Predictors of treatment allocation guesses in a randomized controlled trial testing double-blind injectable hydromorphone and diacetylmorphine for severe opioid use disorder

Abstract

Background: SALOME (Study to Assess Longer-term Opioid Medication Effectiveness) tested in a double-blind non-inferiority clinical trial if hydromorphone could be as effective as diacetylmorphine for severe opioid use disorder. Although participants did not guess treatment correctly beyond what is expected by chance, perceived treatment assignment could affect patients’ response to treatment. This study tested if treatment allocation guess is associated with treatment outcomes and identified predictors of guess.

Methods: Data were obtained through questionnaires and clinical records. Participants were asked what medication they thought they were receiving (diacetylmorphine, hydromorphone or unsure) and their open-ended reasons. Multinomial logistic regression was used to assess the predictors of treatment guess. An inductive thematic analysis was used to code open-ended responses. This clinical trial was registered with U.S. National Institutes of Health (https://clinicaltrials.gov/ct2/show/NCT01447212).

Findings: Participants referred to prior experiences with opioids and the presence or absence of specific drug effects as reasons for their guesses. There were no differences in illicit opioid use and retention by guess; however those who guessed diacetylmorphine had better physical and mental health scores. Participants with a treatment-related observed drowsiness event, and higher perceived drug-related high scores were more likely to guess diacetylmorphine compared to hydromorphone. Guessing hydromorphone was more likely among those who made negative comments as reasons for treatment guesses.

Conclusions: Understanding the clues participants use for treatment allocation guesses and relating them to treatment expectations could be integrated with accurate information about the treatment, providing an opportunity for patient–physician shared decision-making in opioid maintenance treatment.

Clinical trial registration: U.S. National Institutes of Health; ClinicalTrials.gov Identifier: NCT01447212; https://clinicaltrials.gov/ct2/show/NCT01447212

Keywords:

• Blinding
• masking
• opioid dependence
• supervised injectable diacetylmorphine
• supervised injectable hydromorphone
• treatment expectations

1. Introduction

Substitution with medically prescribed opioids has proven to be effective for the treatment of opioid-dependence (Mattick et al. 2009, 2014). Medically-prescribed heroin (i.e. diacetylmorphine) provided in supervised clinics has been shown to be a safe and effective alternative for long-term opioid users not attracted or retained to first line treatments (e.g. oral methadone) that continue injecting street opioids (Perneger et al. 1998; van den Brink et al. 2003; March et al. 2006; Haasen et al. 2007; Oviedo-Joekes et al. 2009; Strang et al. 2010; Ferri et al. 2011; Demaret et al. 2015). However, in many contexts, regulatory and political barriers have limited the expansion of diacetylmorphine treatment. This has led clinical researchers to recently test an alternative medication for these patients. SALOME (Study to Assess Longer-term Opioid Medication Effectiveness), a double-blind non-inferiority clinical trial, demonstrated that injectable hydromorphone, a potent short acting opioid licensed for analgesia, could be utilized for long-term opioid dependence in jurisdictions where diacetylmorphine is not licensed and/or in patients for whom it is contraindicated or unsuccessful (Oviedo-Joekes et al. 2016).

In the SALOME trial, analysis of the blinding showed that participants did not correctly guess their treatment allocation beyond what would be expected by random guessing. Specifically, 48% of participants in the hydromorphone group guessed they were receiving diacetylmorphine or were unsure. Likewise, 64% of participants in the diacetylmorphine group guessed they were receiving hydromorphone or were unsure (Oviedo-Joekes et al. 2016). Although the blinding was maintained in SALOME, evidence suggests that the treatment a participant perceives he or she is receiving may be associated with his or her treatment response (Wright et al. 2012; Buchanan et al. 2013; Correa et al. 2014). For example, Correa and colleagues found that, regardless of treatment allocation (i.e. randomization arm), perceived treatment assignment at the mid-treatment assessment was a significant predictor of self-reported and behavioral smoking endpoint outcomes, suggesting that treatment expectations can affect treatment outcomes (Correa et al. 2014). Similar findings were observed in a double-blind placebo-controlled trial for alcohol dependence. Although outcomes did not differ between treatment groups, participants that perceived they had been assigned to the active treatment had better outcomes (e.g. lower alcohol dependence score) compared to those that expected they were receiving the placebo (Colagiuri et al. 2009). While the mechanisms are not certain, the evidence from such placebo-controlled studies strongly suggests that treatment expectation is an important factor to consider when testing treatment outcomes and when interpreting these outcomes for standard clinical practice (Colagiuri 2010; Dar & Barrett 2014).

It has been suggested that, even if there is an association between treatment guess and treatment outcomes, there is no certainty of what treatment guess is measuring (Schulz et al. 2010). For example, participants may not completely grasp the effects of the medication. As such, their guess will be based on their perceptions and knowledge of the medications, which may not necessarily be accurate (Schulz & Grimes 2002). Participants could also be making guesses (correct or not) based on experiences during the trial that they attribute to the treatment (correct or not). For example, in superiority clinical trials, participants generally expect the ‘new’ medication to have a larger effect (Schulz & Grimes 2002). Thus, it is possible that participants who perceive their treatment to be effective may be more likely to ‘guess’ they are receiving the new treatment (independent of treatment allocation). Finally, perceived treatment assignment and perceived treatment efficacy are two different sets of expectations and how these contribute to the overall outcome expectancy is difficult to disentangle (Colagiuri 2010). To advance understanding of what is being measured when the blinding is tested, the incorporation of qualitative responses of reasons for treatment guess has been suggested, along with asking at multiple points in the study (Schnoll et al. 2008; Thomas et al. 2008; Colagiuri et al. 2009).

