Original Paper

251

Author

M. I. Saleh

Affiliation

Faculty of Pharmacy, The University of Jordan, Amman, Jordan

Key words ▶ opioid ● ▶ detoxification ● ▶ predictors ● ▶ opiates ● ▶ buprenorphine ● ▶ modeling ● ▶ longitudinal study ●

Abstract

received 06.03.2014 revised 18.08.2014 accepted 03.09.2014 Bibliography DOI http://dx.doi.org/ 10.1055/s-0034-1390467 Published online: October 16, 2014 Pharmacopsychiatry 2014; 47: 251–258 © Georg Thieme Verlag KG Stuttgart · New York ISSN 0176-3679 Correspondence M. I. Saleh, PhD Faculty of Pharmacy The University of Jordan Amman 11942 Jordan [email protected]

▼

Introduction:  The present analysis describes the longitudinal change in buprenorphine treat­ ment outcome. It also examines several par­ ticipant characteristics to predict response to buprenorphine. Methods:  Participants (n = 501, age > 15 years) received buprenorphine/naloxone treatment for 4 weeks, and then were randomly assigned to undergo dose tapering over either 7 days or 28 days. An empirical model was developed to des cribe the longitudinal changes in treatment out­ come. Several patient characteristics were also examined as possible factors influencing treat­ ment outcome. Results:  We have developed a model that cap­ tures the general behavior of the longitudinal change in the probability of having an opioidnegative urine sample following buprenorphine treatment. The model captures both the initial increase (i. e., initial response) and the subse­ quent decrease (i. e., relapse to opioid) in the

Introduction

▼

The use and abuse of opioids, including heroin and prescription pain medication, is an enduring public health problem. Globally, opioid use has an estimated prevalence of 0.6–0.8 % in 2010 (between 26 million and 36 million users) [1]. In the USA, results of the 2011 National Survey on Drug Use and Health (NSDUH) showed that an estimated 22.5 million Americans aged 12 years or older, were current or past month illicit drug users. The survey showed that 281 000 Ameri­ cans aged 12 years or older used heroin in 2011 [2]. One of the commonly used FDA-approved medi­ cations for opioid dependence, second to metha­ done, is buprenorphine [3]. Buprenorphine has

likelihood of providing an opioid-negative urine sample. Characteristics associated with success­ ful buprenorphine treatment outcome include: having a negative urine test for drugs, hav­ ing alcohol problems [assessed using alcohol domain of addiction severity index (ASI-alco­ hol)] at screening, being older, and receiving low cumulative buprenorphine dose. However, ASIalcohol values were generally low which make the application of the proposed alcohol effect for patients with more severe alcohol problems questionable. Conclusions:  A novel approach for analyzing buprenorphine treatment outcome is presented in this manuscript. This approach describes the longitudinal change in the probability of provid­ ing an opioid-free urine sample instead of con­ sidering opioid use outcome at a single time point. Additionally, this model successfully describes relapse to opioid. Finally, several patient characteristics are identified as predic­ tors of treatment outcome.

many characteristics that make it an excellent agent for opioid addiction treatment. Since buprenorphine is a partial mu opiate receptor agonist, it is associated with a low incidence of respiratory depression when increased doses of buprenorphine are used [4]. Second, using bupre­ norphine does not usually result in overdose [4]. Finally, interdosing interval of buprenorphine can be extended by doubling or tripling the dose with­ out causing toxicity. This feature can be attributed to the fact that larger doses do not enhance buprenorphine’s agonist activity, but they do extend its duration of action [5]. Studies of predictors of response for buprenor­ phine treatment evaluated outcome at various endpoints. Dreifuss et. al. examined several par­ ticipant characteristics as possible predictors of

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

Modeling Longitudinal Changes in Buprenorphine Treatment Outcome for Opioid Dependence

successful outcome at the end of a 12 week buprenorphine/ naloxone treatment [6]. Another research group identified sev­ eral predictors (e. g., medication type, level of care, opioid with­ drawal severity, baseline anxiety symptoms, and tobacco use) of successful outcome at the end of a 13 day buprenorphine/nalox­ one or clonidine regimen [7]. Recently, Hillhouse et. al. exam­ ined participant characteristics associated with success at the end of a 4 week stabilization with buprenorphine [8]. Despite the fact that these studies identified several participant charac­ teristics as possible predictors for buprenorphine treatment outcome, they examined the outcome at a single pre-specified time point. None of these studies explored the longitudinal change in treatment outcome. The objective of the present study was 2-fold: (i) to describe lon­ gitudinal change in buprenorphine treatment outcome, evalu­ ated by opioid urine sample, using a proposed model and (ii) to identify participant characteristics that affect response to buprenorphine.

opioid dependence. Key exclusion criteria included any of the following: having a medical condition that would make partici­ pation medically hazardous; having a known allergy or sensitiv­ ity to buprenorphine or naloxone; having an acute severe psychiatric condition in need of immediate treatment; having dependence on alcohol, other depressants, or stimulants, requir­ ing immediate medical attention; having a current pattern of benzodiazepine use; having participated in an investigational drug study, including buprenorphine, within the past 30 days prior to screening; have had methadone or levo-alpha acetyl methadol (LAAM) maintenance or detoxification within 30 days of enrollment; having a pending legal action that could prohibit or interfere with participation; being unable to remain in area for the duration of treatment; being pregnant, lactating, or plan­ ning to become pregnant; having a positive urine sample for methadone and benzodiazepine immediately preceding buprenorphine/naloxone induction; or seeking long-term (greater than 2 months) opioid maintenance treatment (see [10] for further details).

