Self-harm, with or without suicidal intent, substantially increases the risk of future suicide Reference Brådvik, Mattisson, Bogren and Nettelbladt1,Reference Crandall, Fullerton-Gleason, Aguero and LaValley2 and is known to be the strongest predictor of completed suicide. Reference Haukka, Suominen, Partonen and Lönnqvist3–Reference Owens, Horrocks and House7 Furthermore, repetition of self-harm is common: Reference Zahl and Hawton8,Reference Gunnell, Bennewith, Peters, Stocks and Sharp9 16% of patients who self-harmed were found to repeat a similar episode within 1 year. Reference Owens, Horrocks and House7 Repetition of self-harm increases the risk of completed suicide. Reference Zahl and Hawton8 One study found overdose to be the most prevalent type of suicide attempt that required admission, Reference Runeson, Tidemalm, Dahlin, Lichtenstein and Långström10 and approximately 80% of self-harm episodes have been reported to involve overdose. Reference Horrocks, Price, House and Owens11 It is therefore necessary to prevent the repetition of self-harm by drug overdose. According to clinical guidelines on the management of self-harm published in 2004, it is recommended that every patient presenting to hospital with self-harm should undergo a psychosocial assessment by specialists before being discharged. 12,13 Despite this recommendation, some studies have found that many patients, especially those with repeated self-harm, Reference Hickey, Hawton, Fagg and Weitzel14 did not actually receive such assessments. Reference Hughes and Kosky15–Reference Okumura, Shimizu, Ishikawa, Matsuda, Fushimi and Ito18 That would suggest that the guideline has not been widely used – possibly because it was not based on firm evidence. There is a lack of data on the influence of psychosocial assessments on preventing repetition of self-harm. Some studies have suggested that such assessments do have an influence, but they were based on a small sample size Reference Hickey, Hawton, Fagg and Weitzel14,Reference Kapur, House, Dodgson, May and Creed19–Reference Kapur, Cooper, Hiroeh, May, Appleby and House21 or on a small number of highly advanced institutions. Reference Kapur, Murphy, Cooper, Bergen, Hawton and Simkin17,Reference Bergen, Hawton, Waters, Cooper and Kapur22,Reference Kapur, Steeg, Webb, Haigh, Bergen and Hawton23 The present study focused on patients with drug overdose who were admitted to emergency centres. Using a national in-patient database in Japan, it aimed to investigate whether psychiatric intervention before discharge was associated with reduced patient readmissions with drug overdose.
The Diagnosis Procedure Combination (DPC) database is a national in-patient database in Japan that includes administrative claims data and detailed patient data. Reference Yasunaga, Horiguchi, Kuwabara, Matsuda, Fushimi and Hashimoto24 As of 2012, the database included the data of approximately 7 million in-patients from more than 1000 hospitals in Japan, representing around 50% of all acute care in-patient admissions. The database consists of the following information: unique hospital identifiers; age and gender main diagnoses, comorbidities present on admission, and complications that occurred after admission recorded with text data in Japanese and using ICD-10 25 codes; procedures; and discharge status. The responsible physicians are obliged to record the diagnoses with reference to medical charts on discharge. For the main diagnosis, the physicians have to enter only one ICD code.
This study was approved by the Ethical Committee, Faculty of Medicine, The University of Tokyo (approval No. 3501). Because of the anonymous nature of the data, informed consent was not required.
Participant selection and data
We identified patients with a diagnosis of drug poisoning (ICD-10 codes: T360–T509) who were discharged from participating hospitals between 1 July 2010 and 31 March 2013 (33 months in total). We included patients aged 12 years or older with a first episode of drug overdose and who visited a hospital with at least one full-time psychiatrist during the study period. We excluded patients who died during admission to hospital.
We identified psychiatric intervention by means of procedure codes for ‘interview for assessment and/or psychotherapy by a psychiatrist’, coded using the Japanese Procedure Codes defined under the fee schedule of the national health insurance system. We divided the patients into two groups: (a) those who received a psychiatric intervention – the psychiatric intervention group; and (b) those who did not – the unexposed group.
