Among presentations with acute recreational drug toxicity, 11.9% self-discharged from the ED. Self-discharge rates varied between centres, from 1.7 to 17.1%. In self-discharge cases the patient stayed shorter, 2 h 20 min vs. 4 h 25 min, and less frequently received treatment beyond mere observation, 33.5 vs. 53.5%. Having taken synthetic cannabinoids and heroin was associated with self-discharge, as were agitation and naloxone treatment. Sedation treatment, flumazenil treatment, and the presence of nearly any clinical feature but agitation were associated with a lower risk of self-discharge.

The overall self-discharge rate of 11.9% was somewhat lower than the 15–19% previously reported among patients treated for recreational drug toxicity [5, 10, 11], but clearly higher than the 1–3% reported in general hospital populations [1,2,3, 5], and also higher than the 6–11% reported among patients with acute poisoning in general [12,13,14]. Considering the increased risk of premature death associated with drug overdose [15, 16] and with self-discharge [1, 2], our study substantiates previous findings that patients self-discharging during treatment for recreational drug overdose are at-risk patients in an at-risk situation [10, 11].

The variation of self-discharge rates across the centres in our study may to some extent reflect different discharge procedures and practices. Though self-discharge in part may be related to patient characteristics, situational factors also are at work [1, 21, 22]. We found higher self-discharge rates overnight and on weekday evenings, which might possibly be due to a larger caseload at these times combined with less staff on night shifts. Except for the two Oslo centres, all our data were collected from regular hospital EDs. In Oslo, the OAEOC is a primary care unit functioning as a pre-ED for the city’s hospitals. The less complicated cases are treated there. Hence, the hospitalized cases treated at the OUH are more complicated, and nearly 90% are admitted to intensive care [23]. This intra-city division of tasks may partly explain the high and low discharge rates at the Oslo centres. Another observation worth exploring is that centres with large volumes of presentations tended to have high self-discharge rates. Still, there are several exceptions to this trend, and even more so when looking at the total number of presentations and not just recreational drug toxicity presentations. Hence, other mechanisms than caseload are probably also at work. Some centres require the patient to sign a form before self-discharge, and others do not. At some centres, self-discharge may not be officially allowed, which would probably impact on how potential self-discharge is dealt with. There may also be architectonical differences making the way out of the ED more or less cumbersome. Differences in discharge procedures should be further investigated to find measures to reduce self-discharge.

Specific interventions to reduce the risk of self-discharge have been tried. In a US study of alcohol-intoxicated ED patients, a protocol for identifying and monitoring incapacitated patients and making it more cumbersome for them to leave until they had regained the capacity to make medical decisions, reduced self-discharge from 15.0 to 7.4%, and increased ED stay by 42 min [24]. The incapacitated patients were placed in a well-supervised area without easy egress, their shoes were removed, and their own clothes were replaced with a gown.

Delay in diagnostic and therapeutic procedures is a frequently stated reason for self-discharge [25]. Waiting is tiresome. Hence, striving for rapid medical clearance makes sense as a measure to reduce self-discharge by reducing time spent waiting for procedures or results. On the other hand, what patients perceive as waiting time, clinicians may perceive as observation time necessary to ensure that the risk of harmful toxicity has passed [26].

Not surprisingly, in self-discharge cases, the patient stayed for a shorter time in the ED. They also to a lesser extent received treatment beyond mere observation. Some patients may have self-discharged before recommended treatment or observation, which would put them at risk. Still, in our clinical experience, important treatment for patients with acute recreational drug toxicity (e.g., naloxone for respiratory depression from opioid overdose and sedation for agitated patients) will often be given early after presentation as this is when the patient is most severely affected. Unfortunately, we were not able to explore this in our data set as the time of giving specific treatments was not registered. Furthermore, many patients treated for recreational drug toxicity would benefit from referral to follow-up for substance use disorders, mental health issues, and somatic co-morbidity [27, 28]. Self-discharge often closes these options. On the other hand, some self-discharging patients may have been less severely sick, not in need of any specific treatment, and consequently preferred to self-discharge rather than wait for medical clearance.

Clinical features of more severe toxicity—hyperthermia, tachypnoea, hypotension, chest pain, arrhythmias, and psychosis—were associated with a reduced risk of self-discharge. The presence of physical symptoms may increase the patient’s concern. Also, clinicians may be more concerned about patients with physical symptoms and therefore put in more effort to prevent them from leaving. In addition, these patients are more likely to be offered treatment which may reduce the risk of self-discharge, e.g., sedation. Still, though only a handful of cases with hyperthermia or arrhythmias self-discharged, substantial numbers with other serious clinical features did.

Naloxone treatment was associated with self-discharge. Though we did not collect information on opioid withdrawal as such, it is possible that naloxone precipitated opioid withdrawal symptoms in some cases leading to self-discharge. This may also partly explain the association between self-discharge and heroin. The median length of stay for self-discharge cases with opioid toxicity receiving naloxone was 2 h 11 min, with one out of four leaving after less than 1 h 3 min. Hence, in many cases, patients with opioid overdose self-discharged sooner than the recommended 2 h of observation after naloxone administration [26]. This may increase their risk of recurrent opioid toxicity, especially in the setting of long-acting opioids, an increasing problem in several European countries [29]. These patients should be targeted for interventions to reduce their risk of self-discharge. Caution should be taken when administering naloxone so as to avoid precipitating withdrawal symptoms. In one out of four cases with opioid overdose, the patient self-discharged after 4 h 39 min, which is long enough for withdrawal symptoms to develop even without naloxone administration. This may also be part of the explanation of the association between self-discharge and heroin. Accordingly, improved management of opioid withdrawal might also reduce self-discharge.

For synthetic cannabinoids, a possible explanation for the association with self-discharge may be agitation, a frequently reported feature of synthetic cannabinoid toxicity [30]. Agitation may also be associated with self-discharge as a feature of opioid withdrawal. Furthermore, it is likely that clinicians let agitated patients leave before medical clearance to reduce risk to staff and as an alternative to sedation treatment. Hence, measures to improve the management of agitated patients may be a fruitful approach in reducing self-discharge, e.g., providing a calm environment, effective and consistent communication, and timely sedation treatment [31,32,33].

Psychosis, hallucinations, and anxiety were associated with a lower risk of self-discharge, in contrast to agitation. Again, these symptoms may increase the patient’s concern. Healthcare workers may also be more concerned about these patients, hence keeping them on for treatment or observation and sometimes initiating mandatory admission to a psychiatric ward.

The drug(s) involved in the presentations were based on patient self-report and the clinical assessment of the doctor treating the patient. This results in some uncertainty as to which drug(s) were taken. However, toxicologic laboratory testing often confirms the drugs reported by the patient and their companions, though frequently more drugs are found on testing than were reported [34]. Still, this is generally the mainstay of the clinical diagnosis with laboratory testing undertaken in only a minority of presentations in most settings.

As we only registered presentations and not individual patients, we were not able to identify whether any patients re-presented, and hence could not assess the risk of unfortunate outcomes for the self-discharging patients. Furthermore, as self-discharging patients tend to present with repeated poisonings [10], the frequency of characteristics associated with self-discharge may be exaggerated as we have counted them for presentations and not patients.

We did not collect information on discharge procedures at the participating centres. Hence, we were not able to explore the possible procedural mechanisms explaining the large variation in self-discharge rates across the centres.

The data material is six years old. However, as far as we know, interventions to reduce the risk of self-discharge have not been broadly implemented since then. Furthermore, we are not aware of changes in procedures that might affect our results, apart from temporary changes during the recent pandemic.