Patients and Methods

Measures

▼

Main study objectives and design

The primary objectives of this clinical study was to compare the relative advantage of 2 rates of buprenorphine/naloxone taper­ ing following a 4 week period of flexible dose stabilization, as reflected by the proportion of participants providing opioid free urine at the end of the taper regimen. Bupreorphine was given in the form of a sublingual combination tablet containing both buprenorphine and naloxone (Suboxone®). Ling et al. hypothe­ sized that a longer taper schedule would result in a higher per­ centage of participants providing opioid-free urine samples at the end of the taper. This hypothesis was based on the finding that the longer taper schedule resulted in more favorable opioid addiction outcomes, as indicated by withdrawal symptoms and opioid-free urines across the taper period [9]. This is a randomized, parallel-group, open-label study. All par­ ticipants underwent an induction phase of 3 days. During the induction phase, buprenorphine dose did not to exceed 8 mg on the first day, 12 mg on the second day and 16 mg on the third day. Following induction, there was a 25 day stabilization period on Suboxone®. Until the beginning of the final stabilization week doses were allowed to be adjusted based on clinical need. Doses were adjusted to a maximum dose of 8, 16, or 24 mg. During the final stabilization week, participants received a constant buprenorphine dose of 8, 16, or 24 mg. Participants were then randomized to one of 2 taper schedules (7 or 28 days). Randomi­ zation was stratified according to buprenorphine dose taken during the last week of stabilization. Participants were followed up weekly over the first 8 weeks post randomization. All partici­ pants were scheduled to have a weekly visit over the first 8 weeks post randomization, one visit 3 months post randomiza­ tion and a final visit 3 months post taper. The primary objective was addressed by Ling et al. The author concluded that buprinorphine dose tapering over 28 days did not provide apparent advantages compared to tapering over 7 days [10].

Study population

Participants (n = 501) met DSM-IV criteria for current opioid dependence were at least 15 years old, provided a positive urine sample for opioids at screening, and were seeking treatment for

The Addiction Severity Index (ASI) is a standardized score to eval­ uate severity profiles in several areas commonly affected by sub­ stance abuse [11]. The ASI covers the following domains: medical, psychiatric, drug and alcohol use, legal, and family/ social employment/support. In the present analysis, the score of each domain of ASI was tested as a possible predictor, separately. The score of each ASI domain was normalized to have a possible range of 0–1. ASI scores were calculated as described previously [12]. The data obtained to calculate the ASI were about problem behavior within the previous 30 days. ASI was administered at screening. The Clinical Opiate Withdrawal Scale (COWS) is an 11-item inter­ viewer administered questionnaire of observable signs and symptoms of opioid withdrawal [13]. COWS was evaluated at screening, at the end of stabilization phase, and at the end of dose tapering phase. The Adjective Rating Scale for Withdrawal (ARSW) is a 16-item questionnaire of signs and symptoms of opioid withdrawal [14, 15]. Each item is rated from 0 (none) to 9 (severe) by the study subject, and is based on his or her subjective withdrawal discomfort [16]. ARSW questionnaire was completed by the par­ ticipant at screening, at the end of stabilization phase, and at the end of dose tapering phase. Visual Analog Scale (VAS) is a 3-item self-report measure that assesses the degree to which the participant experiences any craving for opioids, the severity of withdrawal symptoms, and the extent to which the study medication helps to alleviate crav­ ings (if applicable). Each item is rated from 0 (not at all) to 100 (extremely) by the participant. In the present analysis, the aver­ age score of the 3 items was examined as a predictor of the out­ come. The VAS questionnaire was completed by the participant at screening, at the end of stabilization phase, and at the end of dose tapering phase. Dosing information. Buprenorphine cumulative dose adminis­ tered until the end of stabilization phase was evaluated as potential predictors of response. Demographic information assembled included age, gender, eth­ nicity, employment, and marital status. Demographic character­ istics were collected at screening. Toxicology testing was conducted qualitatively for the following: cocaine, amphetamines, barbiturates, benzodiazepines, meth­

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

252 Original Paper

Original Paper

Data sources

The information reported here results from secondary analyses of data from clinical trials conducted as part of the National Drug Abuse Treatment Clinical Trials Network (CTN) sponsored by National Institute on Drug Abuse (NIDA). Specifically, data from CTN-0003 [Study title: Suboxone (Buprenorphine/Naloxone) Taper: A Comparison of 2 Schedules] were included. CTN data­ bases and information are available at “www.ctndatashare.org”.