The aims of this study are to test if perceived treatment assignment is associated with treatment outcomes and to identify predictors of perceived treatment assignment, including reasons for guesses among participants receiving double-blind injectable hydromorphone or diacetylmorphine. The analysis of treatment expectancies and participants perceptions can be used to complement the findings of double-blind clinical trials and to inform clinical practice.

2. Methods

2.1. Design, setting, participants

SALOME was a randomized, double-blind non-inferiority controlled trial comparing the effectiveness of either injectable hydromorphone or injectable diacetylmorphine received under supervision for six months for the treatment of long-term opioid dependence. The study was conducted in Vancouver (Canada) between December 2011 and December 2013 and received ethical approval from the Providence Health Care/University of British Columbia Research Ethics board. A total of 202 participants provided fully informed consent and were randomized to either injectable hydromorphone or injectable diacetylmorphine. Full details regarding the parent study screening procedures, participant profile, design (including procedures to ensure the blinding was not broken) and main results, have been published elsewhere (Oviedo-Joekes et al. 2015a, 2015b, 2016).

2.2. Procedures and measures

The protocol of this sub-study included questions regarding treatment expectations and perceived treatment assignment. It has been suggested that drawing attention to the blinding may cause unblinding (Kolahi et al. 2009). If drawing attention to the treatment assignment does in fact affect the blinding, it could interfere with the clinical trial for example, by affecting participants’ perceptions of drug effects that could potentially impact treatment adherence. Thus, the design of this study was intended to minimize attention to the blinding by not repeating questions, not prompting participants for clarifications and by interspersing questions throughout the overall evaluation.

Data for the present study were obtained through questionnaires and clinical records. A research team, operating independent of the clinical team, administered questionnaires at baseline and six months. Data collected included socio-demographics, illicit drug use, illegal activities, and physical and mental health (Maudsley Addiction Profile (Marsden et al. 1998)). Treatment assignment expectation questions were asked only at baseline (e.g. what treatment do you wish to receive). Perceived treatment assignment related questions were asked twice: shortly after treatment titration (approximately one week following treatment initiation) and at six months follow-up. Participants were asked, ‘what treatment do you think you are receiving?’ with five options: (1) heroin definitely, (2) heroin possibly, (3) not sure, (4) Dilaudid^® possibly, and (5) Dilaudid^® definitely. Two open-ended questions followed, asking when and why participants thought they were receiving that treatment (or were unsure). No clarifications were asked nor were participants probed to expand on their answers. A Visual Analog Scale (VAS) (Preston & Bigelow 2000) was collected after injecting the medications at the clinic site, and only once (with the first measure of perceived treatment assignment) by a trained peer-interviewer. The VAS asked participants to rate the degree of perceived drug effects and drug liking on a 100-point scale ranging from none to extremely. Data collected at the clinic resulted in complete observations for 177 participants (25 missing). The six-month questionnaires, collected by the research team were obtained from 198 participants (four missing).

Historical administrative prescription records for prior hydromorphone treatment were obtained from the centralized British Columbia (BC) provincial drug dispensation database (PharmaNet). Daily clinical records were used for study medication dose, days of treatment, and adverse and serious adverse events (AEs). All study participants were assessed for AEs, drug reactions or changes in health status before, during and after each injection. Registered nurses conducted these assessments and documented AEs and their relationship to the study medications.

2.3. Analysis

Perceived treatment assignment or ‘treatment guess’ was collapsed into three groups: (1) heroin definitely or possibly = diacetylmorphine, (2) Dilaudid^® definitely or possibly = hydromorphone and (3) Unsure. Analyses are described for the following: (a) participants’ reasons for treatment guess (open-ended comments); (b) differences in treatment efficacy outcomes by treatment guess (treatment guess as the independent variable) and (c) predictors of treatment guess (treatment guess as the dependent variable).

2.3.1. Analysis of open-comments

Thematic analysis of open comments took place in stages and authors (KM, HP and EOJ) remained blinded to randomization arm and guess during the first round of analysis. In the first part, an inductive approach was taken (Starks & Trinidad 2007) where each comment was assigned a code based on its semantic content. The strategy of constant comparison (Schensul et al. 1999) was used to develop themes and relationships across all participants’ comments. After initial independent coding, the authors (KM and HP) met to discuss and review the initial list of themes. Themes were then further refined to ensure congruency between the content and the assigned theme (KM, HP and EOJ). Next, themes were clustered into categories and the content was reviewed again. Four categories emerged for participants’ reasons for perceived treatment assignment: presence of drug effects, absence of drug effects, prior experience with opioids and other (emerged mostly among those guessing unsure). Further attributes, derived from the content of the data, were assigned to each theme: positive (e.g. ‘has legs’), neutral (e.g. ‘because I know heroin’) or negative (e.g. ‘I get itchy after injecting’). At this stage, authors (KM, HP and EOJ) confirmed the coding of each response in the four categories and corresponding attributes. Data are reported for the six-month follow-up because there were no differences in the emerging categories and attributes between the reasons given early in the study and at six months follow-up. A total of 471 references were coded in 198 participants. Therefore, data presented are the number of participants with a reference coded in the given category as one participant could have more than one reference, and references were not mutually exclusive.

