Psychological Medicine Dealing with heterogeneity of cognitive cambridge.org/psm dysfunction in acute depression: a clustering approach Muriel Vicent-Gil1,2, * , Maria J. Portella2, * , Maria Serra-Blasco1 , Original Article Guillem Navarra-Ventura1 , Sara Crivillés1 , Eva Aguilar1, Diego Palao1 *Equal contribution authorship. and Narcís Cardoner1 Cite this article: Vicent-Gil M, Portella MJ, Serra-Blasco M, Navarra-Ventura G, Crivillés S, 1 Mental Health Department, Hospital Universitari Parc Taulí, Institut d’Investigació i Innovació Parc Taulí (I3PT), Aguilar E, Palao D, Cardoner N (2021). Dealing Universitat Autònoma de Barcelona (UAB), Biomedical Research Networking Center Consortium on Mental Health with heterogeneity of cognitive dysfunction in (CIBERSAM), Parc Taulí 1, 08208 Sabadell, Catalonia, Spain and 2Department of Psychiatry, Hospital de la Santa acute depression: a clustering approach. Psychological Medicine 51, 2886–2894. https:// Creu i Sant Pau, Biomedical Research Institute Sant Pau (IIB-Sant Pau), Universitat Autònoma de Barcelona (UAB), doi.org/10.1017/S0033291720001567 Biomedical Research Networking Center Consortium on Mental Health (CIBERSAM), Sant Antoni Mª Claret 167, 08025 Barcelona, Catalonia, Spain Received: 29 November 2019 Revised: 18 March 2020 Abstract Accepted: 8 May 2020 First published online: 1 June 2020 Background. Heterogeneity in cognitive functioning among major depressive disorder (MDD) patients could have been the reason for the small-to-moderate differences reported Key words: Acute episode; analysis; cluster; cognition; so far when it is compared to other psychiatric conditions or to healthy controls. heterogeneity; major depressive disorder Additionally, most of these studies did not take into account clinical and sociodemographic characteristics that could have played a relevant role in cognitive variability. This study Authors for correspondence: Maria Serra- aims to identify empirical clusters based on cognitive, clinical and sociodemographic variables Blasco, E-mail: mserrab@tauli.cat; Narcís in a sample of acute MDD patients. Cardoner, E-mail: ncardoner@tauli.cat Methods. In a sample of 174 patients with an acute depressive episode, a two-step clustering analysis was applied considering potentially relevant cognitive, clinical and sociodemographic variables as indicators for grouping. Results. Treatment resistance was the most important factor for clustering, closely followed by cognitive performance. Three empirical subgroups were obtained: cluster 1 was characterized by a sample of non-resistant patients with preserved cognitive functioning (n = 68, 39%); clus- ter 2 was formed by treatment-resistant patients with selective cognitive deficits (n = 66, 38%) and cluster 3 consisted of resistant (n = 23, 58%) and non-resistant (n = 17, 42%) acute patients with significant deficits in all neurocognitive domains (n = 40, 23%). Conclusions. The findings provide evidence upon the existence of cognitive heterogeneity across patients in an acute depressive episode. Therefore, assessing cognition becomes an evi- dent necessity for all patients diagnosed with MDD, and although treatment resistant is asso- ciated with greater cognitive dysfunction, non-resistant patients can also show significant cognitive deficits. By targeting not only mood but also cognition, patients are more likely to achieve full recovery and prevent new relapses. Introduction Research over the last decade has been mainly focused on cognitive performance after a depressive episode (Semkovska et al., 2019) so as to explore difficulties in cognition as an inde- pendent facet of clinical manifestation of major depressive disorder (MDD). Some studies report improvements in cognitive performance upon remission of depression (Biringer et al., 2007) whereas others suggest that cognitive impairment persists during clinical remis- sion (Bora, Harrison, Yücel, & Pantelis, 2013; Hasselbalch, Knorr, & Kessing, 2011). The dis- parity in these results might be explained by the fact that a number of patients do not show any cognitive impairment over the course of the disorder, and others display significant cognitive © The Author(s), 2020. Published by difficulties even in a non-symptomatic phase. In this regard, a recent study has even found Cambridge University Press. This is an Open discrete neurocognitive subgroups suggesting the presence of substantial heterogeneity in neu- Access article, distributed under the terms of rocognitive performance in MDD patients with current affective stability (Pu, Noda, Setoyama, & the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), Nakagome, 2018). It is noteworthy to mention that although patients were in clinical remis- which permits unrestricted re-use, sion, some of them were classified as globally impaired showing moderate to severe cognitive distribution, and reproduction in any medium, impairment. Therefore, it would be reasonable to think that unresolved cognitive deficits were provided the original work is properly cited. already present during the acute phase of MDD, from which it could be inferred that different cognitive profiles existed among patients. Despite such a heterogeneity being frequently pro- posed as an explanation for disparate findings in cognitive performance during a depressive episode (Hammar & Ardal, 2009; Lee, Hermens, Porter, & Redoblado-Hodge, 2012), it has scarcely been investigated. A recent study by our group has already found two distinguishable