Non-medical use of opioids was found to be comparable in HIV-infected opioid dependent patients receiving methadone or buprenorphine. In addition, our results showed that ongoing use of non-opioids (such as cocaine, cannabis, and benzodiazepines), perception of withdrawal symptoms, and a shorter retention in OMT are associated with non-medical use of opioids. At a time when non-medical use of prescription opioids  and use of opioids by injection [20–22] are growing problems and a real concern for public health, identifying correlates of non-medical use of opioids is of interest and should help physicians to better manage opioid dependence and improve the effectiveness of OMT in not only HIV-infected population but also the whole population of opioid-dependent individuals.
First, it is interesting to note that, according to the results of the univariate analysis, HIV care has a positive impact on decreasing non-medical use of opioids. Patients who were receiving antiretroviral treatment (ART) were less likely to report non-medical use of opioids. This result may be related to the wider access to ART in IDUs who achieved stabilization on OMT . Moreover, it is known that access to ART has a positive impact on drug injecting cessation in HIV-infected IDUs . This impact could be explained by the reality of the relationship between physician and patient based upon trust in the ART initiation process , as suggested in our univariate analysis. This latter result supports the idea that access to ART in HIV-infected opioid dependent individuals could play a harm reduction role.
However, the results of the multivariate analysis show the most important predictors of non-medical use of opioids which are not related to HIV care in our analysis. And after multiple adjustments (including for non random assignment of buprenorphine and methadone), only shorter time since OMT initiation, experience of withdrawal symptoms, and non-opioid drug use were found to be significant determinants of non-medical use of opioids. Among the independent predictors of non-medical use of opioids, some patients' illicit substance use behaviors such as cocaine, daily cannabis, and benzodiazepine use were associated with the outcome. These findings are consistent with those of previous studies showing that non-opioid drug use in OMT-treated patients may negatively influence treatment outcomes. For instance, cocaine use during OMT has previously been shown to be a predictor of buprenorphine diversion by injection in patients receiving prescribed buprenorphine  and to hinder efficacy of OMT . In addition, OMT patients who had used benzodiazepines were more likely to have opioid-positive urine screens during OMT . This is a concern because few guidelines are available to physicians who are required to treat co-dependent drug users. A recent meta-analysis has summarized the most common approaches to treating benzodiazepine dependence  and the most effective is a combination of a gradual dose reduction regimen and psychological interventions. Moreover, a recent study showed that opioid dependent individuals receiving heroin-assisted treatment were more likely to decrease their benzodiazepine use compared to those receiving methadone maintenance treatment . Finally, heavy cannabis use also has been found to be correlated with buprenorphine injection . Polydrug use should be considered more thoroughly not only for measuring the independent effects of multiple drug use but also for investigating the effects of drug use patterns on behavioural outcomes .
Although the association between time since OMT initiation and non-medical use of opioids is not unexpected, it demonstrates that the longer the duration of OMT (retention in OMT), the better achievement of stabilization in terms of heroin use and OMT diversion . The clinical management of opioid dependence has already been described as a very long process with a high risk of cycling in and out of treatment .
Patients who reported withdrawal symptoms during the study period were also found to report higher non-medical use of opioids. This result is clinically very relevant and suggests that physicians should investigate subjective and objective withdrawal symptoms during treatment in order to better understand the reasons that inter-dose opioid withdrawal exist . Their occurrence may be indicative of inadequate OMT dosage [35, 36], either too low, which leads to withdrawal symptoms directly, or too high, which leads to aversive side effects that lead to discontinuation of OMT. However, the results do not support the notion that lower doses of OMT are related to more non-medical use of opioids, probably because low doses were a proxy for less severe opioid dependence and not for inadequate dose prescription. Other factors could be related to disliking of the subjective effects of OMT, or polydrug use, which also may lead patients to interrupt treatment [37, 38]. This issue is all more relevant since Mateu-Gelabert et al showed that withdrawal symptoms could increase risky behaviors regarding HIV and also Hepatitis C virus (HCV) transmission by undermining IDUs' willingness to inject safely .
Recent WHO guidelines recommend that the dose of OMT should be increased progressively according to clinical effect in the individual patient, up to an adequate dosage . In addition, drug interactions represent a potential concern in HIV-infected individuals, especially with regard to methadone and protease inhibitors, and in these circumstances, methadone dose adjustments may be required to avoid withdrawal symptoms . In addition, the experience of withdrawal symptoms among HIV-infected patients who are receiving OMT needs careful surveillance and clinical management since the presence of withdrawal symptoms has been associated with an increased risk of mortality in the same cohort .
After adjustment for non-random allocation to buprenorphine/methadone prescription, we found no difference between methadone and buprenorphine with regard to non-medical use of opioids. These findings are consistent with previous studies showing that methadone and buprenorphine are broadly equivalent at higher dosages . The correction for the bias elicited by non-random assignment was necessary due to the two different models of OMT delivery in France which may lead to the selection of different opioid drug dependent populations. Historically, in France, initiation of methadone is possible only in specialized centers for drug dependence while access to buprenorphine is available in primary care settings . This differential access could explain why methadone in France is more frequently initiated in patients presenting a severe addictive profile. In France, methadone programs provide multidisciplinary care, whereas buprenorphine is delivered through primary care settings, which have minimal ancillary services. Previous research has shown that retention in OMT is influenced by several in-treatment variables such as occurrence of depressive symptoms, social functioning  and trust in the physician . Therefore, although more severely dependent patients were included in the methadone group, it is not entirely surprising that non-medical use of opioids did not differ from the buprenorphine group in our study because the methadone patients had access to more comprehensive care .
Some limitations of this study should be acknowledged. Data on substance use was based on self-reported behaviors, which are often questioned due to possible social desirability bias. However, the validity and reliability of self-reports about active drug use have been established in many studies that used similar methods for collecting information about substance use behaviors [47, 48] as well as in a previous study with a sample of HIV-infected patients in which substantial agreement was documented between self-reported heroin use and morphine detection in urine . Our results do not allow a clear inference about the direction of effects. Although the study started some years ago, this cohort remains a very interesting observational sample of patients, especially for countries that have only recently started to scale up antiretrovirals and OMT for opioid dependent HIV-infected individuals. Finally, although a statistical adjustment was made to allow for non-random assignment to medication type, this study was not a controlled clinical trial and further investigations are required to confirm our findings.