2.3.2. Statistical analyses

To test differences in treatment outcomes by treatment guess, pairwise comparisons were performed, using chi-square test for binary variables (positive for street heroin markers and compliance), and regression (physical and psychological health scores) or negative binomial regression (days of illicit heroin, opiates, crack cocaine, and illegal activities) for continuous variables. Continuous outcomes were adjusted by their baseline values using analysis of covariance. Additionally, statistical interaction between treatment guess and treatment assigned was tested for each outcome to disentangle the role of treatment guess.

For predictors of treatment guess, unadjusted multinomial logistic regression was used to summarize the association between predictors and treatment guess. Multivariable multinomial logistic regression was used for modeling predictors of the nominal outcome guess variable (Diacetylmorphine, Hydromorphone, Unsure). Covariates were selected based on univariate analyses; variables considered in the model had a p value ≤.2 (i.e. an entry significance level of .2). We then constructed the final model using the variables selected in the first step. The selection criteria for the final model were a p value of ≤.1 for each individual variable and a smaller Akaike Information Criterion (AIC) (smaller AIC suggests better model fit) (Akaike 1987). Statistical analyses were conducted separately for the two time points, early in the study and at six-month follow-up. Models for the two time points yielded very similar results and as such we report here only the analysis from six-month follow-up (see Supplement 1 for findings from early in the study). The final model was fit on 175 observations (instead of 198) to be able to include the VAS, which was only assessed once shortly after titration, with responses missing for 25 participants. The VAS was deemed to be clinically important in providing information as to the outcome of treatment guess and thus was retained in the model. All models were fit using Firth’s Penalized Likelihood, a method commonly used to avoid bias typically found when building models with a small number of observations in one or more groups (Bull et al. 2007) (i.e. those that guessed unsure). Data analyses were conducted using SAS version 9.4 and R version 3.3.1 (Cary, NC).

3. Results

At baseline, 83.2% of participants indicated that they wished to be randomized to injectable diacetylmorphine, 5.4% to hydromorphone and 9.4% were unsure (2% missing). When asked if participants would start injectable hydromorphone treatment if it were the only option available, 82.2% responded yes, 5.4% no, and 8.9% were unsure (3.5% missing). Early in the study, shortly after titration, when asked what treatment participants thought they were receiving, out of the 202 participants allocated to receive treatment in the clinical trial, 70 (35%) guessed diacetylmorphine, 61 (30%) hydromorphone, and 46 (23%) were unsure (25 = 12% missing). At six-months, 64 (32%) guessed diacetylmorphine, 95 (47%) hydromorphone, and 39 (19%) were unsure (2% missing). A total of 179 (out of 196, six missing) participants thought that the treatment they were receiving was effective.

3.1. Participants’ reasons for treatment guess

Table 1 shows summary statistics for emerging categories and their association with treatment guess at six-months. The most common reason provided in response to treatment assignment guess was participant’s prior experiences with opioids, primarily ‘heroin’ and ‘Dilaudid^®’ (i.e. hydromorphone), which led participants to ‘know how the drug feels’. Most participants provided such statements as:

The sensation. I have done a variety of opiates and can tell (Participant 6081; Guessed hydromorphone, Treatment Arm = hydromorphone; Gender = Man)

Other participants expanded on why their prior experiences with opioids led to a particular guess by specifying the reactions and sensations they associated with ‘heroin’ or ‘Dilaudid’:

When I have done Dilaudid in the past, the feeling, the high, is in my body, in my muscles. And when I do heroin, I get a hot, itchy feeling. When I do the SALOME treatment, I get that heroin itchy pins and needles feeling. (Participant 6005, Guessed diacetylmorphine, Treatment Arm = diacetylmorphine; Gender = Man)

Leaning towards Dilaudid because the effect that I’m getting is not the same as heroin on the street. When I did heroin on the street I could get things done, I could make phone calls, get out and motivated… this drug makes me really very sleepy, no energy, not getting things done, the house is sloppy. I doze off more frequently and for longer. This all makes me think it’s not heroin. (Participant 6103; Guessed hydromorphone; Randomization = diacetylmorphine; Gender = Man)

There were no differences in this category when examining treatment assignment (randomization arm) by treatment guess. For example, the 47 participants who guessed hydromorphone because the medication ‘does not feel like heroin’ or ‘feels like hydromorphone’, were distributed evenly by treatment arm: 24 were assigned to hydromorphone and 23 to diacetylmorphine.

Table 1. Reasons for perceived treatment assignment guess – summary statistics for emerging categories and association with treatment guess at six months.