Psychological Medicine 2887 cognitive profiles in first-episode patients (Vicent-Gil et al., 2018), studying the cognitive functioning in major depression whereas other reports used mixed samples of mood disorders (Serra-Blasco et al., 2019). The patients fulfilled the inclusion cri- (Cotrena, Branco, Kochhann, Shansis, & Fonseca, 2016; Iverson, teria of a current episode of MDD (DSM-IV-TR criteria). Brooks, Langenecker, & Young, 2011). Diagnosis was double checked by two experienced psychiatrists Previous studies have described that cognitive dysfunction and validated through clinical reports. Exclusion criteria for all could be associated with more severe manifestations of the disease participants were the following: (i) presence of any neurological (Serra-Blasco & Lam, 2019), such as early/late illness onset, symp- disease, (ii) medical illness with a known impact on cognitive tom severity (McDermott & Ebmeier, 2009), number of previous functioning, (iii) intelligence quotient (IQ) <85, (iv) presence of depressive episodes (Semkovska et al., 2019) and a higher level of a comorbid axis I diagnosis with the exception of anxiety disor- resistance to antidepressant strategies (Murrough et al., 2015; ders and dysthymia, (v) past or current substance abuse or (vi) Pimontel et al., 2016; Serra-Blasco et al., 2015). All these studies any axis II diagnosis according to the DSM-IV-TR. Participants did not take into account the possible heterogenic cognitive pro- were on medication at the time of evaluation. Patients were files among the included patients, which could have explained the invited to participate in this cross-sectional study, which included small-to-moderate effect sizes reported so far. Apart from that, a clinical and neuropsychological assessment conducted by other sociodemographic variables, such as years of schooling experienced research neuropsychologists. The study was set fol- (Venezia et al., 2018) and age (Dotson, Resnick, & Zonderman, lowing the principles of the Declaration of Helsinki and was 2008), could also be associated with lower cognitive performance. approved by the Research Ethics Board of the Institut In fact, two recent studies on clustering analysis of cognitive func- d’Investigació i Innovació Parc Taulí (I3PT) at Hospital tioning have shown that the most affected cognitive profile was Universitari Parc Taulí. All participants gave their written characterized by poorer general functioning (i.e. poorer intellec- informed consent after a full and comprehensive explanation of tual ability; Pu et al., 2018; Vicent-Gil et al., 2018). the study. Recent studies have reported an association of cognitive dys- function with worse psychosocial functioning, both at work per- Clinical and demographic assessment formance and in social relationships (Clark, DiBenedetti, & Perez, 2016; Cotrena et al., 2016; Evans, Iverson, Yatham, & Lam, 2014). Clinical and demographic variables were obtained during a semi- Considering the above, it might be useful to identify those structured interview, which covered age, sex, years of schooling, patients with cognitive impairment during an acute episode of age at illness onset, number of episodes and duration of illness. depression, to try to cope with cognitive deficits while at the Medication use at the time of evaluation was categorized as: no same time improving the patients’ psychosocial functioning with- medication, monotherapy with antidepressants, antidepressants out having to wait to treat those deficits after clinical remission. So plus benzodiazepines and combination of antidepressants with far, only one of the studies mentioned in a sample of first episode one or more psychotropic drugs (e.g. antipsychotics, lithium of depression has analyzed neuropsychological heterogeneity in and anticonvulsants). Depressive symptomatology was assessed the acute stage of the illness revealing two distinguishable cogni- using the 17-item Hamilton Depression Rating Scale (HDRS-17; tive profiles (preserved and impaired clusters) (Vicent-Gil et al., Bobes et al. 2003; Hamilton, 1960). The Maudsley Staging 2018). Some patients were classified as cognitively impaired show- Method (MSM; Fekadu, Wooderson, Markopoulou, & Cleare, ing significant deficits in attention/working memory and verbal 2009) was administered to assess the level of treatment resistance. memory and subtle impairment in executive function, whereas MSM includes information about duration and severity of depres- the rest did not show any cognitive deficit. Unfortunately, cogni- sion, antidepressant treatments, augmentation strategies and elec- tion is not routinely evaluated in all MDD patients [as reported by troconvulsive therapy providing two categories: non-resistant McAllister-Williams et al. (2017) in the UK], and the presence of (scores 3–6) and resistant (scores 7–15). cognitive dysfunction in a non-negligible percentage of them still represents an unresolved problem, which derives in increased Neuropsychological assessment social and health costs. Even though there seems to be some evidence about the The neuropsychological battery included the following tests: Digit heterogeneity in cognitive functioning among MDD patients Span of the Wechsler Adult Intelligence Scale version IV (see Douglas et al., 2018), no studies have attempted to identify (WAIS-IV; Wechsler, 2008); Rey Auditory Verbal Learning Test subgroups of acute patients considering the factors mentioned (RAVLT; Rey, 1941); Digit Symbol Substitution Test (DSST; above. The aim of the current study is to identify clusters of WAIS-IV); Trail Making Test Part A (TMT-A), and Trail MDD patients with an acute episode using cognitive, clinical Making Test Part B (TMT-B; Tombaugh, 2004); Wisconsin and sociodemographic measures as indicators of grouping. We Card Sorting Test (WCST; Heaton, 1981); similarities subtest hypothesize that different cluster groups will emerge based on (WAIS-IV); semantic verbal fluency test (category fluency; their cognitive, clinical and sociodemographic characteristics. In Benton & Hamsher, 1976; Peña-Casanova et al., 2009) and phon- addition, we also expect that most cognitively affected patients emic verbal fluency test (PMR, adapted for Spanish speaking will show worse psychosocial adaptation. population; Casals-Coll et al., 2013; Peña-Casanova et al., 2009). Premorbid intelligence (estimated IQ) was assessed with Vocabulary Subtest of the WAIS-IV and it was used to compare Methods cognitive profiles after the clustering analysis. Participants Functional assessment A sample of 174 participants aged 18–65 years old was selected from the outpatient unit at the Psychiatry Department of the Functioning Assessment Short Test (FAST; Rosa et al., 2007) was Hospital Universitari Parc Taulí, from part of a broader project used to evaluate autonomy, occupational functioning, cognitive 2888 Muriel Vicent‐Gil et al. Fig. 1. Clustering summary: relative importance of each indicator. functioning, financial issues, interpersonal relationships and leis- clusters measured with the Silhouette’s index, (ii) the best clinical ure time. The scores range from 0 to 72 with higher values indi- coherence and (iii) an adequate number of patients in each cluster cating more disability. A score of ⩾12 represents a mild to severe to facilitate statistical analyses of comparison. Also, as a final functional impairment (Bonnín et al., 2018). result, the analyses provide a ranking of the importance of each predictor entered in the model. The greater the importance meas- ure, the more relevant the variable is considered in the formation Data analyses of the cluster. Data were analyzed using the Statistical Package for Social Demographic, clinical and functional variables were analyzed Sciences (SPSS), version 21. Neuropsychological raw scores were among resulting clusters in each group of patients by means of converted to z-scores using normative data. To reduce the num- one-way analysis of variances or chi-square when appropriate, ber of neuropsychological variables (Miskowiak et al., 2017; and effect-sizes were reported, as well. Mur, Portella, Martínez-Arán, Pifarré, & Vieta, 2007), cognitive domains were defined using a principal component analysis Results (PCA). The number of retained components was decided upon the resulting scree plot, admitting eigenvalues above or close to The PCA of neuropsychological data extracted four orthogonal 1 (based on theoretically-driven decision). These components dimensions that corresponded to four cognitive domains and (cognitive domains) were then rotated and used for a cluster explained a 74% of cumulative variance. The four cognitive analysis. domains were: (i) Attention/Working Memory domain, which To identify homogeneous subgroups of patients, a two-step included the forward and backward Digit Span; (ii) Verbal clustering analysis was carried out based on cognitive domains, Memory domain, composed of RAVLT first trial, immediate recall stage of treatment-resistance (MSM), depressive symptomatology and delayed recall; (iii) Executive Function domain, with TMT (HDRS-17), number of depressive episodes, age and years of Part A, TMT Part B, DSST and number of categories from schooling. This two-step analysis is designed to use categorical WCST and (iv) Verbal Ability domain which included PMR, and continuous variables in large samples. It represents an extra semantic fluency and similarities. See online Supplementary value to previous studies as other relevant variables such as treat- Table 1 with a brief summary of outcome measures for each test. ment response can be included. During the first step, subjects are The two-step clustering analysis resulted in a three-cluster preclustered into small subgroups using a sequential-clustering solution, which was selected as being the most optimal one approach. In the second step, subclusters from the first are entered based on a fair Silhouette’s index (equal to 0.3, see online as inputs and grouped into the best number of clusters according Supplementary Fig. 1), on the clinical interpretability from previ- to a hierarchical-clustering method (Norusis, 2011). The deter- ous studies, and on the number of patients in each cluster. As can mination of the optimal numbers of clusters is based on the be observed from Fig. 1, the stage of treatment resistance obtained Akaike Information Criterion (AIC) and the log-likelihood dis- the highest relevance for clustering with a predictor importance of tance, taking as the best solution the large ratio of AIC changes 1.0, followed by verbal ability, executive function, attention/work- and the large ratio of distance measures. The final model is ing memory and verbal memory with values ranging between 0.8 based on different criteria (Nylund, Asparouhov, & Muthén, and 0.5. Depressive symptomatology, years of schooling, age and 2007): (i) the highest cohesion and separation of the resulted number of depressive episodes obtained the lowest relevance for Psychological Medicine 