Longitudinal change in opioid use model

An empirical model was used to describe longitudinal changes in buprenorphine treatment outcome. Let P(t) denote the prob­ ability of having a successful buprenorphine treatment outcome, defined as having an opioid-negative urine sample, at time t. As depicted in Eq. 1, a logit transformation of P(t) was applied in order to transform P(t) domain from (0,1) to ( − ∞,  + ∞). The transformed values were used in the model to ensure that model predicted probabilities lie in (0,1). ⎛ P ⎞ R = In ⎜ +  (1) ⎝ 1 − P ⎟⎠ where R is the logit form and α is a parameter included in the logit transformation in order to restrict the domain of R to be positive. The change in the value of R overtime is described according to Eq. 2: dR / dt = K in (1 + e − Kt ) − K out ⋅ M , R(0) =  (2) where Kin is a zero-order constant that govern the increase in the value of R, K is a first-order constant for the initial rapid increase in the value of R (i. e., initial increase in the probability of having a successful treatment outcome), Kout is a first-order constant that govern the decrease in the value of R, M is a modulator to counterbalance the input of R, and β is the initial value for R. The right-hand side of Eq. 2 can be broken down into 2 compo­ nents. The first component (Kin(1 + e − Kt)) represents an input term. The input term describes the increase in the value of R (i. e., increase in the probability of having a successful treatment outcome). The second component (Kout · M) represents a loss term. This term counterbalances the production of R. A new variable, M, was introduced to account for the relapse observed following a successful treatment outcome. The change in the value of M over time is described according to Eq. 3: dM / dt = K rel ⋅ R  K rel ⋅ M , M (0) = 0

(3)

where Krel is a first-order constant that accounts for the relapse observed following a successful treatment outcome. When R increases, as a result of initiating buprenorphine treatment, the value of the modulator, M, will also increase. The increase in M counterbalances R by increasing the rate of loss of R. The increase in the rate of loss of R accounts for the relapse observed among patients treated with buprenorphine.

Patient characteristics that influence treatment outcome

To assess whether patient characteristics influenced the opioid ▶  Table 1), the parameter-covariate relation­ urine test result ( ● ship was explored. First, the model described above was fitted to the data without including parameter-covariate relationship. ▶  Table 2) were included in the mixed-effects Second, covariates ( ● model individually. Third, covariates that resulted in a drop in NONMEM objective function, equal to twice the negative loglikelihood of the data, of more than 3.84 were included together in the mixed effects model to describe parameter-covariate rela­ tionship. Because the change in the objective function of the NONMEM value is approximately chi-square distributed with a degree of freedom of 1, a difference in the NONMEM objective function value of 3.84 is associated with a P-value of less than 0.05 [17, 18]. Finally, several trials of the model formulated in step 3 were conducted. With each trial one of the covariates was excluded. The covariates that did not result in an increase of

Table 1  Percentage of participants with opioid-free urine test at each evaluation time-point. Time-point

screening induction/stabilization (week 0) induction/stabilization (week 1) induction/stabilization (week 2) induction/stabilization (week 3) induction/stabilization (week 4) end of dose tapering 1 month post taper 3 months post randomization 3 months post taper

Percentage of

Number of

participants with

participants

opioid-free urine test

examined

0 % 2 % 58 % 61 % 64 % 62 % 51 % 34 % 33 % 31 %

501 501 490 491 483 498 357 284 185 237

Table 2  Summary of covariates tested for their impact on the longitudinal change in buprenorphine treatment outcome. Categorical patient

Continuous patient characteristics

characteristics Race – white –  African American – hispanic – Multiracial –  Another race Marital status – married –  never married – divorced – separated – widowed Gender – female – male Taper group –  7 days taper –  28 days taper Drug urine test result at screening – positive – negative

Age The Addiction Severity Index (ASI) at screening with the following domains: – medical – employment –  drug use –  alcohol use – legal – family/social – psychiatric Visual Analog Scales (VAS) at: – screening –  end of stabilization –  end of taper Clinical Opiate Withdrawal Scale (COWS) at: – screening –  end of stabilization –  end of taper Adjective Rating Scale for Withdrawal (ARSW) at: – screening –  end of stabilization –  end of taper Cumulative buprenorphine dose

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

amphetamines, phencyclidine (PCP), marijuana, and tricyclic antidepressants. Results of urine drug toxicology at screening and at the end of taper were investigated as a predictor of the outcome. Buprenorphine treatment outcome was evaluated based on urine test results for the following opioids: morphine, methadone, and oxycontin.

253

254 Original Paper

Inter-individual model

A normal distribution was used to describe parameters distribu­ tion:

 i =  + i i ∼ N (0, * 2 )

(4)

where θi denotes the ith individual’s parameter value and is a function of θ, the population value of the parameter, and ηi is the individual random effect (a zero-mean random variable with the variance ω2), which accounts for the difference between the population parameter value and the individual value.

Data analysis and model evaluation

Modeling was performed using the NONMEM computer pro­ gram [1] Version 6 with the COND LAPLACE LIKELIHOOD method to fit the data. NONMEM uses mixed (fixed and random) effects regression to estimate population means and variances of the parameters and to identify factors that influence treatment out­ come. The stability and the performance of the final model were validated by the bootstrap method. 500 data sets were recon­ structed by re-sampling from the original data using the Perlspeaks-NONMEM (PsN) Toolkit Version 3.6.2 [19, 20]. The final model was fitted repeatedly to the 500 bootstrapped samples and parameters were calculated for each dataset.