	Guessed DAM	Guessed HDM	Unsure
Major categories^a	n = 64	n = 95	n = 39
Prior experience with opioids:
Yes (e.g. ‘I know how the drug feels’)^b^,^i^,^j	32 (36.0%)	53 (59.6%)	4 (4.5%)
No	32 (29.4%)	42 (38.5%)	35 (32.1%)
Presence of drug effects:
Made negative comments (e.g. ‘I get itchy’)^c^,^h^,^j	13 (16.9%)	55 (71.4%)	9 (11.7%)
Did not make negative comments	51 (42.1%)	40 (33.1%)	30 (24.8%)
Made positive or neutral comments (e.g. ‘the taste’)^d^,^h^,^i	32 (58.2%)	19 (34.5%)	4 (7.3%)
Did not make positive or neutral comments	32 (22.4%)	76 (53.1%)	35 (24.5%)
Absence of drug effects:
Made positive comments (e.g. ‘has no legs’)^e^,^h	2 (10.0%)	18 (90.0%)	0 (0.0%)
Did not make positive comments	62 (34.8%)	77 (43.3%)	39 (21.9%)
Made neutral comments (e.g. ‘I get pins and needles’)^f	9 (47.4%)	9 (47.4%)	1 (5.3%)
Did not make neutral comments	55 (30.7%)	86 (48.0%)	38 (21.2%)
Other comments:
Yes (e.g. ‘unsure’; ‘inconsistent effects’)^g^,^i^,^j	6 (12.0%)	7 (14.0%)	37 (74.0%)
No	58 (39.2%)	88 (59.5%)	2 (1.4%)

DAM: diacetylmorphine; HDM: hydromorphone.

Data shown are n (%). Significance testing from the unadjusted multinomial logistic regression p<.05.

a Categories for emerging themes from open-ended reasons for perceived treatment assignment are not mutually exclusive since participants could have referenced the presence and absence of effects within the same open comment.

b Reasons for treatment assignment guess that referenced prior experiences with illicit heroin, licit and illicit hydromorphone and morphine were coded in this theme.

c Reasons for treatment assignment guess that referenced undesirable or uncomfortable side effects (e.g. headaches, heavy feelings in chest, anxiety).

d Reasons for treatment assignment guess that referenced general side effects (e.g. the feeling) or positive effects (e.g. pain management).

e Reasons for treatment assignment guess that referenced the absence of a desired drug reaction (e.g. euphoric sensations).

f Reasons for treatment assignment guess that referenced the absence of an anticipated drug effect that was perceived to be associated with DAM or HDM.

g This theme was created to capture references by participants who were unsure, for participants with missing reasons, and references with unclear meanings (e.g. ‘inconsistent’).

h For DAM vs. HDM.

i For DAM vs. unsure.

j For HDM vs. unsure.

The presence of negative drug effects, primarily histamine reactions, such as itchiness, flushing and pins and needles, were referenced by a higher percentage of hydromorphone (55/77 = 71.4%) compared to diacetylmorphine (13/77 = 16.9%) guessers. However, as summarized in the following excerpts, these reactions (e.g. ‘itchiness’) were not unique to one particular guess (or treatment) group.

The way I feel; I get itchy after injection (Participant 6126, Guessed diacetylmorphine, Treatment Arm = diacetylmorphine; Gender = Man)

Because of the reaction, the pins and needles (Participant 6043, Guessed hydromorphone, Randomization = diacetylmorphine; Gender = Man)

It wasn’t like good heroin. The high was different. It seemed like the drug was being switched up. Sometimes I was itchy, sometimes I wasn't. Strength varied too. (Participant 6188, Guessed hydromorphone, Treatment Arm = diacetylmorphine, Gender = Man)

Inconsistent, sometimes feel itchy, sometimes feel nothing at all. (Participant 6121, Guessed Unsure, Treatment Arm = hydromorphone, Gender = Man)

General drug reactions, such as ‘the feeling’ or ‘the taste’, with no added connotation were also frequently referenced. Only eight participants (five diacetylmorphine guessers; three hydromorphone guessers) referenced reactions that were clearly viewed as positive, including reduced withdrawal symptoms, effective pain management and improved clarity or focus.

Experience with Dilaudid proved ineffective for pain… the SALOME drug is effective for pain. (Participant 6037, Guessed diacetylmorphine, Treatment Arm = diacetylmorphine, Gender = Man)

Not nodding, different feeling from heroin, makes me very clear mentally, gives me energy. (Participant 6040, Guessed hydromorphone, Treatment Arm = diacetylmorphine, Gender = Woman)

Compared to the presence of drug effects, the absence of drug effects was less frequently referenced. Among those indicating that positive drug effects were not present, such as euphoric sensations or length of drug effects, 90% (18/20) guessed hydromorphone.

Missing euphoria that comes with heroin (Participant 6048, Guessed hydromorphone, Treatment Arm = diacetylmorphine, Gender = Woman)

The drug is short acting – getting sick by next session. (Participant 6029, Guessed hydromorphone, Treatment Arm = diacetylmorphine, Gender = Man)

The absence of neutral, but expected drug effects, primarily ‘pins and needles’ or ‘the taste’ were also referenced. These reasons were similarly distributed between diacetylmorphine (n = 9) and hydromorphone (n = 9) guessers.

Didn’t get pins and needles. (Participant 6032, Guessed diacetylmorphine, Treatment Arm = diacetylmorphine, Gender = Woman)

Just noticing the effect the medication has on me compared to the others in the group. They are on the nod, and I am not (Participant 6154, Guessed Unsure; Treatment Arm = hydromorphone; Gender = Man)

Finally, participants referencing the ‘other’ category primarily guessed unsure (37/39). While few ‘unsure’ participants expanded beyond this, participants who offered further explanation described experiencing inconsistent effects from the medication.

