From the results of this study, it can be deduced that the presence of alcohol or cannabis in patients hospitalized due to trauma was associated with an increased risk of TR. We also found an important increase in the risk of TR in the polydrug group. However, as alcohol was the substance most frequently found in different combinations with other drugs in the polydrug group (63.72% of patients were positive for alcohol in this group). This could suggest that the association between polydrug and TR was mainly due to this substance. Whether we take into account the first trauma during follow-up after discharge (almost 5 years), we found an increased risk of TR in all consumption profiles with the exception of the psychotropic medications/opioids and cocaine/amphetamines groups. However, the non-significant aHRR estimates in the latter group could be due to the small sample size: only 10 patients were positive screened for cocaine, amphetamines, and / or methamphetamines as a single substance. The consumption of these drugs often appears to be associated with other substances such as alcohol. Therefore, the majority of patients positive for cocaine, amphetamines, and / or methamphetamines were categorized in the polydrug group.

The associations found between the baseline characteristics of the patients and each subgroup of substances were also consistent with most of the previous studies. For example, the patients who tested positive for cannabis were young men (25 years, 95% men), and traffic crashes were the most common mechanism of injury [24]. In contrast, the characteristics of patients exposed to psychotropic medications/opioids differ from other substance users, as they were predominantly women, were older than the other groups, and had falls as the most common mechanism of injury. It is important to note that almost three quarters of patients in the psychotropic medications/opioids group had a history of psychiatric disorder. Given that the history of psychiatric disorders is a well-known risk factor for traumatic injury [18], the inclusion of this variable in the adjusted models is essential to establishing a relationship between the consumption of psychotropic medications/opioids and TR. The increase of 39%-45% in the risk TR that we found associated with mental disorders could be the reason for the absence of a significant risk of TR in the group of psychotropic drug users, despite the fact that in this group, the incidence rate was twice that of the negative subcohort.

In patients positive for the analysed substances, the time to new trauma in the follow-up period was 16 months; however, in the group of patients negative for all substances, the next trauma took twice as long to occur. The results of the adjusted models estimate an increase in risk attributable to the detection of alcohol from 90% in the case of the first new trauma to 133% if we take into account all traumas during follow-up. This strong association of alcohol consumption with TR has been shown in other studies [10,14,16]. In a systematic review, Nunn et al 2016 [10] found that the rate of alcohol-related trauma recidivism ranged from 26.7% to 76.9% among individual studies. They calculated a weighted aggregate estimate of 41.0%. However, this result should be interpreted with some caution due to methodological limitations and considerable variation among the studies included in this review. There may be considerable variations between studies due to the lack of a universally accepted definition for a trauma patient and disparity in the follow-up periods used to determine whether a repeat admission for injury had occurred. Studies also differed in the types of trauma patients they included. For example, Vaaramo et al 2008 [14] was to patients with injuries in a specific body region, such as the head. They found that even head trauma without traumatic brain injury under the influence of alcohol implies an elevated risk of subsequent traumatic brain injury (aHR 2.51, 95% CI 1.38–4.56, p < 0.01). Others have addressed this issue only in motor vehicle collision. With this approach, Fabbri et al 2005 [16] found that alcohol was the most powerful behavioural factor predicting recurrent events in subjects treated in an ED for injury after motor vehicle crash (relative risk: 3.73, 95% CI 3.00–4.64).

The presence of other substances in addition to alcohol is not usually contemplated. The studies that have addressed the relationship between the use of illicit drugs and TR find that cocaine is the substance that is most frequently associated with TR [9,19]. Those results are consistent with ours, if we consider the combined use of cocaine with other substances mentioned above. On the other hand, the increased risk of TR found in cannabis users is of particular relevance if we consider that it was the group with the highest proportion of road traffic crashes. Due to their potential to cause serious injuries, traffic crashes are a public health problem of the first order [20]. In recent years, an increase in cannabis use and a decrease in the perception of the risk of consumption for health in general have been detected in our country [25]. We also find a low perception when cannabis users who have suffered trauma are asked about the perceived risk that consumption entails at the time of suffering road traffic injury [26].

Our study shows that in addition to being common among trauma patients, exposure to alcohol is an important factor when detecting people at risk of new trauma. The clinical usefulness of the approach used in our study is that even if we ignore the history of consumption or its severity, the detection of exposure during systematic screening in patients admitted for trauma is sufficient to conclude that a patient has double the risk of suffering new trauma after discharge if no health education intervention is practiced. This reinforces the relevance of SBIRT programmes [22] in trauma centres and the BMI [23] as a tool to halve the risk of recurrence [21].

Our work has several strengths compared to other study designs. First, our study is based on a programme of systematic substance screening so that the possibility of selection bias in the original sample is lower than in other studies in which the detection of drugs depended on the decision of the physician [12,16]. Second, positive results for benzodiazepines and opiates have not been ruled out, but a distinction has been made between the patients who consumed these substances prior to the trauma and those to whom the substances were administered in the emergency room and by emergency services. This distinction has allowed us to quantify the effect of these substances on recurrence, unlike other similar studies [27,28]. Finally, the categorization of patients into subgroups according to consumption profile allows for a more valid assessment than those performed in other studies, where drugs other than alcohol were included in a single group [27–29].

The main limitation of this study is the confounding bias inherent in any observational study, which leads to some doubt in drawing conclusions regarding the causality of the associations observed. Although we have included the main factors related to both consumption and TR in the model, it is possible that there are other unmeasured confounders. Some candidates could be occupational factors or those related to the level of schooling [18]; however, cohort studies based on large populations of trauma patients [12,30] do not find a relationship between these variables and injuries in favour of other measures, such as young age, male sex or psychiatric disorders. Therefore, we consider that the potential bias of these confounders should not have much impact on our estimates. On the other hand, due to the different window of detection for each drug in urine, interviews were conducted to confirm whether the consumption of drugs other than alcohol occurred 6 hours before the injury. This may introduce social desirability and/or recall bias. However, this seems unlikely after revealing exposure to the substance with a laboratory test. The fact that the sensitivity analysis showed no differences supports this assertion. Another aspect to consider is that the intent of the injury has not been discriminated. However, often the intent is difficult to determine. We were able to collect information about diagnoses of mental disorder. Psychiatric conditions are strongly associated with self-harm, so we consider that an important part of the confusion related to intent is eliminated by introducing mental disorders as a covariate in the models. Finally, selection bias could arise because not all hospitalized trauma patients could be screened and subjects who had BMI were excluded. Nonetheless, we have no reasons to suspect an independent association between the presence of drugs and the probability of being screened or an association between the consumption profile and a higher probability of receiving the BMI.