Prescriber effects on outcome of psychopharmacological treatment of Major Depressive Disorder (MDD)

Lead Investigator: Robert Schoevers, University of Groningen
Title of Proposal Research: Prescriber effects on outcome of psychopharmacological treatment of Major Depressive Disorder (MDD)
Vivli Data Request: 5974
Funding Source: None
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Treatment outcome is the result of both specific and unspecific factors. In Major depressive disorder (MDD) treatment, the primary specific factor is the ingredient of the antidepressant medication. Unspecific factors include treatment form and setting, as well as characteristics of both patient and doctor/therapist. In placebo-controlled studies, these unspecific factors are controlled for, so that the difference between pill and placebo can be fully ascribed to the medication. To maximize outcome for patients, it is relevant to learn more about the influence of therapist factors, and how they could be strengthened. As of yet, very few studies have done such an analysis. A secondary analysis of an Randomized Controlled Trail (RCT) with imipramine by McKay et al (2006) suggested that these therapist effects could be substantial. However, this was a relatively small study, and these results have not been replicated. Therefore, the current study will investigate the magnitude of the influence of individual therapists on the variance in outcomes of MDD treatment in an independent and larger dataset that contains raw individual-level data from multiple previous RCTs comparing duloxetine and placebo. Importantly, investigation of therapist influence – is the prescriber of the antidepressant medication – on the effects of both antidepressants and placebo is possible in this data, as for all participants in the studies an identification code of the individual prescriber is available (“invglobe”).

Relevance:

If results show relevance of variation in prescriber factors for treatment outcome, this knowledge could be used to improve treatment effectiveness, for example by optimizing certain clinical skills.

Statistical Analysis Plan:

Method. (a) Study design:
Multi-site, randomized, double-blind, placebo-controlled trial. Adult outpatients (>18 years) meeting DSM-IV criteria for MDD received either placebo or duloxetine 60mg/day to 120mg/day for 8 weeks.

Method. (b) Statistical analysis:
As in previous work, Hierarchical linear modeling (HLM) with post-treatment depression score as outcome will be applied, given that participants are nested within psychiatrists/therapists, HLM is useful to disentangle the variance in the outcome that is explained by treatment and the variance that is explained by therapists/psychiatrists. Different models are fit to compare the explained variances of these two variables. In these models, patients are at level-1 and therapists are at level-2, as the former are nested under the latter. All models include the baseline depression severity score as covariate to adjust for baseline severity differences between patients.

1. Variance explained by therapist: treatment is used as a fixed factor and therapist is used as a random factor (for estimation of a random intercept) to estimate the variance determined by therapist. The latter is estimated as the ratio of variance due to therapist to the total variance (VAR(ther)/VAR(ther)+VAR(residual); also known as the intraclass correlation; ICC).

2. Variance explained by treatment: the variance explained by treatment is estimated by fitting a baseline model without treatment as independent model and a model with treatment as fixed factor, using the ratio of the difference in explained variance to the baseline model variance as an estimate of explained variance (R2) by treatment.
As both the estimates from the first and second models are ratio’s they can be compared to gain insight into the proportions of variance explained by therapist and treatment.

3. To investigate if treatment varied significantly across therapists, a third model is fit with both treatment and therapist as random effects. In this model, the variance of the effect of treatment across therapists is estimated in the level-2 model: b(treatment)=γ1+u1. Here a significant variance indicates that the treatment effect varies significantly across therapists.

Effect measure of interest:
The 17-item Hamilton rating scale for depression (HAMD17) total score after 8 weeks of treatment.

Planned adjustment for covariates:
Relevant covariates will be included in the HLM to adjust for their potential effects on the estimated explained variance by treatment and therapist, respectively.

Meta-analysis methods:
Individual Patient Data (IPD) meta-analysis. This meta-analysis method involves combining and analyzing the quantitative evidence from the above selected studies to produce results based on a whole body of research. We will use a two-step approach;

The two-step approach, the individual participant data are first analyzed in each separate study independently by using a statistical method appropriate for the type of data being analyzed. This step produces aggregate data for each study. These data are then synthesized in the second step using a suitable model for meta-analysis of aggregate data.

Power to detect a clinically important effect:
The majority of the current study focuses on estimation of estimated variance in the outcome by therapist and by treatment rather than null-hypothesis testing (model 1 and 2).

In the third model, a H0-test actually does apply. Here, we test the H0 that u1>0. We simulated the required sample size (level-2 n therapists, level-1 n patients) to detect a significant medium slope variance across therapists (=0.32=0.09) using a standard Likelihood Ratio Test comparing a model with random slope to a model without random slope. Monte Carlo estimation with 1000 draws showed that a significant non-zero variance (at alpha<0.05) could be observed with 80% power at a sample size of at least 30 therapists, with at least 35 patients per therapist. If more therapists are available, less patients per therapist are needed (e.g. 50 therapists with each 30 patients, 60 therapists, with each 25 patients). If less data are available, only larger effects will be detectable.

Planned sensitivity analysis:
Not applicable.

Planned subgroup analysis:
Not applicable.

Handling of missing data:
All analyses will be conducted on an intent-to-treat basis

Requested Studies:

Duloxetine Versus Placebo in the Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAQ(A)

Duloxetine Versus Placebo in the Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAQ(B)

Duloxetine Versus Placebo and Paroxetine in the Acute Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAT(A)

Duloxetine Versus Placebo and Paroxetine in the Acute Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAT(B)

Duloxetine Versus Placebo and Paroxetine in the Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAY(A)

Duloxetine Versus Placebo and Paroxetine in the Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMAY(B)

Duloxetine Once-Daily Dosing Versus Placebo in the Acute Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMBH(A)

Duloxetine Once-Daily Dosing Versus Placebo in the Acute Treatment of Major Depression
Sponsor: Eli Lilly and Company
Sponsor ID: F1J-MC-HMBH(B)

Duloxetine Hydrochloride (LY248686) Protocol F1J-US-HMCB Duloxetine Once-Daily Dosing Versus Placebo in Patients With Major Depression and Pain
Sponsor: Eli Lilly and Company
Study ID: NCT00036335
Sponsor ID: 6353

Duloxetine Versus Escitalopram and Placebo in the Treatment of Patients With Major Depression
Sponsor: Eli Lilly and Company
Study ID: NCT00073411
Sponsor ID: 7978

Duloxetine Versus Placebo in the Long-Term Treatment of Patients With Late-Life Major Depression
Sponsor: Eli Lilly and Company
Study ID: NCT00406848
Sponsor ID: 10815

Summary of results:

Our analysis plan was based on the idea that the personal relation with a care provider would be one of the factors that could determine treatment outcome (together with medication/placebo and other possible predictors). Unfortunately, we have not been able to find the identifiers of the treating clinicians that we were looking for. It cost us a lot of time to work through the available information as this was sometimes ambiguously formulated, but we had to conclude that the data are not suited for this analysis.

We would very much like to thank you for your willingness to enable such an analysis and are sorry that this is the result. We appreciate the fact that these data are made available for research and will definitely keep this possibility in mind for future plans.