3.2. Efficacy outcomes by treatment guess

Six-month primary and secondary efficacy outcomes by treatment guess are presented in Table 2. Treatment efficacy at six-months did not differ significantly by treatment guess regarding prior month number of days of street heroin use, total use of any street opioid (including heroin), crack cocaine use, and involvement in illegal activities. Retention in treatment of at least 28 days in the prior month and proportion of urine toxicology tests positive for street heroin markers were not significantly different by treatment guess. The frequency of physical and psychological health symptoms was the only outcome found to differ significantly by treatment guess. Those who guessed diacetylmorphine reported significantly better physical health symptom scores (mean =9.59; standard deviation [SD] = 7.58) compared to those who guessed hydromorphone (mean = 12.57; SD = 8.39) or were unsure (mean =12.61; SD = 8.51). Those guessing diacetylmorphine also reported significantly better psychological health symptom scores (mean = 6.98; SD = 7.30) compared to those who guessed hydromorphone (mean = 9.35; SD = 8.44) or were unsure (mean = 9.00; SD = 9.40).

Table 2. Six-month primary and secondary efficacy outcomes by treatment guess.

	Guessed DAM	Guessed HDM	Unsure
Outcome variable	n = 64	n = 95	n = 39
Prior 30 days street heroin use	4.33 ± 7.94	4.58 ± 8.82	3.33 ± 6.10
Positive for street heroin markers	15 (23.8%)	24 (26.4%)	10 (25.6%)
Prior 30 days any street opioid use^a	5.47 ± 9.02	5.36 ± 9.24	4.74 ± 7.55
Prior 30 days crack cocaine use	7.03 ± 11.23	8.14 ± 11.22	8.79 ± 12.14
Prior 30 days illegal activities	5.41 ± 10.44	2.76 ± 6.59	3.53 ± 8.53
MAP Physical health score^b^,^d^,^e	9.59 ± 7.58	12.57 ± 8.39	12.61 ± 8.51
MAP Psychological health score^b^,^d^,^e	6.98 ± 7.30	9.35 ± 8.44	9.00 ± 9.40
SALOME compliance^c	54 (84.4%)	76 (80.0%)	27 (69.2%)

DAM: diacetylmorphine; HDM: hydromorphone.

Data shown are n (%) and mean ± standard deviation. Pairwise comparisons were performed using: (1) chi-square test for positive for street heroin markers and compliance; (2) regression for physical and psychological health scores; or (3) negative binomial regression for days of illicit heroin, opiates, crack cocaine, and illegal activities for continuous variables. Continuous outcomes were adjusted by their baseline values using analysis of covariance.

Data not shown: Results from the analysis of statistical interaction between treatment guess and treatment assigned on outcomes revealed a significant interaction only for physical health score (p = .03). Specifically, physical health was significantly better among those randomized to HDM and guessing DAM (mean ± SD = 9.00 ± 7.95) compared to those randomized to HDM and guessing HDM (mean ± SD = 13.14 ± 8.59; p = .017). This interaction was not observed among those randomized to diacetylmorphine and guessing diacetylmorphine (mean ± SD = 10.03 ± 7.38) compared to those randomized to diacetylmorphine and guessing hydromorphone (mean ± SD = 11.88 ± 8.19; p = 1.00).

a Includes street use of heroin, morphine, hydromorphone, and speedball (combined street opioids and stimulants).

b Maudsley Addiction Profile: Physical and psychological health scores range from 0 to 40, higher scores indicate poorer health.

c Participants receiving at least 28 out of 30 days treatment based on clinical medication dispensation records at the six-month follow-up date.

d p < .05 for DAM vs. HDM.

e p < .05 for DAM vs. unsure.

Results from the analysis of statistical interaction between treatment guess and treatment assigned on outcomes revealed a significant interaction for physical health score (p=.03), with no other tested outcomes demonstrating a significant interaction. Specifically, physical health was significantly better among those randomized to hydromorphone and guessing diacetylmorphine (mean ± SD = 9.00 ± 7.95) compared to those randomized to hydromorphone and guessing hydromorphone (mean ± SD = 13.14 ± 8.59; p=.017). Among those randomized to diacetylmorphine, there was no significant difference in physical health score between those who guessed diacetylmorphine (mean ± SD = 10.03 ± 7.38) or hydromorphone (mean ± SD = 11.88 ± 8.19; p = 1.00).

3.3. Predictors of treatment guess

Table 3 shows socio-demographic, and treatment factors by treatment guess at six months. Socio-demographic and baseline characteristics were not significantly associated with guess, with the exception of age, where older participants were more likely to guess hydromorphone compared to diacetylmorphine. Participants guessing hydromorphone or diacetylmorphine early in the study were significantly more likely to repeat their guess at six months: 39 (55.7%) of those who chose diacetylmorphine early in the study did so at six months, 34 (57.6%) of those who chose hydromorphone did again, while only 9 (20.0%) continued to be unsure. Regarding the VAS score obtained early in the study, guessing diacetylmorphine was associated with higher scores in ‘drug liking’, ‘good-effect’, and ‘drug-related high’. Average dose of hydromorphone or diacetylmorphine received in the prior month and past use (before randomization) of street or prescribed hydromorphone was not associated with treatment guess. Number of related AEs was not associated with treatment guesses; however guessing diacetylmorphine, compared to hydromorphone, was more likely among those who had one or more observed and related drowsiness event. No relationships were found between related histamine reactions and treatment guess. Guessing hydromorphone was more likely among those with negative reasons for their treatment guess. Those who were more likely to be ‘unsure’ about their treatment assignment provided less neutral or positive comments for treatment guess and had less number of days in treatment, compared to those who guessed diacetylmorphine or hydromorphone.

