This retrospective descriptive study included patients receiving oral MS, buprenorphine, or methadone in OUD context. It conformed to the RECORD-Pharmacoepidemiological recommendations (23–26).

Data were extracted from the French national healthcare system data (SNDS), often used for public health and pharmacoepidemiological research, between 01/01/2015 and 12/31/2015. SNDS covers 98.8% of the population, comprising exhaustive anonymous individual administrative, medical, and pharmacy data (27, 28). Anonymous identifiers link health reimbursement data, diagnoses codes from hospitalization discharge databases using the 10th revision of the international statistical classification of diseases (ICD-10), and the death registry. Administrative data provides sociodemographic information: year of birth, sex, date of death, free complementary medical cover (CMUc) for low-income status, and any recognized chronic conditions from the list of 30 long-term/major diseases (LTD-30) that are guaranteed full reimbursement for any medical fees. This list is reviewed annually by the government and includes diseases that require particularly costly medical treatment for at least 6 months, like cancer, diabetes, severe heart disorder, chronic psychiatric, neurological or muscular diseases, and chronic lung disease, etc. (29). Pharmacy data comprise anonymous pharmacy identifiers and exhaustive claims for all reimbursed medications dispensed in pharmacies (substances and quantities supplied, dates of prescription, and dispensing) including opioid medications, enabling the daily dosage given to regular users to be estimated. Medical data comprise anonymous doctor identifiers and the specialty of the prescribers.

This study was approved for medical research by the French institute for health data privacy (INDS, no. 176) and the French national data protection commission (CNIL, no. 1946535). French law prohibits the authors from directly sharing the data used for this study, but access can be requested directly from SNDS (website: https://www.health-data-hub.fr).

The criteria used for patient selection were validated by a previous study (30). In accordance with OMT prescription recommendations, we included all men and women aged 15 years and over to whom MS, buprenorphine or methadone was dispensed at least once in 2015. Patients who received regular and concomitant OMT and MS prescriptions were excluded from the analysis due to the inability to link potential complications to either of the two opioids.

OUD patients were identified as:

All patients diagnosed with cancer or receiving palliative care, as well as patients with chronic pain, were excluded. Patients with chronic pain were identified based on:

All ICD-10 codes applied to select the patients are outlined in Supplementary Table 1.

Medications were identified by their Anatomical Therapeutic Chemical (ATC) codes (“N02AA01” and “N02AA51” for morphine alone and in combination, respectively, “N07BC01” for buprenorphine, and “N07BC02” for methadone). Dates of dispensings were used to determine frequency of use.

For MS, buprenorphine and methadone “capsule,” regularity was defined as receiving the medication over at least three consecutive months, during which the treatment was regularly dispensed, i.e., with <35 days between each pharmacy dispensing. This 35-day threshold corresponds to French legal restrictions on opioid medications, which limit their prescription and dispensing to a maximum of 28 days, to which a grace period of 7 days was added in order to avoid overestimating medication discontinuation. Methadone 'syrup' is subject to stricter legislation, with a maximum dispensing period of 14 days, making it necessary to adapt the regularity criterion to 18 days (14 days, plus 4 days of grace period). Patients who fulfilled these criteria were considered regular medication users.

Four groups have been formed. The first comprised all patients with regular MS use in OUD context in 2015. Those who did not fulfill these MS regular-user criteria, but received at least two MS doses in 2015 were included in the second group. The last two groups comprised all OUD control patients receiving regular OMT, separated into buprenorphine on one hand and methadone on the other.

All administrative, medical and pharmaceutical data mentioned in “data source” were extracted. All ICD-10 codes applied to identify outcomes and covariates are outlined in Supplementary Table 1.

Diagnosis associated with hospital discharge were extracted to identify unintentional opioid overdoses.