Results

▼

Study sample characteristics

The age range was 18.3–71.1 years with a sample mean of 35.9 years (standard deviation, 10.5); 70 % were White, 11 % were African American, 7 % were Hispanic, 9 % were Multiracial, and 3 % were of another race; 33 % were females, and 67 % were males. Regarding the marital status, 24 % were married, 51 % were never married, 17 % were divorced, 6 % were separated, and 2 % were widowed.

predictors and various parameters of the model described ear­ lier was described by Eqs. 5–7. The model parameters are sum­ marized in ●  ▶  Table 3. K in = K in0 + B1 ⋅ ASI Alcohol + B2 ⋅ Age 0 K out = K Out + B3 ⋅ Drug



(5) (6)

(7)  =  0 + B4 ⋅ Dose were B1 is the regression coefficient for the effect of ASIAlcohol on Kin, B2 is the regression coefficient for the effect of age on Kin, B3 is the regression coefficient for the effect of drug urine test on Kout, B4 is the regression coefficient for the effect of cumulative buprenorphine dose on Kout, ASIAlcohol is the ASI score value (alco­ hol index), Age is the age in years, Drug is the urine drug test result (1 for positive, 0 otherwise), and Dose is the cumulative dose of buprenorphine administered over the study period (in mg). Older participants tend to have a better treatment outcome. A comparison between 2 participant groups is presented in ●  ▶  Fig. 2. The early age group represents participants with age lower than the 25th percentile of the age value computed for all par­ ticipants. The late age group represents participants older than the 75th percentile of the age value for the investigated popula­ tion. The observed likelihood of having an opioid negative urine sample for older age group is consistently higher than the early age group. The influence of alcohol use at screening on treatment outcome is presented in ●  ▶  Fig. 2. 2 groups of participants were selected based on their value of ASI-alcohol score. The first group repre­ sents participants with low ASI-alcohol values (e. g., participants with modest alcohol addiction problems) denoted by low ASIalcohol group. The ASI-alcohol value for this group is below the 25th percentile for the ASI-alcohol values. The second group rep­ resents participants with high ASI-alcohol values (e. g., partici­ pants with severe alcohol addiction problems). This group is referred to as high ASI-alcohol group. Participants in this group have an ASI-alcohol value above the 75th percentile for the ASIalcohol. As presented in ●  ▶  Fig. 2, the high ASI-alcohol group has

Longitudinal changes in opioid addiction status model

0.8 Probability of opioid-negative urine test results

A summary of percentage of participants with opioid-free urine samples at each evaluation time point is presented in ●  ▶  Table 1. The longitudinal values of the probability of having an opioidnegative urine sample is presented in ●  ▶  Fig. 1. The model cap­ tures the general behavior of the longitudinal change in the probability of having an opioid-negative urine sample. Identi­ fied predictors of outcome include: age {39.9 (18.3–71.1) years, [mean (range)]}, buprenorphine cumulative dose [536.6 (56– 772) mg], drug urine test result at screening (293 positive and 208 negative), and alcohol use status (ASI score) at screening [0.06 (0–0.93)]. Drug urine tests include the following: cocaine, amphetamines, barbiturates, benzodiazepines, methampheta­ mines, phencyclidine (PCP), marijuana, and tricyclic antidepres­ sants. Participant drop out or failure to attend an evaluation appointment resulted in missing outcome measure at certain evaluation points. Using non-linear mixed effects does not require all participants to have the same evaluation time-points. As a result, missing outcome measures were handled naturally. ▶  Table 2) were handled by using However, missing covariates ( ● the average value. The relation between each of these selected



0.7 0.6 0.5 0.4 0.3 0.2 0.1

0

50 100 Time following the initiation of buprenorphine therapy (days)

150

Observed probability Predicted probability 95% confidence interval of the predicted probability

Fig. 1  Observed and predicted (with 95 % confidence interval) probability of successful buprenorphine treatment outcome.

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

more than 3.84 in the objective function upon removal were excluded from the final model.

Original Paper

255

Table 3  Final parameter estimates of the proposed model that describe longitudinal changes in buprenorphine treatment outcome. Parameter, unit

Description

Population typical

0 , 1/day K in K, 1/day

a zero-order constant that govern the increase in the value of R *  *  a first-order constant for the initial rapid increase in the value of R (i. e., initial increase in the probability of ­having a successful treatment outcome) a first-order constant that govern the decrease in the value of R a first-order constant that accounts for the relapse observed following a successful treatment outcome a parameter included in the logit transformation in order to restrict the domain of R to be positive the initial value for R A parameters that describe the influence of ASIAlcohol on Kin A parameters that describe the influence of age on Kin A parameters that describe the influence of drug urine test result on Kout A parameters that describe the influence of cumulative buprenorphine dose on α A parameters that describe the interindividual variability (standard deviation) of Kin

value (RSE) * 

0.075 (10 %) 1.03 (17 %)  − 6.28 (6 %) 5 (fixed) 0.236 (42 %) 0.003 (36 %) 0.012 (26 %)  − 0.003 (21 %) 0.030 (18 %)

