Comparing sensitivity to change using the 6-item versus the 17-item Hamilton depression rating scale in the GUIDED randomized controlled trial

Roughly half of patients with major depressive disorder (MDD) fail to respond to treatment with an antidepressant medication, and approximately two-thirds fail to achieve remission [1]. These inadequate outcomes have sparked great interest in exploring biological subtypes of depression that correlate with variability in medication response [2]. Pairing clearly defined subtypes with validated biomarkers such as genetic and epigenetic, proteomic, metabolomic, inflammation, neuroimaging, and electroencephalography measures might enable more precise treatment selection and response monitoring.

Genetic variation is an important biological contributor to both MDD development [3, 4] and to treatment response [5, 6]. On their own, individual gene variants explain little of the variance in disease risk or outcomes; rather, clinical manifestation of MDD and treatment response appear to result from the combined effects of many genes, along with other clinical and environmental factors. Combinatorial pharmacogenomic tests, which evaluate the weighted effects of genetic variants to predict which medications may be impacted by gene-drug interactions, hold promise for aiding patient-specific treatment selection [7]. Recently, the Genomics Used to Improve DEpression Decisions (GUIDED) randomized controlled trial (RCT) reported on the efficacy of using a combinatorial pharmacogenomic test in medication selection (guided-care), compared with treatment as usual (TAU), for patients with treatment non-responsive MDD [8]. This trial differed from traditional drug studies in that patients in both arms received active treatment. GUIDED approached but did not achieve a statistically significant difference between guided-care versus TAU for its primary outcome, percent symptom improvement at week 8 (p = 0.069; intent-to-treat [ITT] cohort), as assessed by the Hamilton Depression Rating Scale, 17-item (HAM-D17). However, significantly more patients achieved the secondary outcomes, response (p = 0.007) and remission (p = 0.005) at week 8, measured using HAM-D17, when they received pharmacogenomics-guided care.

The results observed in the GUIDED trial highlight the challenges in detecting clinically and statistically significant differences in randomized trials when patients in all study arms receive active treatment. This is especially true in psychiatry, where several well-powered randomized trials comparing active MDD treatments have failed to show differences in efficacy, including the Sequenced Treatment Alternatives in Depression (STAR*D) trial [9], the Genome-Based Therapeutic Drugs for Depression (GENDEP) trial [10], and the Combining Medications to Enhance Depression Outcomes COMED trial [11]. Lack of significant differences in efficacy extends even to large trials that compare psychotherapy, antidepressant medications, or their combination [12, 13]. Such equivalent outcomes, despite the treatments’ distinct mechanisms, raise the possibility that the assessment metrics used are flawed [14].

The Hamilton Depression Rating Scale (HAM-D) is the most widely used outcome measure in MDD clinical trials, with the 17-item version (HAM-D17) originally published in 1960, serving as the standard [15, 16]. Over the past four decades, however, researchers have raised concerns about the ability of the HAM-D17 scale to assess accurately the severity of and change in depression symptoms [17‐19]. Factor analyses of HAM-D17 have determined that the scale is not a unidimensional measure of depression severity but rather consists of two to eight symptom factors [20]. Although multidimensionality in a scale is useful for detecting a broad array of clinical features, a multidimensional (or multifactorial) scale may reduce the ability to detect change over time, because some factors may not adequately distinguish groups when valid differences exist [21]. The ability to scale appropriately with illness severity is a fundamental aspect of construct validity. Medication side effects may affect some factors on multidimensional scales more than others, potentially producing total score changes that do not align with changes in core depressive symptoms [22, 23]. In studies such as GUIDED that allow concomitant treatments (e.g., sedative hypnotics for insomnia and anxiety in conjunction with antidepressant medication), assessing efficacy with HAMD-17 becomes even more problematic, as the uncontrolled additional medications can result in score changes unrelated to antidepressant treatment.

To address these shortcomings, researchers developed abbreviated, more focused versions of HAM-D17 [24]. Of these, the most widely used is the six-item subscale of HAM-D17, known as the HAM-D6 or melancholia subscale [23, 25]. The HAM-D6 scale is specific to the core depressive symptoms of depressed mood, guilt, work and activities, psychomotor retardation, psychic anxiety, and general somatic symptoms (energy and physical pain), and it is unidimensional [26]. HAM-D17 symptoms omitted from the HAM-D6 scale include suicidal thoughts, initial insomnia, middle insomnia, late insomnia, psychomotor agitation, somatic anxiety, gastrointestinal symptoms [appetite], sexual disturbances, hypochondriasis [somatization], insight, and weight loss. The HAM-D6 scale correlates better with the Clinical Global Impressions Scale-Severity than does the HAM-D17 scale, particularly among more severely ill patients [21]. It has repeatedly demonstrated greater effect sizes for second-generation antidepressants than has HAM-D17, as well as similar effect sizes for medications that have sedating side effects, such as TCAs and mirtazapine [27‐29].