In the same way, diagnosis associated with hospital discharge and LTD codes were extracted to identify the covariates: human immunodeficiency virus (HIV), hepatitis B virus (HBV), hepatitis C virus (HCV), and the main infection complications described as potentially related to intravenous drug injection (16–19). Arterial and venous thrombosis complications (20–22) and various data on comorbidities: severe chronic psychiatric disorders (LTD-23), alcohol use disorders (by ICD-10 code and specific treatments (33), [i.e., disulfiram (“N07BB01”), acamprosate (“N07BB03”), naltrexone (“N07BB04”), and nalmefene (“N07BB05”)], benzodiazepine [anxiolytics (“N05BA”) and hypnotics (“N05CD,” “N05CF”)], and gabapentinoids [pregabalin (“N03AX16”) and gabapentin (“N03AX12”)]) concomitant treatments were also collected. Coprescription was defined as receiving dispensings of the medications involved on exactly the same date, suggesting that the treatments were simultaneously on the same prescription.

Doctor shopping behavior (DSB) was measured in regular morphine, buprenorphine and methadone groups. DSB was defined as a combination of overlapping prescriptions for a specific medication from several different prescribers, dispensed in different pharmacies, to the same patient. This practice enables patients to increase the amount of medications they receive (34, 35) and is typically associated with high levels of misuse and/or diversion (36–40). In this study, the threshold defining a DSB was fixed as:

These thresholds correspond to those established by previous studies assessing DSB scores for opioid analgesics, ensuring the comparability of results (41, 42). DSB is measurable only for regular substance use and so is not applicable to the occasional MS group.

The daily opioid dose was calculated for the three regular groups of patients and its oral morphine equivalent was evaluated with a fixed 30:1 ratio for buprenorphine (43) and a validated variable ratio ranging from 4:1 to 12:1 based on the daily dose for methadone (44).

The Charlson Comorbidity Index (CCI), extensively applied in clinical research to account for the confounding influence of comorbidities was calculated (45–47). The CCI assesses the level of comorbidity by considering the level of severity of 19 predefined comorbid disorders, as well as the number of disorders present among them by means of a score (48).

Categorical variables were expressed as frequencies and associated percentages, and quantitative variables as mean ± standard deviation (SD) or median and interquartile range (IQR), according to their statistical distribution (normality assessed using the Shapiro-Wilk test). The comparison between groups was performed using the chi-squared test for categorical data or Fisher's test where appropriate, with a variance analysis for continuous variables or the Kruskal-Wallis test if normality was rejected.

To determine the influence of various factors associated with overdose in opioid patients, a univariate logistic regression model was performed. The associated p-values were computed with their corresponding odds ratios (ORs) and their 95% confidence intervals (95% CI). To study the factors associated with opioid overdose, a multivariate logistic regression analysis was performed. All variables associated with p < 0.25 in univariate analysis were included in the model. Age and sex were forced in the model. The corresponding adjusted ORs were calculated with their 95% CIs. All statistical analyses were conducted using SAS-9.4 software (SAS Institute, USA) and STATA-14.2 (StataCorp, USA).

From 1 January to 31 December 2015, 101,453 patients with OUD were included, among whom 2,237 patients receiving MS in the context of OUD (2.2%). MS groups were divided between the 1,288 patients who regularly received MS (1.3%), and the 949 who received it occasionally (0.9%) (see flow chart, Figure 1).

A total of 99,216 OMT controls were included in the same period, with 64,578 (63.7%) patients treated with buprenorphine, and 34,638 (34.1%) with methadone (see Supplementary Figure 1). Of the 100,504 OUD patients regularly receiving a regular opioid substitution (MS, buprenorphine or methadone), 1.3% (n = 1,288) were regular prescribed MS users.

The study population is described in Table 1. Mean ages were similar in MS and buprenorphine groups; methadone control patients were younger. All groups displayed the same sex ratio, four men to one woman. Over a third of patients receiving MS were considered low-income, based on their CMUc status. The poverty level was higher in regular MS users than occasional ones (p < 0.01), and overall, it was higher in MS groups than controls (p < 0.01) of which only a quarter benefited from CMUc.

MS users presented the highest prevalence of psychiatric disorders, with similar rates in both regular and occasional users (p = 0.1). Alcohol use disorder was more frequent for occasional than regular MS users (p < 0.01) and controls. Control groups were not different (p = 0.15). Regular MS users received benzodiazepines coprescriptions more frequently, with rates similar in other groups. Gabapentinoids coprescription rates were higher in MS users than in control groups, which were comparable between them (p = 0.5). Regular MS users received gabapentinoids more frequently (pregabalin and/or gabapentin) coprescriptions than occasional ones (p = 0.01).

