Epidemiology 200C Homework #2

Epidemiology 200C Homework #2

Due 4/22/10

Read Oviedo-Joekes et al., “Diacetylmorphine versus Methadone for the Treatment of Opioid Addiction,” and answer the following questions:

1.Although patients assigned to diacetylmorphine treatment had better reported outcomes than those assigned to methadone maintenance, the authors recommend that methadone remain the “treatment of choice for the majority of patients.” Aside from more severe side effects (including possible overdose), and possible expenses and logistic problems in delivering injectable diacetylmorphine, why do you think that the authors make this recommendation (hint – look at the inclusion criteria for the study).

The study was limited to patients who had relapsed to injection drug use after at least one attempt at methadone maintenance. In other words, the trial was restricted to a subset of opioid users for whom methadone was an ineffective treatment. Since methadone is an effective treatment for many opioid users, it is possible that the relative efficacy of methadone compared to diacetylmorphine is very different in the population of opioid users as a whole.

2.The authors state that “Some persons probably volunteered for thestudy in the hope of being assigned to receivediacetylmorphine; hence, we anticipated a substantialdropout rate in the methadone group.” They also state that “All participantslost to follow-up were considered not tohave been retained in treatment and not to have a response.” Draw a DAG showing the relationship between randomization status, actual treatment, loss to follow-up, actual outcome, and measured outcome. Is a test of the treatment assignment and the measured outcome a valid test of the treatment effect of methadone versus diacetylmorphine on the outcomes under study other than retention?

No, a test of the association between R and O* is not a valid test of the effect of T on O because there is a selection bias induced by the loss to follow-up whichintroduces biasing paths from R to O* (see the attached DAGs for the paths). Notice, however, the paths from R via D to O or O* are blocked by D=1. Only those who were retained had their outcome measured (O*); we can also include arrow from D to O if we assume like the authors that only those retained actually had a response (O)

3. What about the differences in methods of administering treatment could have affected the results of the study?

The methadone was administered orally and the diacetylmorphine was injected so patients were fully aware whether they were receiving methadone or not, and would have had different incentives to remain in the trial based on the drug effects. This difference in the method of administration thus could have affected both measured outcomes and differences in loss to follow-up rates.

The patients receiving injectable drugs could at any time switch partially or totally to oral methadone if such a switch was deemed appropriate by a physician (30 patients out of the 115 in the injectable diacetylmorphine group did this, whether voluntarily or not). Such a high rate of people switching from the intervention assigned to the other intervention (given that the opposite was not possible) could have led to bias in estimates of effect.

4. Why do you think the authors stratified the randomization of treatment according to center and to the number of previous methadone treatments? Do you have evidence that this prevented confounding in the resulting analysis?

The authors probably stratified on these factors because they thought that these could be factors which affected the results of the study. There could be differences in the people who attended one center or another, or differences in the people who had a few previous methadone treatments and several previous methadone treatments that could have impacted the measured outcome of retention rates and reduction in use. If the distribution of these factors was uneven in the treatment groups, they likely thought it would affect the results from the study of the effect of treatments on the measured outcomes.

Randomization does not prevent confounding however. Although they attempted to ensure the distribution of treatments was relatively even among the groups they stratified on, we do not know if the groups were comparable on the unmeasured factors that could have affected compliance and the outcomes. Randomization-based statistics such as intent-to-treat tests would largely account for thissource of uncertainty, but not the problems of follow-up.