Table 3. Summary statistics and unadjusted multinomial regression analysis of predictors of treatment guess at six months.

	Guessed DAM	Guessed HDM	Unsure
Variable	n = 64	n = 95	n = 39
Age^g	42.69 ± 10.60	45.82 ± 8.61	42.44 ± 9.83
Gender
Women	22 (35.5%)	26 (41.9%)	14 (22.6%)
Men	42 (30.9%)	69 (50.7%)	25 (18.4%)
Ancestry
Aboriginal	18 (29.5%)	29 (47.5%)	14 (23.0%)
Non-Aboriginal	46 (33.6%)	66 (48.2%)	25 (18.2%)
Randomization arm
DAM	35 (35.7%)	44 (44.9%)	19 (19.4%)
HDM	29 (29.0%)	51 (51.0%)	20 (20.0%)
Week 1 Guess:^a^,^g^,^h
DAM	39 (55.7%)	23 (32.9%)	8 (11.4%)
HDM	9 (15.0%)	34 (56.7%)	17 (28.3%)
Unsure	10 (22.2%)	26 (57.8%)	9 (20.0%)
Week 1 VAS score post-drug intake^b
Drug effect	63.43 ± 27.45	55.82 ± 31.13	52.79 ± 25.68
Drug liking^g	69.30 ± 30.14	57.75 ± 32.54	61.44 ± 32.47
Good effect^g	71.79 ± 27.51	61.20 ± 33.06	61.47 ± 31.42
Bad effect	9.29 ± 19.49	10.88 ± 21.83	10.88 ± 18.81
Drug-related high^g^,^h	59.60 ± 32.15	41.54 ± 35.98	40.94 ± 32.87
Sick	6.49 ± 19.23	6.75 ± 19.74	5.53 ± 13.25
Ever used prescribed or illicit HDM
Yes	58 (33.7%)	82 (47.7%)	32 (18.6%)
No	6 (23.1%)	13 (50.0%)	7 (26.9%)
Days in SALOME treatment from 0 to 6 months^h	169.89 ± 21.03	159.69 ± 42.62	142.92 ± 57.67
Days with SALOME treatment in prior 30 days	27.00 ± 7.59	25.52 ± 9.73	23.28 ± 11.39
Average dose received in prior 30 days	517.29 ± 249.68	472.02 ± 242.47	447.27 ± 327.35
Number of related AEs	2.81 ± 4.12	2.53 ± 5.17	2.74 ± 5.23
Number of related SAEs	0.14 ± 0.47	0.09 ± 0.41	0.23 ± 0.71
Had an observed drowsiness event related to the medication^c^,^g
Yes	30 (46.2%)	23 (35.4%)	12 (18.5%)
No	34 (25.6%)	72 (54.1%)	27 (20.3%)
Had an observed allergic reaction related to the medication^d
Yes	11 (25.6%)	25 (58.1%)	7 (16.3%)
No	53 (34.2%)	70 (45.2%)	32 (20.6%)
Any negative reason for perceived treatment assignment^e^,^g^,^i
Yes	15 (16.9%)	65 (73.0%)	9 (10.1%)
No	49 (45.0%)	30 (27.5%)	30 (27.5%)
Any neutral or positive reason for perceived treatment assignment^f^,^g^,^h^,^i
Yes	40 (56.3%)	26 (36.6%)	5 (7.0%)
No	24 (18.9%)	69 (54.3%)	34 (26.8%)

DAM: diacetylmorphine; HDM: hydromorphone; VAS: Visual Analog Score; AEs: adverse events; SAEs: serious adverse events.

Data shown are n (%) and mean ± standard deviation. Significance testing from the unadjusted multinomial logistic regression p<.05.

a Treatment assignment guesses from post-titration/early in the study.

b Visual analog scale (VAS) for participant’s subjective evaluation of drug-related high score post-drug intake collected post-titration/early in the study by a trained peer-interviewer at the clinic. Completed by 175 participants.

c Related drowsiness events defined and assessed objectively by clinic nurses during the post-drug intake assessments. A related drowsiness event includes mild (easy to arouse, does NOT require intervention, able to pass post-assessment), moderate (failure of first post-assessment) and severe (requiring any treatment or intervention, such as oxygen, naloxone).

d Related allergic reactions were defined and assessed objectively by clinic nurses during the post-drug intake assessments. A related allergic reaction includes any histamine reaction localized to the application site or reactions occurring immediately post-injection.

e Participants providing an open-ended reason for perceived treatment assignment in which the discourse referenced negative or undesirable aspects of the medication. Specific themes collapsed into this category included the presence of negative drug effect or absence of a positive drug effect. Note that one participant’s reason for treatment assignment guess may have been coded at both negative and neutral. For example, seven participants’ reasons were coded at both the presence of negative and neutral drug effects.

f Participants providing an open-ended reason for perceived treatment assignment in which the discourse was neutral or positive in nature. Specific themes collapsed into this category included the presence of positive or neutral drug effects, or absence of negative or neutral drug effects.

g Significance testing from the unadjusted multinomial logistic regression p < 0.05 for DAM vs. HDM.

h Significance testing from the unadjusted multinomial logistic regression p < 0.05 for DAM vs. unsure.

i Significance testing from the unadjusted multinomial logistic regression p < 0.05 for HDM vs. unsure.