Based on the protocol of Quan et al, we converted ICD-10 codes of comorbidities that were present on admission into scores for each patient to calculate the Charlson comorbidity index (CCI). Reference Quan, Sundararajan, Halfon, Fong, Burnand and Luthi26 This index is used to predict mortality by classifying or weighting comorbidities to assess disease burden and case mix. Hospital volume was defined as the number of patients with the diagnosis of drug overdose annually at each hospital; it was classified into three categories (low, medium and high volume), with approximately equal numbers of patients in each group.
The primary outcome was readmission to the same hospital due to repeated drug overdose.
We conducted one-to-one matching between the psychiatric intervention group and unexposed group based on the estimated propensity score of each patient. Reference D'Agostino27 This approach avoided treatment selection bias, which is inherent in observational data analysis. In this approach, every patient in the intervention group was matched with a patient in the unexposed group based on the estimated propensity score, the probability of undergoing the intervention calculated using the observed potential confounders. The matched patients constituted two groups with similar characteristics, which resembled a randomised experiment-like situation. To estimate the propensity score, we fitted a logistic regression model with receipt of the psychiatric intervention as the outcome variable and the following as independent variables: age; gender; ICD-based information on toxic agents (non-opioid analgesics, anti-pyretics and anti-rheumatics [T39]; anti-epileptic, sedative, hypnotic and anti-Parkinsonian drugs [T42]; other psychotropic drugs [T43]; other drugs [T36–38, T41, T44–T49]; and unspecified drugs [T50]); ICD-based diagnoses of mental disorders (schizophrenia [F2]; mood disorders [F3]; organic mental disorders [F0]; mental disorders due to psychoactive substance use [F1]; disorders of personality and behaviour [F6]; and other mental disorder); Japan Coma Scale Reference Ohta, Kikuchi, Hashi and Kudo28 on admission; CCI; tracheal intubation; haemodialysis; type of hospital (academic or non-academic); hospital volume category and fiscal year of discharge. The C-statistic was calculated to evaluate the goodness of fit.
We conducted one-to-one matching between the psychiatric intervention and unexposed groups using nearest-neighbour matching within a calliper. One unexposed patient with the closest propensity score was selected for each intervention patient – provided that the difference in propensity score was within a certain amount (a calliper). We set a calliper as 0.20 of the standard deviation of the estimated propensity scores to achieve good balance between the intervention and unexposed groups. We used standardised differences to compare the prevalence of characteristics between the two groups. Reference Austin29 An absolute standardised difference of >10 has been suggested as signifying meaningful imbalance. Reference Austin29 We performed a chi-squared test to compare the proportions of readmission between the psychiatric intervention and unexposed groups among the propensity-matched patients. Logistic regression analysis for readmission was performed to calculate the odds ratio and 95% confidence interval (CI) of the psychiatric intervention group with respect to the unexposed group. We performed subgroup analysis on propensity-matched patients by age group. The threshold for significance was P<0.05. All statistical analyses were conducted using IBM SPSS Statistics, version 22.0 (IBM SPSS, Armonk, New York, USA).
We identified 29 564 eligible patients from 368 hospitals during the study period; they comprised the psychiatric intervention group (n=13 035; 44.1%) and unexposed group (n=16 529; 55.9%). In total, 1961 patients (6.6%) required repeated admission due to drug overdose. Using one-to-one propensity score matching, we obtained 7938 pairs of the psychiatric intervention and unexposed groups. The C-statistic for goodness of fit was 0.768. Table 1 shows the demographic characteristics of all patients (n=29 564) and the propensity score-matched patients (n=15 876). Patients in the psychiatric intervention group were more likely to have the following characteristics: be younger and female; have schizophrenia, mood disorder, or personality or behaviour disorders; take psychotropic drugs during their overdose episode; have a worse consciousness level; undergo tracheal intubation and haemodialysis; and be discharged after 2012. Academic hospitals and higher volume hospitals were more likely to perform psychiatric interventions. After propensity score matching, the patient distributions were closely balanced between the two groups.
