001), and bipolar disorder (OR = 18.48, 95 % CI = 10.13–33.7, p < 0.0001) diagnoses were significantly associated with the diagnosis of DSP at EDs.

After adjusting for age, sex, anxiety disorder, sleep disorder, depression, schizophrenia, and bipolar disorder as potential confounders, BZD use remained significantly associated with DSP (AOR = 2.47, 95 % CI = 1.93–3.17, p < 0.0001; Table 2). All average cumulative BZD dose ranges analyzed were significantly associated with DSP (<0.005 g/year, AOR = 1.75, 95 % CI = 1.28–2.39, p = 0.0004; 0.005–0.04 g/year, AOR = 2.57, 95 % CI = 1.93–3.42, p < 0.0001; and >0.04 g/year, AOR = 4.30, 95 % CI = 3.17–5.83, p < 0.0001). Regression analysis also showed a significant dose–response association between average BZD dose and DSP diagnosis (p < 0.0001).

Table 2 Effects of BZDs and dose response on psychiatric patients receiving BZD prescriptions
Full size table

The relationship between duration of BZD use before DSP events and DSP was analyzed; regardless of the duration of BZD treatment (i.e., within 1 week, 1 month, or 1 year prior to the index date of the DSP event at EDs), patients receiving BZD treatment were significantly associated with DSP events at EDs (Table 3). Patients with a BZD treatment duration of only 1 week were more likely to present with DSP than those treated for 1 month, 3 months, or 1 year (<1 week, AOR = 5.67, 95 % CI = 3.90–8.25; <1 month, AOR = 5.47, 95 % CI = 4.25–7.04; <3 months, AOR = 4.66, 95 % CI = 3.71–5.85; <6 months, AOR = 4.23, 95 % CI = 3.40–5.26; <1 year, AOR = 3.72, 95 % CI = 3.01–4.59, all p < 0.0001).

Table 3 Effects of BZDs on psychiatric patients receiving BZD prescriptions, by various periods of BZD use
Full size table

Limitation

The limitations of our study include the limited nature of the clinical data that were collected from the NHIRD, the difficulty associated with controlling for confounding factors because of the retrospective study design, and the incomplete verification of data in the NHIRD. The NHIRD does not provide detailed information regarding smoking habits, disease severity markers, alcohol consumption, body mass index, physical activity, socioeconomic status, or family history, which all represent possible confounding factors for our analysis. The registries in the NHI claims system were primarily designed for administrative billing, and the registry data are not subjected to the stringent levels of verification that are appropriate for many types of scientific studies. We were unable to contact the patients directly to obtain more information on their use of BZDs because the participants remained anonymous. Furthermore, dose equivalence weighting and disease severity analysis were difficult to apply due to the lack of laboratory data and detailed clinical information. The relationship between substance abuse, DSP, and BZD prescriptions was also not established because the substance abuse information database was not connected to the NHIRD. However, the data from the NHIRD regarding BZD prescriptions, the diagnosed psychiatric illness, and the coding of DSP patients visiting EDs were highly reliable, and the results of our study indicate that further studies, such as population-based, unbiased, randomized, observational trials, are warranted to confirm the causal relationships between DSP and BZD use.

Discussion

Prescribing BZDs for sleep and anxiety disorders is a common psychiatric practice because of their relative safety and the rare occurrence of death following overdose (Longo and Johnson 2000). There is little doubt of the therapeutic efficacy of BZDs in reducing anxiety, inducing sleep, and quelling panic symptoms (Longo and Johnson 2000; Bandelow et al. 2012). The anxiolytic and hypnotic efficacy of BZDs has been well established by numerous studies. Compared with antidepressants, which have a longer onset of action but are the best agents for the long-term treatment of anxiety disorders, and anticonvulsants and antipsychotics, which are effective but have an intermediate onset of action, patients with these disorders often prefer BZDs because these agents have immediate effects to treat fluctuating conditions if patients use BZDs judiciously (Longo and Johnson 2000; American Psychiatric Association 1990). However, the greatest asset is also their greatest liability: drugs that work immediately tend to be addictive. It is difficult to evaluate and optimally prescribed dosage for anxiety and insomnia patients taking BZDs during their routine follow-up because most patients typically ask for more drugs, not fewer (Prescott et al. 2009; Sansone and Sansone 2012).

The relationships among BZD therapy, DSP, and suicide have been reviewed (Neutel and Patten 1997; Neale and Smith 2007; Centers for Disease Control and Prevention 2012). Previous studies have suggested that BZD use may be associated with DSP events and suicide attempts (Neutel and Patten 1997; Neale and Smith 2007), particularly in elderly people (Carlsten et al. 2003; Ticehurst et al. 2002). In our study, the cumulative BZD doses were significantly associated with increased DSP events at EDs, and the effects of BZD usage were significant if the prescription duration was greater than 1 month. Patients with underlying psychiatric illnesses, such as schizophrenia, depression, and bipolar disorder, and patients who just started to receive BZD treatment had an even higher risk of DSP. After we adjusted for age, sex, anxiety, sleep disorder, depression and schizophrenia, and DSP history, the adjusted OR for psychiatric patients visiting the ED due to DSP in patients who took BZDs remained 2.5-fold higher than that in patients who did not take BZDs. This finding suggests that psychiatrists typically prescribe or that patients typically take higher doses of BZDs to relieve psychiatric symptoms such as anxiety and poor sleep quality, which are the most common symptoms for patients with unstable psychiatric illness (Li et al. 2010; Baylé et al. 2011). Patients taking BZDs have been observed to develop tolerance, dependence, and dose escalation, and studies have shown that both patients and psychiatrists are likely to choose higher dosages to relieve persistent symptoms (Prescott et al. 2009; Sansone and Sansone 2012); therefore, patients who do not respond favorably despite a high therapeutic BZD dosage may choose to take a much higher dose of BZDs, resulting in DSP events. Thus, psychiatrists should be aware of this phenomenon and focus more on patients' unstable emotions and underlying psychiatric illnesses to provide other psychological or social assistances, prescribe antidepressants or antipsychotics that have stable and prolonged effects, and treat potential addiction behaviors, rather than prescribing BZDs, which often only have immediate and short-term effects (Baylé et al. 2011).

As in other countries, DSP events at EDs in Taiwan are regarded as an important warning sign for suicide behaviors that require action. Previous studies have established a relationship between DSP and suicide behaviors and have suggested that aggressive actions should be taken to follow up after DSP events to prevent suicide (Skegg 2005; Hawton et al. 2003). In Taiwan, psychiatric patients with a high-risk status and those that have previously attempted DSP have been required to report to a suicide prevention center since 2006 to undergo aggressive suicide prevention strategies (National Taiwan Suicide Prevention Center 2006; Department of Health 2011).

In conclusion, BZDs prescribed to psychiatric outpatients were associated with the incidence of DSP at EDs. Furthermore, DSP events at EDs were significantly associated with BZD doses, and the effects of BZD usage were significant if the prescription duration was greater than 1 month. Our results indicate that vigilant evaluation of the psychiatric status of patients receiving BZD therapy is critical for the prevention of DSP events.