Most comparisons of the efficacy of antidepressants have relied on the assumption that missing data are randomly distributed. Dropout rates differ between drugs, suggesting this assumption may not hold true. This paper examines the effect of non-random dropout on a comparison of two antidepressant drugs, escitalopram and nortriptyline, in the treatment of major depressive disorder. The GENDEP study followed adult patients with major depressive disorder over 12 weeks of treatment, and the primary analysis found no difference in efficacy of the two antidepressants under missing at random assumption. By applying the recently developed Muthén–Roy model, we compared the relative efficacy of these two antidepressants taking into account non-random distribution of missing outcomes (NMAR). Individuals who dropped out of the study were those who were not responding to treatment. Based on the best fitting NMAR model, it was found that escitalopram reduced symptom scores by an additional 1.4 points on the Montgomery–Åsberg Depression Rating Scale (p = 0.02), equivalent to 5% of baseline depression severity, compared to nortriptyline. We conclude that association between dropout and worsening symptoms led to an overestimate of the effectiveness of treatment, especially with nortriptyline, in the primary analysis. These findings review the primary analysis of GENDEP and suggest that, when non-random dropout is accounted for, escitalopram is more effective than nortriptyline in reducing symptoms of major depression.
Antidepressants are the primary treatment for moderate and severe depression. It can take up to 6–8 weeks of treatment for symptoms to decrease (Anderson et al., 2008; Uher et al., 2011). However many individuals do not complete treatment (Lingam and Scott, 2002; Olfson et al., 2006). The reasons for discontinuing treatment vary, and include lack of response, side-effects, and remission of symptoms. In a clinical trial these factors can make dropout systematically related to outcome. This is especially important in the comparison of antidepressants that differ in the burden of side-effects and the percentage of individuals who complete treatment. For example, clinical trials comparing tricyclic antidepressants (TCAs) and selective serotonin reuptake inhibitors (SSRIs) have reported higher rates of drop-out in the TCAs (Arroll et al., 2005; Hirschfeld, 1999; MacGillivray et al., 2003; Uher et al., 2009b), potentially complicating the comparison of efficacy.
When making the decision whether to continue or stop medication, the patient and clinician often weigh the perceived therapeutic effect against the burden of side effects. This systematic relationship between efficacy, side effects and discontinuation can produce data not missing at random (NMAR) (Little and Rubin, 2002). This means that missing data differ systematically from observed values. It is often described as informative or non-ignorable missingness, and differs from data missing completely at random (MCAR) and data missing at random (MAR). With MCAR, the outcome variable is not related to the probability of dropout. In MAR, the observed values of the outcome variable are related to the probability of dropout, but the unobserved outcomes are not, after accounting for other covariates included in the analysis. In MNAR, the unobserved outcomes are related to the probability of dropout. An example of NMAR would be when individuals stop improving and dropout of the study before assessment, and so are lacking measurements showing the lack of improvement from which the cause of dropout could be established. Whether missing data are considered informative depends on the method of analysis, specifically which types of missingness it can account for. For instance, in general estimating equations both MAR and NMAR non-ignorable, while in likelihood based estimation only NMAR is non-ignorable. As a result, conventional methods of assessing and comparing the efficacy of antidepressants may produce biased results unless NMAR data are explicitly modelled and taken into account. In this case, the unobserved cause of missingness may be related to the trajectory of response to anti-depressants, and so captured by latent variables representing the slope or intercept of response. Several previous studies have shown the benefits of trajectory modelling in the analysis of clinical data (Gueorguieva et al., 2011; Marques et al., 2011; Stauffer et al., 2011; Uher et al., 2010a). A method based on trajectory modelling has been proposed to account for NMAR and has been previously applied to dropout in level I of the STAR*D study where all patients were treated with the same SSRI antidepressant (Muthén et al., 2011). This model looks for an association between patterns in dropout during the study and trajectories of response to treatment. It has also been used as a secondary analysis of a comparison of duloxetine against SSRIs and placebo treated groups (Gueorguieva et al., 2011). Here, we apply this method to the comparison of the efficacy of two antidepressants in the GENDEP study: escitalopram (an SSRI) and nortriptyline (a TCA). While the primary analysis of GENDEP showed no difference in efficacy between the two antidepressants (Uher et al., 2009b), they differed in percentage of individuals who dropped out of the study. Our aim is to examine if the differential dropout has affected the efficacy comparison.