|All patients||Propensity-matched patients|
|Unexposed group (n=16 529)||Psychiatric intervention group (n=13 035)||Absolute standardised difference||Unexposed group (n=7938)||Psychiatric intervention group (n=7938)||Absolute standardised difference|
|Non-opioid analgesics, anti-pyretics and anti-rheumatics||651||3.9||730||5.6||8.0||383||4.8||365||4.6||0.9|
|Anti-epileptic, sedative-hypnotic and anti-Parkinsonism drugs||4685||28.3||4109||31.5||7.0||2423||30.5||2469||31.1||1.3|
|Other psychotropic drugs||1172||7.1||1186||9.1||7.3||701||8.8||697||8.8||0.0|
|Classification of mental disorder|
|Organic mental disorders||260||1.6||178||1.4||1.6||114||1.4||127||1.6||1.6|
|Mental disorders due to psychoactive substance use||451||2.7||351||2.7||0.0||260||3.3||271||3.4||0.6|
|Disorders of personality and behaviour||261||1.6||461||3.5||12.1||210||2.6||217||2.7||0.6|
|Other mental disorder||1190||7.2||2042||15.7||26.9||1012||12.7||990||12.5||0.6|
|Level of consciousness on admission|
|Charlson comorbidity index|
|0||12 251||74.1||11 200||85.9||29.8||6775||85.3||6640||83.6||4.7|
|Hospital volume groups, per year|
|Fiscal year of discharge|
Table 2 shows the proportion of readmissions due to drug overdose in each subgroup in the propensity score-matched groups. In the matched patients, 1304 patients (8.2%) required repeated admission due to drug overdose. Patients who were younger females, had personality disorders and took other psychotropic drugs during their overdose episode were more likely to be readmitted as a result of drug overdose. Patients who were admitted to higher volume hospitals or discharged before 2011 were also more likely to be readmitted.
|No. of patients||Readmission||%|
|Non-opioid analgesics, anti-pyretics and anti-rheumatics||748||35||4.7|
|Anti-epileptic, sedative-hypnotic and anti-Parkinsonism drugs||4892||395||8.1|
|Other psychotropic drugs||1398||121||8.7|
|Classification of mental disorder|
|Organic mental disorders||241||10||4.1|
|Mental disorders due to psychoactive substance use||531||28||5.3|
|Disorders of personality and behaviour||427||57||13.3|
|Other mental disorders||2002||154||7.7|
|Level of consciousness on admission|
|Charlson comorbidity index|
|Hospital volume groups, per year|
|Fiscal year of discharge|
The propensity score-matched analysis showed a significant difference in readmission due to drug overdose between the psychiatric intervention and unexposed groups (7.3% v. 9.1% respectively; P<0.001). Logistic regression analysis showed that the psychiatric intervention group had a significantly lower proportion of readmission through drug overdose than the unexposed group (adjusted odds ratio 0.79; 95% CI 0.71–0.89; P<0.001; Table 3). Subgroup analysis showed that psychiatric intervention was significantly associated with lower readmission in two age groups – patients in their 20s and 40s (Table 4).