 * RSE (i. e., relative standard error) was calculated using bootstrap  *  * R is the logit form of the probability of having a successful buprenorphine treatment outcome

Probability of opioid-negative urine test results

Age 0.8

0.6

0.6

0.4

0.4

0.2

0.2

0

0

50

100

150

0

Late age group (predicted) Early age group (predicted) Late age group (observed) Early age group (observed)

Drug urine at screening

0.8

0.8 0.6

0.4

0.4

0.2

0.2 0

50

0

50

100

Buprenorphine cumulative dose

0 150 100 0 50 Time following the initiation of buprenorphine therapy (days)

+ve urine drug group (predicted) –ve urine drug group (predicted) +ve urine drug group (observed) – ve urine drug group (observed)

150

High ASI-Alcohol group (predicted) Low ASI-Alcohol group (predicted) High ASI-Alcohol group (observed) Low ASI-Alcohol group (observed)

0.6

0

Fig. 2  The influence of various identified predictors on the longitudinal change in buprenorphine treatment outcome.

Addiction severity index (Alcohol)

0.8

100

150

High buprenorphine dose group (predicted) Low buprenorphine dose group (predicted) High buprenorphine dose group (observed) Low buprenorphine dose group (observed)

a higher likelihood of having a favorable outcome. This remark is consistent for observations starting from the beginning of buprenorphine administration (i. e., day zero) until the end of the study. Increasing the exposure to buprenorphine has been associated with the worst treatment outcome. ●  ▶  Fig. 2 represents a com­ parison between low buprenorphine dose and high buprenor­ phine dose groups. Low buprenorphine dose and high buprenorphine dose groups represent participants with cumu­ lative buprenorphine dose below and above the 25th and the 75th percentiles of cumulative buprenorphine dose, respectively. High buprenorphine dose groups consistently have a lower like­ lihood of having a favorable opioid use outcome. The trend observed in ●  ▶  Fig. 2 highlights the proposed effect of buprenor­ phine dose on buprenorphine treatment outcome. In this analysis, participants with drug positive urine sample at screening were associated with less favorable outcome. This observation is demonstrated by the increased likelihood of hav­ ing an opioid negative urine sample for participants with a nega­ ▶  Fig. 2). tive drug urine at screening ( ●

Discussion

▼

We have developed an empirical model that describes the longi­ tudinal change in buprenorphine treatment outcome. We have also explored factors predicting the time course of buprenor­ phine treatment outcome. Our analysis indicated that the likeli­ hood of successful treatment outcome is increased for participants who presented with a negative urine test for drugs at screening, presented with alcohol problems (assessed by ASIalcohol score) at screening, were older, and/or received lower cumulative buprenorphine doses. However, ASI-alcohol values were generally low which make the application of the proposed alcohol effect for patients with more severe alcohol problems questionable. Opioid treatment outcome changed dynamically following pharmacological detoxification with buprenorphine. At screen­ ing, all participants provided a positive urine sample for opioids. Following detoxification, a fraction of participants provided a successful treatment outcome. Later, a fraction of patients with successful treatment outcome relapsed and provided a urine

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

Kout, 1/day Krel, 1/day α β B1, 1/day B2, 1/(year.day) B3, 1/day B4, 1/mg BSV_Kin

0.428 (12 %) 0.047 (14 %)

256 Original Paper

K rel ⋅ R = K rel ⋅ M R=M



(8) (9)

Similarly, the rate of change in R equal zero: K in (1 + e − Kt ) − K out ⋅ M = 0

(10) At the end of treatment time approaches infinity, as a result: e − Kt = 0

(11)

From Eqs. 9–11, the value of R at the end of study is: K in K out

(12) From Eq. 1 and Eq. 12, the odds off having a successful treatment R=

outcome ⎛⎜ P ⎞⎟ ⎝1− P ⎠

become:

⎛ K ⎞ P = exp(R − ) = exp ⎜ in − ⎟ 1− P K ⎝ out ⎠

(13)



Using Eqs. 5–7, the influence of covariate can be incorporated into Eq. 14 to give: ⎛ K 0 + B1 ⋅ ASI Alcohol + B2 ⋅ Age ⎞ P = exp ⎜ in − 0 + B4 ⋅ Dose⎟ 0 1− P + B3 ⋅ Drug K Out ⎠ ⎝

(14)

