The response pattern to SSRIs as assessed by the Montgomery‐Åsberg Depression Rating Scale: a patient‐level meta‐analysis

The effect size for antidepressants vs. placebo varies considerably among the 17 symptoms rated by the Hamilton Depression Rating Scale (HDRS) ¹ . Using patient‐level data (N=~13,000) from the development programs of citalopram, duloxetine, paroxetine and sertraline, we reported that there are sizeable effects on HDRS items such as depressed mood and psychic anxiety, which appear already after one week of treatment, but negligible effects, throughout the treatment period, on items that may capture side effects of selective serotonin reuptake inhibitors (SSRIs), such as insomnia, somatic anxiety, gastrointestinal symptoms, genital symptoms, and weight change1, 2, 3. Other authors have reported similar findings4, 5.

While the Montgomery‐Åsberg Depression Rating Scale (MAD­RS) overlaps with the HDRS ⁶ , there are significant differences between the two scales with respect to how the various symptoms are described. Moreover, the MADRS includes some key depressive symptoms not explicitly rated by the HDRS, such as inability to feel and concentration difficulties. Patient‐level analyses of the impact of SSRIs on individual MADRS items may thus allow us to assess to what extent symptom‐level findings based on HDRS ratings generalize to other instruments, and may further our understanding of the effects of SSRIs on different depressive symptoms.

We report here symptom‐level MADRS ratings from 4,243 subjects participating in twelve acute phase placebo‐controlled trials of an SSRI in major depression (see supplementary information). Our aims were: a) to investigate the time‐course and magnitude of the effects of SSRIs on individual MADRS items; b) to assess the relation of individual MADRS items to the MADRS total score; and c) to compare drug‐placebo differences for the total score of a six‐item unidimensional MADRS subscale (MADRS‐6) ⁷ – consisting of the items reported sadness, apparent sadness, inner tension, lassitude, inability to feel, and pessimistic thoughts – with those for the total score of the full MADRS.

Outcome measures were assessed using linear mixed models. The models included baseline score on the outcome parameter as a covariate, fixed effects for time (week), treatment (SSRI or placebo), trial, and the interaction between treatment and time. Within‐subject correlations were modelled using an unstructured covariance matrix, and denominator degrees of freedom were estimated using the Kenward‐Roger approximation. The parameters of interest were treatment group means over time, as well as effect sizes and levels of significance for the between‐treatment comparisons.

We assessed the following outcome measures: a) the total score of the MADRS scale; b) the total score of the MADRS‐6 subscale; and c) the scores on all individual MADRS items. Week 6 was selected as endpoint since that was the last available evaluation in 9 out of 12 trials. All analyses were conducted in SAS version 9.4.

All individual items showed statistically significant separation between drug and placebo at endpoint, with endpoint effect sizes ranging between 0.08 (reduced appetite) and 0.38 (apparent sadness as well as reported sadness). Five MADRS items (i.e., apparent sadness, reported sadness, inner tension, pessimistic thoughts, and suicidal thoughts) showed significant separation in favor of SSRIs after one week of treatment, with increasing separation over time until endpoint. Three items (i.e., concentration difficulties, lassitude, and inability to feel) showed significant separation in favor of SSRIs after two or three weeks of treatment and onwards. One item, reduced appetite, separated in favor of placebo after one and two weeks of treatment, but not thereafter. Reduced sleep displayed a nominally negative effect size (–0.05) at week 1, but a small though significant effect size favoring SSRIs at endpoint (0.09). The effect size was 0.37 at endpoint for the full MADRS and 0.40 for the MADRS‐6 subscale (see also supplementary information).

Thus, with respect to items that overlap in content, the response pattern to SSRIs observed with MADRS appears very similar to that seen using HDRS ¹ . Taken together, these results suggest that the general response pattern for SSRIs in depression is not conditional on the specific features of any particular rating instrument, but reflect true symptom‐level effects of these treatments.

The effects on the three MADRS items that have only partially corresponding items in the HDRS were positive, but took somewhat longer time to develop. While concentration difficulties and anhedonia improved significantly after two weeks of treatment and onwards, lassitude required three weeks of treatment to display significant improvement (which also lasted throughout the trial).

Recent reports suggest a blunting of emotions to be a possible side effect of SSRIs, while also acknowledging a correlation between the presence of this symptom and depression severity ⁸ . It is hence of interest to note that SSRIs reduced the score on the inability to feel item in the MADRS – a symptom not included in the HDRS – with an effect size (0.32) similar to those noted for inner tension and the two sadness items (0.35‐0.38). Thus, while the possibility that drug treatment may elicit this symptom in some subjects should not be overlooked, and requires further study, the net impact on emotional blunting of treating depressed subjects with an SSRI for 6 weeks appears favorable rather than harmful.

The effect sizes for individual MADRS items at endpoint generally displayed less dissimilarity than those reported for HDRS items. With the exception of reduced sleep and reduced appetite, which displayed endpoint effect sizes of 0.09 and 0.08, respectively, item‐level effect sizes ranged from 0.23 (concentration difficulties) to 0.38 (reported and apparent sadness). In line with this, the difference between the full scale total score on the one hand, and the total score for the 6‐item subscale, or the scores on the best performing individual items, on the other, was markedly smaller for the MADRS as compared to the HDRS ¹ .

In summary, symptom‐level analyses on MADRS data show a response pattern highly similar to that seen in HDRS analyses, suggesting that the observed effects are not related to particularities of any chosen scale, but do reflect the true symptom‐level profile of SSRIs when used for depression. While the MADRS, like the HDRS, includes some items (i.e., reduced sleep and reduced appetite) that may be contaminated by SSRI side effects ³ , hence introducing a negative bias as compared to placebo, the total score of the MADRS is less impacted by the inclusion of such items than is the HDRS total score.