|Unexposed group (n=7938)||Psychiatric intervention group (n=7938)|
|n||%||n||%||P||Odds ratio (95% CI)|
|Unexposed group||Intervention group|
|Age, years||No. of patients||Readmission||%||No. of patients||Readmission||%||P||Odds ratio (95% confidence interval)|
Using a national in-patient database in Japan, this study compared repeated admission due to drug overdose between a psychiatric intervention and an unexposed group. Only 44% of the admitted patients underwent psychiatric intervention. A propensity-matched analysis demonstrated that the proportion of repeated admission through drug overdose was lower in the psychiatric intervention than in the unexposed group. Although not statistically significant for some age groups, the results were consistent across the various age groups. Psychiatric intervention was associated with lower readmission in younger subgroups, which is consistent with the findings of other reports. Reference Kawanishi, Aruga, Ishizuka, Yonemoto, Otsuka and Kamijo30,Reference Erlangsen, Lind, Stuart, Qin, Stenager and Larsen31
Several studies have investigated the effect of psychiatric intervention before discharge on preventing repetition of self-harm. However, the generalisability of those reports was limited because they were based on small sample sizes Reference Hickey, Hawton, Fagg and Weitzel14,Reference Kapur, House, Dodgson, May and Creed19–Reference Kapur, Cooper, Hiroeh, May, Appleby and House21 or were restricted to data from highly advanced institutions with specialist self-harm teams. Reference Kapur, Murphy, Cooper, Bergen, Hawton and Simkin17,Reference Bergen, Hawton, Waters, Cooper and Kapur22,Reference Kapur, Steeg, Webb, Haigh, Bergen and Hawton23 One strength of the present study was the that it was representative of the general in-patient population, being based on nationwide data from various types of hospitals.
Previous investigations have shown mixed results on the effect of psychiatric intervention. Reference Kapur, Murphy, Cooper, Bergen, Hawton and Simkin17 Most of those studies adopted a conventional regression model, which failed to adjust for patient backgrounds and hospital factors. Our study included various factors that could affect the probability of undergoing psychiatric intervention. Those key factors enabled us to conduct a propensity score analysis, which further reduced selection bias when estimating intervention effects from observational data. The model used in the matching method exhibited good discriminating ability in estimating receipt of psychiatric intervention (area under the receiving operating characteristic curve 0.768; 95% CI 0.763–0.773).
The proportion of readmission within 1 year (4.4%) was lower than the proportions of repeated self-harm (16%) reported in one systematic review. Reference Owens, Horrocks and House7 A recent systematic review of brief contact interventions also found a higher rate of repeated self-harm (intervention 9.8%; unexposed 11.1%). Reference Milner, Carter, Pirkis, Robinson and Spittal32 The reviewed studies identified repeated episodes using various methods such as the use of catchment areas for including patients, follow-up interviews and checking medical records. In the present study, we were able to identify only same-hospital readmission due to overdose, and this was a potential source of underestimation.
Our findings suggest that psychiatric intervention following admission due to drug overdose was associated with reduced readmission. Clinical guidelines from the National Institute for Health and Care Excellence in 2004 12 and Royal College of Psychiatrists in 2004 13 recommend that a psychosocial assessment by a trained mental health specialist be carried out for all patients who self-harm. In the present study, interventions by psychiatrists were associated with reduced risk of readmission, which suggests that such interventions are effective. However, because we were unable to assess the effect of intervention by other specialists, our findings may not apply to hospitals without consultation liaison services provided by psychiatrists.
The intervention in the present study included two aspects – assessment and psychotherapy. Because the DPC database lacked information regarding classification of the performed intervention, we were unable to distinguish between assessment and psychotherapy; likewise, we could not identify which elements of the intervention were effective. In the context of brief hospital admission, however, we assume that the intervention reflected the effects of assessment.
Several limitations of this study warrant consideration. First, the database we used did not include the severity of mental disorder, which may have influenced the probability of undergoing psychiatric intervention. Second, the recorded diagnoses in administrative claims databases are less well validated than those based on prospective cohorts or registries. Third, a large proportion of unspecified drugs may have caused confounding bias and led to underestimating the true effect of the intervention.
In conclusion, our study demonstrated that psychiatric intervention by psychiatrists before discharge was associated with reduced risk of repeated admission to emergency centres. These findings indicate the importance of psychiatric intervention for drug overdose patients admitted to emergency centres in preventing repeated admission.
This study was supported by a grant from the Ministry of Health, Labour and Welfare, Japan (Research on Policy Planning and Evaluation Grant No. H26-Policy-011).