Table 4 shows the adjusted model with factors independently associated with treatment guess. Compared to hydromorphone, the odds of guessing diacetylmorphine were significantly higher with each unit increase in VAS ‘drug-related high’ score and for those with at least one observed drowsiness event at the clinic after injecting the medications. Compared to diacetylmorphine and unsure, the odds of guessing hydromorphone were higher among those participants who provided negative comments as reasons for their guesses. Finally, the odds of guessing diacetylmorphine or hydromorphone, compared to unsure, were higher among those who made any positive or neutral comment regarding their treatment guess.

Table 4. Multivariable multinomial logistic regression models for predictors of treatment guess at six months.

	DAM vs. HDM	DAM vs. unsure	HDM vs. unsure
Predictor	OR (95% CI)	OR (95% CI)	OR (95% CI)
Week 1 VAS drug-related high score	1.02 (1.00, 1.03)^a	1.01 (1.00, 1.03)	1.00 (0.98, 1.01)
Had an observed drowsiness event related to the medication:
Yes vs. no	3.32 (1.44, 7.97)^a	1.81 (0.60, 5.68)	0.54 (0.18, 1.65)
Any negative reason for perceived treatment assignment:
Yes vs. no	0.14 (0.06, 0.31)^a	2.52 (0.75, 9.29)	18.30 (6.12, 64.33)^a
Any neutral or positive reason for perceived treatment assignment:
Yes vs. no	1.91 (0.68, 5.84)	34.77 (10.8 132.54)^a	18.22 (6.02, 65.39)^a

DAM: diacetylmorphine; HDM: hydromorphone; OR: odds ratio; 95% CI: 95% confidence interval; VAS: Visual Analog Score.

All models were fit using Firth’s Penalized Likelihood. Data shown are OR (95% confidence intervals) for the odds of guessing DAM vs. HDM, DAM vs. Unsure or HDM vs. Unsure.

a Indicates a significant association (p<.05) based on the multivariable multinomial logistic regression model, which was built on 175 observations (since the VAS was completed by 175 participants and entered the model).

4. Discussion

The present study tested if perceived treatment assignment was associated with treatment outcomes among participants receiving double-blind injectable hydromorphone or diacetylmorphine. It also identified predictors of perceived treatment assignment, including reasons for participants’ guesses. Among primary and secondary study outcomes, perceived treatment assignment was associated with physical and psychological health scores, but not with street opioid use, crack cocaine use, retention in treatment or illegal activities. Participants were more likely to expect they were receiving diacetylmorphine compared to hydromorphone if they had an observed drowsiness event related to the medication, if they gave a higher rating score for ‘drug-related high’ and if they did not provide a negative reason for their treatment expectancy.

In the parent trial, analysis by randomization arm showed that hydromorphone did not significantly differ from diacetylmorphine in the main outcomes tested in this study (Oviedo-Joekes et al. 2016). However, when these outcomes were analyzed by treatment expectancy, physical and mental health scores were better among those who thought they were receiving diacetylmorphine compared to hydromorphone. These results suggest that self-reported health outcomes might have been affected by guess. Furthermore, the testing of an interaction between treatment guess and treatment assigned on the outcome of physical health revealed that those randomized to hydromorphone and guessing diacetylmorphine had significantly better physical health compared to those randomized to hydromorphone and guessing hydromorphone. In SALOME, hydromorphone had significantly less related AEs compared to diacetylmorphine. Further, diacetylmorphine was the medication that participants wished to receive (not hydromorphone). This association may arise in response to two different mechanisms: (a) participants who felt better physically and mentally (due to the treatment or not) might have attributed this improvement to diacetylmorphine (the preferred treatment), or (b) those who believed they were receiving diacetylmorphine may have expected and then experienced health improvements (Colagiuri & Boakes 2010). While the design of our randomized clinical trial cannot differentiate between these two mechanisms, they are not necessarily mutually exclusive. If a participant feels physically better, that might make him or her think he or she is receiving diacetylmorphine, and thus (due to the expectations attributed to the treatment) experience more improvement. Little evidence is available to disentangle these two mechanisms or suggest other potential explanations. However, recent experimental studies suggest that it is more likely that a participant’s treatment allocation beliefs and the expectations associated with these beliefs affect their symptoms or responses, rather than the other way around (Colagiuri & Boakes 2010; Brown et al. 2014). In making their treatment guesses, participants might have taken clues from the environment or from their symptoms and assumed they were receiving diacetylmorphine or hydromorphone based on their expectations.

The analysis of factors that independently explained participants’ guesses provides some insight into participants’ expectations of both diacetylmorphine and hydromorphone. Overall, expecting hydromorphone was more likely when participants did not like the effect of the drug, or did not feel as well as they seemed to have expected, likely due to their prior experience with injecting street heroin. When analyzing the specific perceived drug effects, the participants used to conclude what drug they were receiving, there were no differences between the expected treatments as well as the actual treatments received. For example, feeling ‘pins and needles’ was mentioned as an undesirable or neutral effect for either medication, by treatment arm and by treatment expectancy. Diacetylmorphine was the drug participants wished to receive and had more positive expectations associated with it. As such, it is possible that when participants were bothered by a side effect such as histamine reactions, they attributed it to the drug they preferred less or for which they had lower expectations (i.e. hydromorphone).