There were significant differences across all MS patients and OMT control groups (p < 0.01) in terms of the prevalence of overdose. Occasional MS patients presented the highest prevalence, followed by regular MS users (p < 0.01). The controls were less affected, particularly those taking buprenorphine compared to methadone (p < 0.01). Compared to controls (no difference between buprenorphine and methadone, p = 0.88), the mortality rate was higher in MS users (p < 0.01), and significantly higher for occasional MS users than regular ones (p < 0.01).

DSB were significantly higher in regular MS users compared to controls (p < 0.01). Buprenorphine controls exhibited significantly higher DSB prevalence than methadone controls (p < 0.01).

Compared to OMT controls, between which no difference was found (p ≥ 0.1), the prevalences of HIV and HBV infections were significantly higher in MS users, although there was no difference between the regular and occasional groups (p ≥ 0.1). It is noteworthy that only the prevalence of HCV infection was different (p < 0.01) across all groups, as it was higher among MS (regular > occasional) users compared to controls (methadone > buprenorphine) (p < 0.01). The prevalence of bacterial infections was 3.5 times higher in MS groups, p < 0.01, with no difference between regular and occasional MS users, p = 0.62. The prevalence of thrombotic complications was higher in MS groups than in controls (p < 0.01), with comparable prevalence in occasional and regular MS users (p = 0.3).

The pharmaceutical product Skenan^®, a sustained-release MS capsule, was ahead of the other prescribed MS forms featuring on 91.3 and 86.9% of prescriptions dispensed to regular and occasional MS users, respectively. The pharmaceutical product Actiskenan^®, an immediate-release morphine capsule, was the second most frequent MS dispensed to patients with OUD, featuring on 18.8 and 31.5% of prescriptions to regular and occasional MS users, respectively. The pharmaceutical product Moscontin^®, an extended-release pill, came in third position, featuring in 5.2 and 2.7% of prescriptions dispensed to regular and occasional MS users, respectively (see Supplementary Table 2).

Analysis of Skenan^® prescriptions showed a preference for the highest unit doses among regular users, less so among occasional users. For Actiskenan^®, the dose distribution was more evenly distributed for regular users and low doses were more frequent in the occasional user group (see Supplementary Table 2). Regular MS users presented a median daily dose of 443.1 mg/day [IQR (192.8–758.4)], vs. 8.0 mg/day [IQR (4.1–14.5)] for buprenorphine-treated controls and 47.7 mg/day [IQR (28.9–71.8)] for those on methadone, corresponding to 240.7 mg/day [IQR (122.2–435.1)] and 286.3 mg/day [IQR (173.5–574.2)] equivalent oral morphine, respectively (Table 1).

The prescriptions mainly came from private practice medicine at similar rates among MS patients (regular: 87.4% of prescriptions, occasional: 85.4%) and buprenorphine controls (86.8%), slightly less for methadone controls (67.8%), mainly from general practitioners (GPs) for MS patients (regular: 97.8%, occasional: 97.7%) and similar in control groups (buprenorphine: 98.5%, methadone: 98.1%). Psychiatrists were the second main prescribers, accounting for <1.8% of prescriptions in each group. Occasional MS users more frequently received punctual conventional OMT prescriptions (26.2%) alongside those of MS (≥3 prescriptions/year) than regular MS users (20.8%).

In univariate analysis (Supplementary Table 3), general characteristics associated with opioid overdose were young age (p = 0.04), low-income (p < 0.01), receiving morphine rather than a validated OMT, particularly in the case of occasional MS use (p < 0.01), receiving high oral morphine equivalent (p < 0.01), and having treatment misuse behaviors according to DSB (p < 0.01). Having multiple comorbidities (history of severe chronic psychiatric pathologies, alcohol use disorder and systemic infectious complications (Supplementary Table 1) described as potentially related to intravenous drug injection, arterial, and venous thrombosis complications or according to the CCI score) was significantly associated with opioid overdose risk in univariate analysis. Receiving concomitant benzodiazepines or gabapentinoids and opioid prescriptions was associated with overdose risk in univariate analysis (p < 0.01).