We used Eq. 14 to predict the influence of each covariate on the odds of having a sustained remission as presented below. The finding that being older significantly predicted better treat­ ment outcomes is consistent with previous reports [25, 26]. The association between age and pharmacological detoxification outcome was also reported with other opioid dependence med­ ications such as methadone [27]. Using model estimated param­ ▶  Table 3), the longitudinal regression model predicts 40 % eters ( ● increase in odds of success, at the end of the study, with every 10-year increase in participant age. Participants with more problematic alcohol use at screening are predicted to have better outcomes. Alcohol use status was esti­ mated using the alcohol domain of the ASI score. Surprisingly, previous reports showed more frequent alcohol use was predic­ tive of poorer treatment outcomes [25, 28]. A possible explana­ tion of this inconsistency is that the ASI-alcohol score themselves were generally low (mean of 0.06 and inter-quartile range of [0–0.08] out of a possible range of 0–1). Overall, the limited range of the ASI-alcohol score in this trial is insufficient to address the influence of alcohol. Drug urine test result (at screening) was identified as a strong predictor of treatment success. Previous drug use in general has previously been associated with poorer opioid withdrawal out­ come [8, 29]. Adjusting for other predictors included in the final model, the odds of having a successful outcome for a participant with a negative drug urine test result at screening is 2.16 times that of a participant who had a positive drug urine test result at screening. Interestingly, participants receiving lower buprenorphine cumulative dose, who might be expected to have worse out­ comes [30], had better treatment outcomes. A speculative expla­ nation is that higher buprenorphine doses have been reported to be associated with pain [31]. Participants with extreme pain are expected to have worse outcomes [32]. As a result, participants receiving higher buprenorphine doses had worse outcome because they might have more extreme pain compared to par­ ticipants receiving lower buprenorphine doses. An alternative explanation is that patients with better outcomes might be less

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

sample that was positive for opioids. This trend was also observed by other research groups [10, 21]. The current study directly accounts for relapse observed follow­ ing an initial response to buprenorphine. Previous analysis on the present data explored predictors of successful outcome at the end of stabilization but did not account for the relapse observed among participants [10]. The percentage of partici­ pants with successful outcome decreases from 62 % at the end of ▶  Table 1). This stabilization phase to 31 % at the end of study ( ● highlights the magnitude of relapse observed in opioid treat­ ment programs. It is important to consider the time relative to buprenorphine dosing when describing the efficacy of buprenorphine treat­ ment. To further explain this point we refer to a study by Ling et al. [10]. They compared the effects of 2 buprenorphine tapering schedules, 7 days and 28 days, on opioid addiction outcome. There was a significant difference in the fraction of participants with opioid-free urine samples between the 2 tapering regi­ mens at the end of taper (p = 0.0007). However, there was no sig­ nificant difference in the fraction of participants with opioid-free urine samples between the 2 tapering regimens at 1 month and 3 months post-taper (p > 0.05). To summarize, different conclu­ sions can be drawn from a single study regarding treatment out­ come depending on the timing of the endpoint. At the time of preparing this manuscript, this is the first trial describing the longitudinal change in buprenorphine treatment outcome. Several publications explored predictors of pharmaco­ logical detoxification outcome [6–8]. However, these studies explored treatment outcome at a single endpoint that is usually at the end of treatment. Buprenorphine treatment outcome is a dynamically changing factor that cannot be described by a snap­ shot. Instead, we propose a model that is described by differen­ tial equations to account for the dynamic change in treatment outcome. Additionally, the influence of various predictors was explicitly incorporated in the model which can translate in a more individualized prediction. Other research groups described the longitudinal change in opi­ oid addiction outcome using survival models such as KaplanMeier and the Cox proportional hazard [22,  23]. Survival functions are characterized by being non-increasing functions [24]. Since the longitudinal change in the probability of treat­ ment success is characterized by an initial increase followed by a decrease, survival functions are not suitable to describe the lon­ gitudinal change in detoxification treatment outcome. Instead, survival functions were used to describe remission process and the relapse process separately [22, 23]. This model can be used to predict the fraction of patients with sustained remission that is attained at the end of the study (i. e., 3 months post taper). As presented in ●  ▶  Table 1, the percentage of participants with successful outcome did not change signifi­ cantly (p > 0.05) 1 month post taper until the end of the study. From mathematical point of view, once sustained remission is achieved then the change in the probability of remission and relapse are equal to zero. The following equations are used to derive the odds of having a successful outcome at the end of study. At the end of treatment the rate of change in M equal zero, then:

addicted and hence they needed lower buprenorphine dose. This explanation is supported by the observation that a higher buprenorphine dose is required by patients with more frequent heroin use [33]. The longitudinal regression model predicts 38 % increase in odds of success, at the end of the study, with every 100-mg decrease in the cumulative buprenorphine adminis­ tered over the study period. The present methodology offers several advantages. First, it allows the investigation of the change in buprenorphine treat­ ment outcome over time. Second, studying multiple measure­ ments for each subject allows each subject to be his or her own control. This reduces subject-to-subject variation and directly increases power. The increase in power allows the detection of significant predictors with reduced number of subjects. Third, it can accommodate missing data as opposed to dropping all data from every subject who is missing one or more measurements, and it accommodates unequal and/or irregular spacing of repeated measurements. These advantages are direct implica­ tions of using mixed models [34]. The present analysis can be expanded by exploring additional covariates as possible predictors of buprenorphine treatment outcome. Charney et al. identified depression and anxiety as sig­ nificant predictors of addiction treatment outcome [35]. Khant­ zian et al. argue that substance use disorders can result from preexisting mental disorders (including anxiety and depression) [36]. For example, opioids can be used to alleviate anxiety and cocaine or other stimulants can be used to relieve depression [36]. Depression was positively correlated with continued opioid use [37] and was identified as a predictor of both methandone [38] and buprenorphine [39] treatment outcomes. Prevalence of anxiety in opioid dependent patients ranged from 26 % to 35 % [40]. Opioid dependent patients with anxiety have higher prob­ ability of treatment discontinuation [41]. As a result, additional clinical measures are needed to prevent premature treatment discontinuation [41]. Sleep disturbances is another potential predictor of treatment outcome. Sleep medications improved the retention of opioid dependent patients receiving buprenor­ phine treatment [42]. This association is possibly because sleep medications relief sleep disturbances experienced during opioid addiction treatment [42]. In conclusion, a novel approach for analyzing buprenorphine treatment outcome was presented in this manuscript. This approach describes the longitudinal change in the probability of getting an opioid negative urine sample. Additionally, this model successfully describes relapse to opioid. Finally, several patient characteristics were identified as predictors of responsiveness to buprenorphine treatment.