Participants who rated the ‘drug-related high’ higher and who had an observed drowsiness event related to the medication were more likely to think they were receiving diacetylmorphine. Information about the study medications provided to participants in the consent form listed over-sedation as a possible side effect of both drugs. During the study period however, diacetylmorphine had significantly more related drowsiness events than hydromorphone (Oviedo-Joekes et al. 2016). There was no prior evidence that diacetylmorphine would have significantly more drowsiness than hydromorphone, therefore, it is possible that participants guessed right (Boutron et al. 2005). Moreover, these drowsiness events were registered at the clinic, and were defined and assessed objectively by clinic nurses who were blinded to the study medications. The design of this study does not allow for conclusions regarding whether the participants ‘expected’ diacetylmorphine to induce more over-sedation compared to hydromorphone, or how this side effect was perceived (e.g. desired or indifferent). However, it is important to note that when asked why participants thought they were receiving either drug, few mentioned symptoms of over-sedation and their answers were not related to treatment guess. On the other hand, desire to experience the drug-related high is considered a determining factor for using street heroin (Cicero et al. 2014). In this sense, this finding adds to the possibility that positive expectations were associated with diacetylmorphine compared to hydromorphone.

Treatment expectancy can refer to either expectancy of efficacy or expectancy of allocation, and both perceptions interact giving an overall perception of treatment assignment (Colagiuri 2010). In the case of our study, the majority of participants thought that the treatment was effective (179 out of the 196), and most of the primary and secondary outcomes did not differ by treatment expectancy. However, physical and mental health scores were better among those who believed they were receiving diacetylmorphine compared to those who thought they were receiving hydromorphone or were unsure. Moreover, an important minority of participants reported being unsure which treatment they were receiving. These findings bring forward noteworthy differences between the double-blind randomized controlled trial and standard clinical practice. Specifically, research on the role of expectancies has theorized that independent of treatment arm, the association between perceived treatment assignment and outcomes potentially underestimates the effect of treatment when delivered in standard clinical practice, since participants are aware of which treatment they are receiving (Colagiuri et al. 2009; Colagiuri 2010). Double-blind studies will not necessarily predict the treatment effect in actual clinical practice, where treatments are offered open-label (i.e. the patient knows what she is receiving). However, most of the scientific literature on treatment expectancy in double-blind clinical trials comes from placebo-controlled superiority studies. The present study is unique because it was an active-control (where an effect is expected from both medications) non-inferiority study, where there is an assumption than both treatments are somehow effective.

The analysis of treatment expectancies and participants perceptions can be used to complement the findings of double blind clinical trials and to inform clinical practice. While the effect of diacetylmorphine might have been underestimated under a double-blind design, participants’ reasons for treatment expectancy could be used to support treatment decision-making and positively affect the treatment planning process (Crow et al. 1999). For example, in the context of opioid maintenance treatment, increasing knowledge about the inter-individual variability in patients’ response to opioids and what can be expected may help manage undesirable side effects. While hydromorphone and diacetylmorphine seem to be very similar, our clinical trial showed that there were significant differences at least in the frequency of side effects when these medications were prescribed for opioid substitution. Although in different study populations, these findings correspond with laboratory studies (Walker & Zacny 1999; Walker et al. 2001) as well as with the ample literature in palliative care showing that patients respond differently to similar opioids and side effects may improve with opioid rotation (Drewes et al. 2013). Thus, integrating patients’ beliefs regarding opioids in opioid maintenance treatment and supporting patients to achieve accurate expectations might help patients and physicians to reach informed and patient-centered decisions (Mead & Bower 2000; Marchand et al. 2011; Trujols et al. 2012; Oviedo-Joekes et al. 2015).

5. Conclusions

The present study provides evidence of the clues that participants used to make their treatment guesses and the potential impact these beliefs have on health outcomes. Understanding these clues (e.g. histamine reactions) and relating them to patients’ expectations (e.g. better health) could be integrated with accurate information about the treatment and provide an opportunity for patient–physician shared decision making in opioid maintenance treatment. Further research on treatment expectations and how these affect outcomes is needed, as this has implications for clinical practice. Such research would also help the interpretation of clinical trial outcomes, whose design does not allow for the direction of the relationship between treatment guess and treatment outcomes to be disentangled or for underlying mechanisms to be explored.

Acknowledgements

First and foremost, we would like to acknowledge the contribution and commitment of the study participants who made it possible to continue advancing this research while overcoming its many challenges. Also, at Providence Health Care, Justin Karasik and the communications team; Julie Foreman and the clinical team at Providence Crosstown Clinic; Amin Janmohamed and the pharmaceutical team at Providence Crosstown Clinic. Finally, we wish to acknowledge all members of the Community Advisory Board and Data Safety Monitoring Board, staff of the Centre for Health Evaluation and Outcome Sciences and Salima Jutha, the SALOME investigators and research team.

Disclosure statement

The authors report no conflicts of interest.

Additional information

Funding

Canada Research Chairs, 10.13039/501100001804; Canadian Institutes of Health Research, 10.13039/501100000024 [MCT ? 103817]; Michael Smith Foundation for Health Research, 10.13039/501100000245; Providence Health Care; Vancouver Coastal Health; InnerChange Foundation; Providence Health Care Research Institute; St. Paul's Hospital Foundation.