2889 Table 1. Clustering summary: centroids depressive episodes, age and years of schooling. Treatment resistance was the variable with the highest importance of clus- C1 C2 C3 Combined tering, closely followed by cognitive performance (verbal abil- Categorical indicators ity, executive function, attention/working memory and verbal memory). The rest of variables obtained the lowest importance MSM non-resistanta 80% 0% 20% 100% on identifying distinct subgroups of patients. Three empirical a MSM-resistant 0% 74.2% 25.8% 100% clusters were determined: cluster 1, characterized by non- Quantitative indicators (means) resistant patients with preserved cognitive performance; cluster 2, composed of resistant patients with selective impairment and Verbal ability −0.46 −0.71 −1.86 −0.88 cluster 3, grouped by resistant and non-resistant patients with a Executive function −0.45 −0.79 −1.83 −0.9 global cognitive impairment. As hypothesized, the latest was Attention/working −0.51 −0.74 −1.71 −0.88 related to worse clinical and psychosocial outcomes. These memory findings may indicate the existence of different subgroups of Verbal memory −0.49 −1.07 −1.92 −1.04 patients, determined by clinical variables – as well-established in the literature – and by cognitive symptoms, which have HDRS-17 17.15 19.33 23.25 19.38 not received enough attention for decades and may be under- Years of schooling 10.5 9.97 8.05 9.74 lying poor outcomes. Age 52.16 54.97 51.95 53.18 To our knowledge, this study is the first to show different cog- nitive profiles during an acute phase of the illness, considering not Depressive episodes 2.19 2.41 2.43 1.35 only cognitive performance but also clinical and sociodemo- MSM, Maudsley Staging Method; HDRS, Hamilton Depression Rating Scale. graphic factors, which are likely to have contributed to divergent a MSM non-resistant (scores 3–6), MSM-resistant (scores 7–15). results in the last few decades. Hammar and Ardal (2009) already suggested that no single cognitive functioning profile could char- acterize all depressed patients, and that not all patients were to be impaired in the same degree during the acute phase (Hammar & clustering. Table 1 presents the centroids for the predictors, mean Ardal, 2009). The existence of such heterogeneity among mood values of quantitative variables and percentage distribution for the disorders’ cognitive functioning is supported by previous cluster categorical variable treatment resistance. analyses, especially in bipolar disorder (Burdick et al., 2014; Figure 2 illustrates a radar chart that shows the profile of the Cotrena, Branco, Ponsoni, Shansis, & Fonseca, 2017; Lima et al., clusters within the different predictors of the model (quantitative 2019; Solé et al., 2016). These studies found three-cluster solu- variables were transformed into z-scores for a better comprehen- tions based on cognitive performance, corresponding to intact sion of the figure). The first cluster (C1) was classified as cogni- or preserved, selectively impaired and globally impaired patients. tively preserved (n = 68, 39%) and it was characterized by a Only two studies have been carried out clustering analysis with sample of non-resistant patients. The second cluster (C2) exclusively MDD patients, and their findings point toward cogni- included 66 patients (38%) classified as selectively impaired and tive heterogeneity along the different stages of the disorder. One all patients were treatment-resistant. And cluster 3 (C3) included included patients in partial remission and reported three clusters globally impaired patients (n = 40, 23%) with significant deficits (Pu et al., 2018) in accordance with the above-mentioned studies, in all neurocognitive domains, being 23 patients classified as and the other was centered into first-episode patients and showed resistant (58%) and 17 as non-resistant (42%). two clusters, one preserved and one impaired (Vicent-Gil et al., Tables 2 and 3 display comparisons among the three clusters 2018). These previous studies, except for the one with first- for demographic, clinical and functional variables and for co- episode patients, endorse our current findings of a subgroup of gnitive tests. The three clusters differed in terms of cognitive per- intact or mostly preserved patients, of a globally impaired sub- formance (see online Supplementary Fig. 2), in which C1 patients group with a general cognitive affectation, and of a subgroup (cognitively preserved) scored within normal range in almost all with specific domains impaired. Although some of the studies the tests (with the exception of number of categories). C2 (Burdick et al., 2014; Pu et al., 2018) claim to have found ‘discrete patients (selectively impaired) obtained scores below 1S.D. in neurocognitive subgroups’, this may be straightforward for pre- specific tests evaluating memory and executive functioning. served and globally impaired patients, because these individuals And C3 patients (globally impaired) showed significant altera- can be easily detected even in clinical settings. By contrast, the tions in almost all the neuropsychological tests (with z-scores selectively impaired subjects may not constitute a clearly discrete ranging between −1 and −2). Post-hoc comparisons of C3 (glo- neurocognitive subgroup given that other characteristics may bally impaired patients) to C1 and C2 showed large effect sizes interact with cognition making difficult to detect specific cogni- in all cognitive variables (Cohen’s d > 0.8, with