Conflict of Interest

▼

The author declares no conflict of interest.

References

1 Beal SL, LB S, Boeckmann AJ. NONMEM Users Guides (1989–2006). Ellicott City, Maryland, USA: Icon Development Solutions, 2006 2 Substance Abuse and Mental Health Services Administration. Results from the 2011 National Survey on Drug Use and Health: Summary of National Findings. NSDUH Series H-44, HHS Publication No (SMA) 12-4713. Rockville, MD: Substance Abuse and Mental Health Services Administration, 2012 1 online resource

3 Center for Substance Abuse Treatment. Medication-Assisted Treatment for Opioid Addiction in Opioid Treatment Programs. Pharmacology of Medications Used To Treat Opioid Addiction. Medication-Assisted Treatment for Opioid Addiction in Opioid Treatment Programs. Rock­ ville (MD): 2005 4 Walsh SL, Preston KL, Stitzer ML et al. Clinical pharmacology of buprenorphine: ceiling effects at high doses. Clin Pharmacol Therap 1994; 55: 569–580 5 Chawarski MC, Schottenfeld RS, O’Connor PG et al. Plasma concentra­ tions of buprenorphine 24 to 72 hours after dosing. Drug Alcohol Depend 1999; 55: 157–163 6 Dreifuss JA, Griffin ML, Frost K et al. Patient characteristics associated with buprenorphine/naloxone treatment outcome for prescription opioid dependence: Results from a multisite study. Drug Alcohol Depend doi:10.1016/j.drugalcdep.2012.12.010 7 Ziedonis DM, Amass L, Steinberg M et al. Predictors of outcome for short-term medically supervised opioid withdrawal during a rand­ omized, multicenter trial of buprenorphine-naloxone and clonidine in the NIDA clinical trials network drug and alcohol dependence. Drug Alcohol Depend 2009; 99: 28–36 8 Hillhouse M, Canamar CP, Ling W. Predictors of outcome after shortterm stabilization with buprenorphine. J Subst Abuse Treatment 2013; 44: 336–342 9 Amass L, Bickel WK, Higgins ST et al. A preliminary investigation of outcome following gradual or rapid buprenorphine detoxification. J Addict Dis 1994; 13: 33–45 10 Ling W, Hillhouse M, Domier C et al. Buprenorphine tapering schedule and illicit opioid use. Addiction 2009; 104: 256–265 11 McLellan AT, Kushner H, Metzger D et al. The Fifth Edition of the Addic­ tion Severity Index. J Subst Abuse Treatment 1992; 9: 199–213 12 McGahan P, Griffith J, Parente R et al. Addiction Severity Index: com­ posite scores manual. Philadelphia, PA: Veterans Administration Press, 1986 13 Wesson DR, Ling W. The Clinical Opiate Withdrawal Scale (COWS). J Psychoact Drugs 2003; 35: 253–259 14 Amass L, Kamien JB, Mikulich SK. Efficacy of daily and alternate-day dosing regimens with the combination buprenorphine-naloxone tab­ let. Drug Alcohol Depend 2000; 58: 143–152 15 Bickel WK, Stitzer ML, Bigelow GE et al. A clinical trial of buprenor­ phine: comparison with methadone in the detoxification of heroin addicts. Clin Pharmacol Therap 1988; 43: 72–78 16 Oreskovich MR, Saxon AJ, Ellis ML et al. A double-blind, double-dummy, randomized, prospective pilot study of the partial mu opiate agonist, buprenorphine, for acute detoxification from heroin. Drug Alcohol Depend 2005; 77: 71–79 17 Takama H, Tanaka H, Nakashima D et al. Population pharmacokinetics of intravenous busulfan in patients undergoing hematopoietic stem cell transplantation. Bone Marrow Transpl 2006; 37: 345–351 18 Stroup TS, Lieberman JA. Antipsychotic trials in schizophrenia: the CATIE project. Cambridge, UK; New York: Cambridge University Press, 2010 19 Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit – a collection of com­ puter intensive statistical methods for non-linear mixed effect mod­ eling using NONMEM. Comp Meth Progr Biomed 2005; 79: 241–257 20 Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN)–a Perl module for NONMEM related programming. Comp Meth Progr Biomed 2004; 75: 85–94 21 Woody GE, Poole SA, Subramaniam G et al. Extended vs. short-term buprenorphine-naloxone for treatment of opioid-addicted youth: a randomized trial. JAMA 2008; 300: 2003–2011 22 Hasin D, Liu X, Nunes E et al. Effects of major depression on remission and relapse of substance dependence. Arch Gen Psychiatry 2002; 59: 375–380 23 Joe GW, Simpson DD, Sells SB. Treatment process and relapse to opioid use during methadone maintenance. Am J Drug Alcohol Abuse 1994; 20: 173–197 24 Dawar A, Queiroz R. Logic, Language, Information and Computation 17th International Workshop, WoLLIC 2010, Brasilia, Brazil, July 6–9, 2010. Proceedings. Lecture Notes in Computer Science. Berlin, Heidel­ berg: Springer-Verlag Berlin Heidelberg, 2010 25 Marsch LA, Stephens MA, Mudric T et al. Predictors of outcome in LAAM, buprenorphine, and methadone treatment for opioid depend­ ence. Exp Clin Psychopharmacol 2005; 13: 293–302 26 Soyka M, Zingg C, Koller G et al. Retention rate and substance use in methadone and buprenorphine maintenance therapy and predictors of outcome: results from a randomized study. Int J Neuropsychophar­ macol 2008; 11: 641–653