Opioid dose in oral morphine equivalent and shopping behavior, reflecting misuse of the medication, were removed from the final logistic regression model because they could not be assessed for occasional MS users.

In the multivariate model, compared to regular MS users, occasional MS users had an increased risk of overdose [aOR = 2.2 (1.5–3.3)], while the risk was reduced in the buprenorphine group [aOR = 0.5 (0.4–0.7)] and not significantly different for methadone [aOR = 1.0 (0.7–1.4)]. When buprenorphine was used as a reference, occasional MS users were at the highest risk of overdose [aOR = 4.5 (3.4–6.0)], followed by methadone controls [aOR = 2.1 (1.9–2.3)], and regular MS users [aOR = 2.0 (1.5–2.8)]. The graphical representation of the resulting multivariate logistic regression model corresponds to the forest plot in Figure 2 (see also Supplementary Table 4). The area under the curve for multivariate model was equal to 0.822 ± 0.18 (Supplementary Figure 2).

Other factors associated with opioid overdose were age, with a risk decreasing over time regardless of the quartile evaluated (p < 0.01), being considered low-income [aOR = 1.1 95% CI (1.1–1.3), p < 0.01], history of mental health disorders [aOR = 1.7 (1.6–1.9), p < 0.01], alcohol use disorder [aOR = 4.9 (4.4–5.5), p < 0.01], infection complications described as potentially related to intravenous drug injection [aOR = 1.8 (1.5–2.2), p < 0.01], arterial and venous thrombosis complications [aOR = 2.1 (1.6–2.7), p < 0.01], HBV [aOR = 1.7 (1.1–2.5), p < 0.01], and HCV [aOR = 1.3 (1.1–1.5), p < 0.01]. Receiving concomitant benzodiazepines, anxiolytic [aOR = 1.8 (1.6–2.0), p < 0.01] or hypnotic [aOR = 1.5 (1.4–1.7), p < 0.01], or pregabalin [aOR = 2.1 (1.5–2.9), p < 0.01], and opioid prescriptions were associated with an increased overdose risk in multivariate analysis.

These findings should be interpreted by taking into account the strengths and weaknesses inherent in all pharmacoepidemiological studies using healthcare databases. The main risk is that the population included does not resemble that typically encountered in clinical practice. The characteristics of the patients included and controls were similar between groups and consistent with previous field studies (age, sex ratio, low-income status), as were our findings on complications (viral and bacterial infections, DSB) and medication dosages (9–12, 30, 60). This similarity validates our patient selection, despite a selective methodology, and shows the lack of influence of the impossibility to include the 10% of patients receiving their OMT dispensing in an addiction center rather than in pharmacies (9, 10).

Regarding the data source, the SNDS database only provides diagnosis codes attributed on hospital discharge (≥24 h), excluding care provided by emergency services, and we only had access to data on pharmacy dispensing, without details on illicit-market sales. These missing data minimize the size of the groups and the risks to which they are exposed. While our results add to the body of knowledge on opioid overdose and MS use, and are consistent with the results of a previous incidence study on the issue (30), a causal inference cannot be determined directly owing to the cross-sectional design of the study. Further longitudinal studies are needed to explore the temporal relationship between the factors we have identified and opioid overdose. Although oral morphine use disorder seems to be rather specific to France, the main risk factors for opioid overdose do not seem to be specific to this population. The generalization of the study results beyond the French territory must be done with caution and conditioned on future investigations. However, these results may be of interest to countries exposed to the intravenous diversion of oral opioid medications or which wish to offer patients an injectable opioid substitution.

This study also presents significant strengths, notably its ability to recruit a large number of patients for assessing rare issues. This pharmacoepidemiologic approach using a nationwide database is alone in providing sufficient cohort sizes to supply reliable data at the population level, particularly for the patients described here, who are few in number and difficult to follow-up in conventional clinical studies.