the exception of tive alterations. At this point, our current study highlights the first trial in RAVLT in the contrast C2 v. C3). With regard to importance of adding clinical information to the clustering, as functional assessment, significant differences were also it may be crucial for a more comprehensive classification of observed, in which globally impaired patients showed the patients, by capturing other factors that may blur those patients worst outcomes. in C2. Treatment resistance was the most important variable in the clustering process, which embraces a lack of response (and/or Discussion remission) and greater severity of clinical symptoms. The totality This work explores the existence of empirical clusters for MDD of participants in the cognitively preserved cluster was non- taking into account patients’ cognitive performance, stage of treat- resistant and similarly, the 100% of the participants in the select- ment resistance, depressive symptomatology, number of ively impaired cluster were resistant. In the group of globally 2890 Muriel Vicent‐Gil et al. Fig. 2. Radar chart for the distribution of the indi- cators of the model. HDRS-17: Hamilton Depression Rating Scale. impaired patients, however, up to 42% of patients were non- to moderate the relationship between mood and cognition resistant. A possible explanation is that alterations of memory (Opdebeeck et al., 2017). Thus, cognitive reserve should not be and executive function are more specifically related to treatment neglected when assessing MDD patients. resistance (Pimontel et al., 2016; Rao et al., 2019), as observed By considering both cognitive and other illness-related vari- in C2, in consistence with previous evidence of hippocampus ables, the three clusters may reflect variations of the disorder and prefrontal cortex alterations in treatment-resistant depression due to differences in underlying pathophysiology, in response to (Ge et al., 2019). Thus, a global alteration of cognitive function treatment or in illness trajectories, and not merely in cognitive may be reflecting a different phenotype, in which the main char- subdivisions on a linear continuum (Carruthers, Van Rheenen, acteristic would be cognitive impairment and not exclusively Gurvich, Sumner, & Rossell, 2019). Consequently, the clinical linked to treatment resistance – this may not be unreasonable implication of the present results refers to the necessity of consid- as the majority of antidepressant treatments do not target cogni- ering cognitive functioning in clinical settings in all patients diag- tive symptoms. These results might indicate that although treat- nosed with major depression. First, clinicians should seriously ment resistance would be associated with greater cognitive contemplate addressing cognitive symptoms when a given patient impairment, the existence of various cognitive profiles has to be begins to show resistance to treatment, as cognitive difficulties taken into account beyond the usual clinical variables, as non- may be related to worse treatment outcomes. Therefore, the inter- resistant patients can still display cognitive symptoms. In fact, vention should be directed not only to clinical symptoms, but also even though all patients of the study were acutely depressed, to cognitive dysfunction. Second, 42% of patients who respond 39% showed no cognitive impairment. These findings also help adequately to antidepressants may also present a global cognitive to explain the small-to-moderate effects – depending on the dysfunction. In clinical practice, some patients with a good domain – of cognitive dysfunction in previous studies that com- response to antidepressant treatment complain about a lack of pare patients with MDD with healthy controls, considering the complete recovery and of difficulties to perform daily activities, apriorism of cognitive homogeneity among MDD patients may partly due to their perceived cognitive problems. Detecting such have led to such disperse results. The current study, which is data- unidentified cognitive difficulties is of great importance as they driven, demonstrates the relevant role of cognitive functioning in are associated with worse psychosocial functioning (Cambridge, major depression, pointing at the existence of potential cognitive Knight, Mills, & Baune, 2018; Knight & Baune, 2018) and with dysfunction in a high percentage of patients who will require the low rates of recovery (Groves, Douglas, & Porter, 2018). more tailored treatments. Hence, treating cognitive dysfunction together with clinical symp- Socio-demographic variables have a low relative importance in tomatology in an active episode of depression could probably the cluster formation (Fig. 1). However, years of schooling and IQ result in a better response to treatment. Different pharmacological significantly differ among clusters, where the most impaired strategies (e.g. vortioxetine, duloxetine, modafinil or erythropoi- patients – cluster 3 – had on average 2 years less of schooling etin), non-pharmacological approaches (cognitive remediation and almost one standard deviation less of IQ than the cognitively or aerobic exercise) and neurostimulation interventions have preserved group. Differences in IQ could be tautological as posi- been shown to be effective in the treatment of cognitive dysfunc- tive associations have been described between IQ and cognitive tion while improving psychosocial functioning and quality of life performance. However, in clinical