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

257

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

Original Paper

27 Mancino M, Curran G, Han X et al. Predictors of attrition from a national sample of methadone maintenance patients. Am J Drug Alcohol Abuse 2010; 36: 155–160 28 Stenbacka M, Beck O, Leifman A et al. Problem drinking in relation to treatment outcome among opiate addicts in methadone maintenance treatment. Drug Alcohol Rev 2007; 26: 55–63 29 Joe GW, Simpson DD, Broome KM. Retention and patient engagement models for different treatment modalities in DATOS. Drug Alcohol Depend 1999; 57: 113–125 30 Schottenfeld RS, Pakes J, Ziedonis D et al. Buprenorphine – dose-related effects on cocaine and opioid use in cocaine-abusing opioid-depend­ ent humans. Biol Psychiat 1993; 34: 66–74 31 Chakrabarti A, Woody GE, Griffin ML et al. Predictors of buprenorphinenaloxone dosing in a 12-week treatment trial for opioid-dependent youth: secondary analyses from a NIDA Clinical Trials Network study. Drug Alcohol Depend 2010; 107: 253–256 32 Clark MR, Stoller KB, Brooner RK. Assessment and management of chronic pain in individuals seeking treatment for opioid dependence disorder. Can J Psychiatry/Rev can psychiatrie 2008; 53: 496–508 33 Lintzeris N, Bammer G, Rushworth L et al. Buprenorphine dosing regime for inpatient heroin withdrawal: a symptom-triggered dose titration study. Drug Alcohol Depend 2003; 70: 287–294 34 Seltman HJ. Experimental Design and Analysis. Pittsburgh, PA: Carn­ egie Mellon University, 2013 35 Charney DA, Palacios-Boix J, Negrete JC et al. Association between con­ current depression and anxiety and six-month outcome of addiction treatment. Psychiatric Serv 2005; 56: 927–933

36 Khantzian EJ. The self-medication hypothesis of addictive disorders: focus on heroin and cocaine dependence. American J Psychiatry 1985; 142: 1259–1264 37 Brewer DD, Catalano RF, Haggerty K et al. A meta-analysis of predictors of continued drug use during and after treatment for opiate addiction. Addiction 1998; 93: 73–92 38 Teesson M, Mills K, Ross J et al. The impact of treatment on 3 years’ outcome for heroin dependence: findings from the Australian Treat­ ment Outcome Study (ATOS). Addiction 2008; 103: 80–88 39 Dreifuss JA, Griffin ML, Frost K et al. Patient characteristics associated with buprenorphine/naloxone treatment outcome for prescription opioid dependence: Results from a multisite study. Drug Alcohol Depend 2013; 131: 112–118 40 Fatseas M, Denis C, Lavie E et al. Relationship between anxiety disor­ ders and opiate dependence – a systematic review of the literature: implications for diagnosis and treatment. J Subst Abuse Treatment 2010; 38: 220–230 41 Benningfield MM, Dietrich MS, Jones HE et al. Opioid dependence dur­ ing pregnancy: relationships of anxiety and depression symptoms to treatment outcomes. Addiction 2012; 107 (Suppl 1): 74–82 42 Warden D, Subramaniam GA, Carmody T et al. Predictors of attrition with buprenorphine/naloxone treatment in opioid dependent youth. Addict Behav 2012; 37: 1046–1053

Saleh MI. Modeling Buprenorphine Treatment Outcome …  Pharmacopsychiatry 2014; 47: 251–258

This document was downloaded for personal use only. Unauthorized distribution is strictly prohibited.

258 Original Paper

Copyright of Pharmacopsychiatry is the property of Georg Thieme Verlag Stuttgart and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.