This comparative, post hoc analysis of the HAM-D6 and HAM-D17 depression scales in the GUIDED trial for MDD treatment found greater sensitivity to differences in treatment effects with the abbreviated version of the scale. This result likely is due to the narrower focus of the HAM-D6 scale, as compared with HAM-D17, for the core symptoms of depression. Furthermore, although both versions of the scale achieved statistically significant differences for the response and remission outcomes, the greater differences for symptom improvement seen with the HAM-D6 scale suggest that HAM-D6 provided a more precise measure for MDD outcome assessment. This is supported further by the observation of the high sensitivity of the HAM-D6 scale in the subset of patients who entered the trial on medications predicted by the pharmacogenomic test to have gene-drug interactions. Mokken scale analysis further supported the increased homogeneity of HAM-D6 relative to HAM-D17. Altogether, these results mirror those seen in many placebo-controlled pharmacological trials, wherein the HAM-D17 scale failed to identify an antidepressant effect, while the HAM-D6 scale did [27].

Although the percent differences in response and remission rates were generally similar for HAM-D17 and HAM-D6, the slightly lower difference in remission rate between arms (0.8%) as assessed by the HAM-D6 (∆ = 4.6%) versus the HAM-D17 (∆ = 5.4%) is of interest. A concern in the field is that the standard HAM-D17 remission threshold of ≤7 may be high, capturing many patients who continue to experience impairment or distress from persisting symptoms [35, 36]. Thus, low levels of core symptoms, as determined by the standard HAM-D6 remission threshold (≤4), might comprise a more valid measure for defining the state of clinical remission. The question of whether the HAM-D6 or the HAM-D17 remission threshold better predicts restoration of function and long-term wellness should be a focus of future work.

The importance of maximizing signal detection through use of the most sensitive scale to detect treatment effects is of particularly great importance for comparative effectiveness studies and for biomarker-based clinical trials, both of which provide active treatment to all patients [37]. Over the past several decades, adequately powered MDD trials comparing active treatments, be they medications or psychotherapies, have found no difference between treatment arms [9‐11]. Notably, all of these large trials have used either the HAM-D17 scale, the Montgomery-Åsberg Depression Rating Scale (MADRS), or the Quick Inventory of Depressive Symptomatology-Self-Report (QIDS-SR) as the efficacy measure, each of which contains numerous items unrelated to core depression symptoms captured by the HAM-D6 scale. Trials such as GUIDED that allow concomitant medications for specific symptoms, such as sedative hypnotics for anxiety or insomnia, can further diminish the ability to identify a difference between treatment arms when the outcome measure includes non-core depressive symptoms [38]. Consequently, future randomized trials applying biomarker-based approaches to treatment selection for MDD may benefit from using the HAM-D6 or a similar, more focused symptom scale.

The greater discriminative ability of the HAM-D6 scale also allows for smaller sample sizes to test hypotheses about efficacy [39‐41]. Given the greater precision and numerous advantages of HAM-D6, it is difficult to justify continued use of the full HAM-D17 scale as the sole primary outcome measure in MDD treatment trials. In the future, HAM-D17 could be used to enable historic comparisons of baseline severity among trials, but new study protocols should consider specifying the HAM-D6 or a similarly more precise assessment of core symptoms [42, 43] as the primary efficacy variable for analysis. Administering the shorter version can have the added benefit of reducing time burdens on clinical trial participants.

This analysis had several strengths that were inherent to the GUIDED primary analysis. First, the diversity of the study cohort mirrors that seen across varied clinical scenarios for MDD treatment, including clinicians in both psychiatric specialty and primary care clinics. Second, the study’s two active treatment arms reflect real-world clinical practice and provide a relevant evaluation of clinical utility. The limitations of the primary GUIDED analysis also apply to this study [8]. Specifically, the treating clinician was not blinded to study arm, though this limitation was mitigated somewhat by using blinded central raters, and by keeping the site raters and patients blinded to study arm until after week 8. The impact of polypharmacy is another intrinsic limitation; however, as was discussed in the primary analysis, confounding effects likely would be equivalent between study arms. A specific limitation of using the HAM-D6 scale is that it does not assess for some important depressive symptoms, including physical symptoms [24] and suicide. The routine use of separate, more comprehensive suicide assessments in modern clinical trials of MDD treatments reduces concern about this limitation.

The results of this analysis are consistent with a substantial body of published evidence showing that HAM-D6, which is focused more precisely on core depressive symptoms, is more sensitive than HAM-D17 in assessing depression symptom improvement in patients with MDD. The demonstrated utility of pharmacogenomics-guided treatment over TAU as detected by the HAM-D6 in the GUIDED trial highlights its value for future biomarker-guided trials comparing multiple active treatment arms.