practice it is usually observed (Salagre et al., 2017; Zuckerman et al., 2018). that patients with normal or intact premorbid IQ show relevant The current study was subject to some limitations. First, due to cognitive impairment, and the other way around (patients with cross-sectional design it was not possible to assess the long-term limited IQ who do not show any cognitive deficits). Although glo- stability of the cognitive profiles. Second, the sample included bally impaired patients had the lowest IQ in this study, their intel- outpatients treated in a specialized clinical setting, and may not lectual ability fell within normality, and their cognitive comprise worldwide clinical practice as mild outpatients may be performance was below normality (1.5S.D. below, on average). underrepresented. Third, the mean age of the sample was older All the above-mentioned variables, which are normally used as than other studies, which could have facilitated the inclusion of proxies for cognitive reserve, showed a significant importance in patients with cognitive deficits caused by other conditions (such previous studies of cluster analysis (Pu et al., 2018; Vicent-Gil as neurodegenerative processes). In any case, the presence of a et al., 2018). And cognitive reserve itself has specifically shown neurological condition was an exclusion criterion which was Psychological Medicine Table 2. Mean scores (standard deviation) for demographic and clinical variables across clusters Clusters Statistics Post-hoc Cohen’s da C1 Preserved C2 Selectively C3 Globally (n = 68) impaired (n = 66) impaired (n = 40) F or χ2 p C1 v. C2 C1 v. C3 C2 v. C3 C1 v. C2 C1 v. C3 C2 v. C3 Age (years) 52.16 (8.62) 54.97 (6.22) 51.95 (7.97) 2.94 0.056 0.104 1.0 0.150 0.37 0.03 0.42 Sex, female n (%) 49 (72.1) 40 (60.6) 33 (82.5) 5.9 0.052 0.16 0.22 0.018 0.12 0.12 0.23 Educational level 10.5 (3.36) 9.97 (3.29) 8.05 (2.03) 8.29 <0.001b 0.96 <0.001b 0.006 0.16 0.88 0.7 (years) Estimated IQ, 51.68 (8.27) 49.52 (7.49) 43.3 (3.67) 17.63 <0.001b 0.246 <0.001b <0.001b 0.27 1.31 1.05 T-score Age at illness onset 42.13 (12.12) 42.17 (9.76) 38.8 (12.23) 1.35 0.262 1.0 0.423 0.418 0.00 0.27 0.30 (years) Number of 2.19 (1.42) 2.41 (1.2) 2.43 (1.47) 0.57 0.566 1.0 1.0 1.0 0.17 0.17 0.01 episodes, n Illness duration 127.3 (122.07) 171.68 (110.38) 171.63 (145.52) 2.67 0.073 0.118 0.221 1.0 0.38 0.33 0.00 (months) HDRS-17, total 17.15 (6.03) 19.33 (6.56) 23.25 (4.9) 12.99 <0.001b 0.110 <0.001b 0.004 0.35 1.11 0.68 score Comorbidities, n 20 (29.4) 21 (31.8) 18 (45.0) 0.294 0.230 0.762 0.101 0.173 0.03 0.16 0.13 (%) Anxiety 3 (4.4) 6 (9.1) 6 (15.0) 0.361 0.164 0.279 0.055 0.352 0.09 0.19 0.09 disorders, n (%) Dysthymia, n (%) 17 (25.0) 17 (25.8) 12 (30.0) 0.349 0.840 0.920 0.571 0.635 0.01 0.05 0.05 MSM, resistant n 0 (0) 66 (100) 23 (58) 134.9 <0.001b <0.001b <0.001b <0.001b 1.0 0.68 0.56 (%) FAST, total score 36.49 (13.39) 42.68 (14.19) 47.93 (13.48) 9.17 <0.001b 0.03 <0.001b 0.176 0.45 0.85 0.38 b b Current 18.98 <0.001 <0.001 0.01 0.482 0.38 0.29 0.07 medication, n (%) AD 30 (44.1) 7 (10.6) 8 (20) AD + BZD 22 (32.4) 22 (33.3) 11 (27.5) AD + Others 14 (20.6) 37 (56.1) 20 (50) Medication-free 2 (2.9) 0 (0) 1 (2.5) IQ, intelligence quotient; HDRS, Hamilton Depression Rating Scale; MSM, Maudsley Staging Method; FAST, Functional Assessment Short Test; AD, antidepressant; BZD, benzodiazepines; others, antipsychotics; lithium or anticonvulsants. a Cramer’s V for categorical variables. b Analysis of variance or χ2 tests statistically significant after applying Bonferroni correction for multiple comparisons ( p < 0.004). 2891 2892 Table 3. Mean scores (standard deviation) for cognitive variables across clusters Clusters Statistics Post-hoc Cohen’s d C1 Preserved C2 Selectively C3 Globally (n = 68) impaired (n = 66) impaired (n = 40) F p C1 v. C2 C1 v. C3 C2 v. C3 C1 v. C2 C1 v. C3 C2 v. C3 Attention/working −0.51 (−0.88) −0.74 (0.62) −1.71 (0.45) 54.41 <0.001a 0.071 <0.001a <0.001a 0.30 1.72 1.79 memory Forward digit −0.7 (0.66) −0.89 (0.77) −1.9 (0.59) 41.32 <0.001a 0.331 <0.001a <0.001a 0.26 1.92 1.47 (WAIS-IV) Backward digit −0.33 (0.79) −0.6 (0.75) −1.5 (0.54) 34.80 <0.001a 0.091 <0.001a <0.001a 0.35 1.73 1.38 (WAIS-IV) Verbal memory −0.49 (0.74) −1.07 (0.72) −1.92 (0.85) 44.52 <0.001a <0.001a <0.001a <0.001a 2.14 1.79 1.08 First trial (RAVLT) −0.65 (0.81) −0.92 (0.78) −1.49 (1.14) 11.22 <0.001 a 0.251 <0.001 a 0.005 0.34 0.85 0.58 Immediate recall −0.56 (0.87) −1.26 (1.0) −2.38 (0.98) 46.42 <0.001a <0.001a <0.001a <0.001a 0.75 1.96 1.13 (RAVLT) Delayed recall −0.27 (0.88) −1.02 (0.92) −1.89 (0.96) 40.53 <0.001a <0.001a <0.001a <0.001a 0.83 1.76 0.93 (RAVLT) Executive function −0.45 (0.72) −0.79 (0.63) −1.83 (0.48) 60.9 <0.001a 0.007 <0.001a <0.001a 0.50 2.26 1.86 TMT part A −0.27 (0.97) −0.5 (0.87) −1.76 (0.82) 37.61 <0.001 a 0.426 <0.001 a <0.001 a 0.25 1.66 1.49 TMT part B −0.01 (1.05) −0.61 (0.93) −1.98 (0.62) 38.66 <0.001a 0.003 <0.001a <0.001a 0.60 2.28 1.73 DSST (WAIS-IV) −0.29 (0.88) −0.66 (0.74) −1.82 (0.76) 47.01 <0.001 a 0.022 <0.001 a <0.001 a 0.46 1.86 1.55 Categories (WCST) −1.11 (0.82) −1.34 (0.76) −1.84 (0.44) 12.01 <0.001 a 0.240 <0.001 a 0.003 0.29 1.11 0.81 Verbal ability −0.46 (0.64) −0.71 (0.67) −1.86 (0.48) 68.06 <0.001 a 0.06 <0.001 a <0.001 a 0.38 2.47 1.97 Phonemic fluency −0.45 (0.77) −0.67 (0.73) −1.68 (0.66) 37.58 <0.001a 0.255 <0.001a <0.001a 0.29 1.72 1.45 (PMR) Semantic fluency −0.6 (0.86) −0.98 (0.83) −1.97 (0.78) 34.31 <0.001a 0.025 <0.001a <0.001a 0.45 1.67 1.23 (Animals) Similarities −0.34 (1.0) −0.49 (1.02) −1.93 (0.82) 37.12 <0.001a 1.0 <0.001a <0.001a 0.15 1.74 1.56 (WAIS-IV) Muriel Vicent‐Gil et al. WAIS: Wechsler Adult Intelligence Scale, RAVLT: Rey Auditory Verbal Learning Test, TMT: Trail Making Test, DSST: Digit Symbol Substitution Test, WCST: Wisconsin Card Sorting Test. a Analysis of variance or χ2 tests statistically significant after applying Bonferroni correction for multiple comparisons ( p < 0.003). Psychological Medicine 2893 strictly evaluated. Fourth, although two-step clustering analysis is Ethical standards. The authors assert that all procedures contributing to one of the most robust techniques to classify individuals (as it this work comply with the ethical standards of the relevant national and insti- combines k-means and hierarchical approaches), the generaliz- tutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. ability of the findings is one of the main drawbacks as the results are very sample-specific. The lack of external replication with an independent dataset represents a limitation of these results. In any case, our findings are fairly similar to the scarce literature. Fifth, References individual scores were corrected with demographic-adjusted Benton, A. L., & Hamsher, K. (1976). Multilingual aphasia examination. Iowa norms from a similar population, which might affect the inter- City: University of Iowa. pretation of the present findings; therefore, future research should Biringer, E., Mykletun, A., Sundet, K., Kroken, R., Stordal, K. I., & Lund, A. consider the cognitive heterogeneity within the norm samples. To (2007). A longitudinal analysis of neurocognitive function in unipolar depres- overcome this limitation, composite scores for cognitive perform- sion. Journal of Clinical and Experimental Neuropsychology, 29, 879–891. ance were used as a more objective patient’s cognitive perform- Bobes, J., Bulbena, A., Luque, A., Dal-Ré, R., Ballesteros, J., & Ibarra, N. (2003). ance outcome compared to the use of single tests (Miskowiak Evaluación psicométrica comparativa de las versiones en español de 6, 17 y et al., 2017). Sixth, the variety of impaired cognitive domains 21 ítems de la escala de valoración de Hamilton para la evaluación de la within the ‘selectively impaired cluster’, when compared to depresión. Medicina Clinica, 120, 693–700. Bonnín, C. M., Martínez-Arán, A., Reinares, M., Valentí, M., Solé, B., Jiménez, other studies, reflects another kind of heterogeneity that cannot E., … Rosa, A. R. (2018). Thresholds for severity, remission and recovery be resolved with cluster analysis, as the solution depends on the using the functioning assessment short test (FAST) in bipolar disorder. samples used in each study. However, by using a two-step cluster- Journal of Affective Disorders, 240, 57–62. ing, we detected two clear and extreme subgroups (i.e. preserved Bora, E., Harrison, B. J., Yücel, M., & Pantelis, C. (2013). Cognitive impair- and globally impaired) together with a subgroup of selective ment in euthymic major depressive disorder: A meta-analysis. impairment, in which other factors, beyond cognition, help to Psychological Medicine, 43, 2017–2026. better characterize them. Finally, concomitant medication could Burdick, K. E., Russo, M., Frangou, S., Mahon, K., Braga, R. J., Shanahan, M., be a possible confounder because of its side effects on cognition. & Malhotra, A. K. (2014). Empirical evidence for discrete neurocognitive But due to clinical reasons and ethical concerns, it was not subgroups in bipolar disorder. Psychological Medicine, 44, 3083–3096. adequate to discontinue the medication. Cambridge, O. R., Knight, M. J., Mills, N., & Baune, B. T. (2018). The clinical relationship between cognitive impairment and psychosocial functioning in In conclusion, the current study shows the existence of distin- major depressive disorder: A systematic review. Psychiatry Research, 269, guishable subgroups in a sample of acute depressed patients, in 157–171. which treatment resistance and cognitive performance are rele- Carruthers, S. P., Van Rheenen, T. E., Gurvich, C., Sumner, P. J., & Rossell, S. vant factors to take into account. The current design provides evi- L. (2019). Characterising the structure of cognitive heterogeneity in schizo- dence for the heterogeneity of cognitive dysfunction in MDD. phrenia spectrum disorder: a systematic review and narrative synthesis. Therefore, future clinical research should consider the existence Neuroscience and Biobehavioral Reviews, 107, 252–278. of potential cognitive dysfunction in all MDD patients, to tailor Casals-Coll, M., Sánchez-Benavides, G., Quintana, M., Manero, R. M., new strategies to achieve a full clinical and functional recovery Rognoni, T., Calvo, L., … Peña-Casanova, J. (2013). Estudios normativos and to prevent new relapses. españoles en población adulta joven (proyecto NEURONORMA jóvenes): normas para los test de fluencia verbal. Neurologia (Barcelona, Spain), 28, 33–40. Supplementary material. The supplementary material for this article can Clark, M., DiBenedetti, D., & Perez, V. (2016). Cognitive dysfunction and work be found at https://doi.org/10.1017/S0033291720001567. productivity in major depressive disorder. Expert Review of Acknowledgements. We gratefully acknowledge patients who participated Pharmacoeconomics & Outcomes Research, 16, 455–463. in this study. We also thank the staff of the Department of Psychiatry, Cotrena, C., Branco, L. D., Kochhann, R., Shansis, F. M., & Fonseca, R. P. Hospital Universitari Parc Taulí for their assistance, especially to Dr (2016). Quality of life, functioning and cognition in bipolar disorder and Figuereo, Dr Ros, Dr Acebillo and Dr Coronas for their kind cooperation in major depression: a latent profile analysis. Psychiatry Research, 241, 289–296. patient recruitment. To end, we are most grateful to Roser Granero for her Cotrena, C., Branco, L. D., Ponsoni, A., Shansis, F. M., & Fonseca, R. P. (2017). helpful comments regarding statistical analysis. Neuropsychological clustering in bipolar and major depressive disorder. Journal of the International Neuropsychological Society, 23, 1–10. Financial support. This work was supported by Lundbeck España Dotson, V. M., Resnick, S. M., & Zonderman, A. B. (2008). Differential asso- (Institutional ethical committee # 2015630). ciation of concurrent, baseline, and average depressive symptoms with cog- nitive decline in older adults. American Journal of Geriatric Psychiatry, 16, Conflict of interest. Dr Maria J. Portella is funded by the Instituto Carlos III 318–330. through a research contract Miguel Servet II (CP16/00020) and from the Douglas, K. M., Gallagher, P., Robinson, L. J., Carter, J. D., McIntosh, V. V., Health Research and Innovation Strategic Plan (PERIS SLT006_17_177). She Frampton, C. M., … Porter, R. J. (2018). Prevalence of cognitive impairment has also received honoraria as academic from Lundbeck. Dr Sara Crivillés in major depression and bipolar disorder. Bipolar Disorders, 20, 260–274. has received honoraria as consultant, advisor or CME speaker for Evans, V. C., Iverson, G. L., Yatham, L. N., & Lam, R. W. (2014). The re- Janssen-Cilag, Lundbeck/Otsuka, Lab Esteve and Lab Servier. Dr Eva lationship between neurocognitive and psychosocial functioning in major Aguilar has received grants and served as consultant, advisor or CME speaker depressive disorder: A systematic review. Journal of Clinical Psychiatry, for Janssen-Cilag and Lundbeck/Otsuka. Dr Diego Palao has received research 75, 1359–1370. funding from the Spanish Ministry of Science and Innovation, Instituto Carlos Fekadu, A., Wooderson, S. C., Markopoulou, K., & Cleare, A. J. (2009). The III and from Health Research and Innovation Strategic Plan (PERIS). Dr Maudsley Staging Method for treatment-resistant depression: Prediction Narcís Cardoner declares that he has received funding from the Universitat of longer-term outcome and persistence of symptoms. Journal of Clinical Autònoma de Barcelona, from the Spanish Ministry of Science and Psychiatry, 70, 952–957. Innovation, Instituto Carlos III and from Health Research and Innovation Ge, R., Torres, I., Brown, J. J., Gregory, E., McLellan, E., Downar, J. H., … Strategic Plan (PERIS) 2016–2020. He has also received honoraria as consult- Vila-Rodriguez, F. (2019). Functional disconnectivity of the hippocampal ant, advisor or CME speaker from Janssen, Lundbeck, Pfizer, Exeltis and MSD network and neural correlates of memory impairment in treatment- within the last 3 years. The rest of authors have nothing to disclose. resistant depression. Journal of Affective Disorders, 253, 248–256. 2894 Muriel Vicent‐Gil et al. Groves, S. J., Douglas, K. M., & Porter, R. J. (2018). A systematic review of cog- Pimontel, M. A., Rindskopf, D., Rutherford, B. R., Brown, P. J., Roose, S. P., & nitive predictors of treatment outcome in major depression. Frontiers in Sneed, J. R. (2016). A meta-analysis of executive dysfunction and anti- Psychiatry, 9, 382. depressant treatment response in late-life depression. The American Hamilton, M. (1960). A rating scale for depression. Journal of Neurology Journal of Geriatric Psychiatry, 24, 31–41. Neurosurgery & Psychiatry, 23, 56–62. Pu, S., Noda, T., Setoyama, S., & Nakagome, K. (2018). Empirical evidence for Hammar, A., & Ardal, G. (2009). Cognitive functioning in major depression – discrete neurocognitive subgroups in patients with non-psychotic major a summary. Frontiers in Human Neuroscience, 3, 1–7. depressive disorder: clinical implications. Psychological Medicine, 48, Hasselbalch, B. J., Knorr, U., & Kessing, L. V. (2011). Cognitive impairment in 2717–2729. the remitted state of unipolar depressive disorder: a systematic review. Rao, D., Xu, G., Lu, Z., Liang, H., Lin, K., & Tang, M. (2019). Comparative Journal of Affective Disorders, 134, 20–31. study of cognitive function between treatment-resistant depressive patients Heaton, R. K. (1981). Wisconsin card sorting test manual. Odessa: and first-episode depressive patients. Neuropsychiatric Disease and Psychological Assessment Resources. Treatment, 15, 3411–3417. Iverson, G. L., Brooks, B. L., Langenecker, S. A., & Young, A. H. (2011). Rey, A. (1941). L’examen clinique en psychologie. Paris: Presses Universitaires Identifying a cognitive impairment subgroup in adults with mood disor- de France. ders. Journal of Affective Disorders, 132, 360–367. Rosa, A. R., Sánchez-Moreno, J., Martínez-Aran, A., Salamero, M., Torrent, C., Knight, M. J., & Baune, B. T. (2018). Cognitive dysfunction in major depressive Reinares, M., … Vieta, E. (2007). Validity and reliability of the functioning disorder. Current Opinion in Psychiatry, 31, 26–31. assessment short test (FAST) in bipolar disorder. Clinical Practice and Lee, R. S. C., Hermens, D. F., Porter, M. A., & Redoblado-Hodge, M. A. (2012). Epidemiology in Mental Health, 3, 1–10. A meta-analysis of cognitive deficits in first-episode major depressive dis- Salagre, E., Solé, B., Tomioka, Y., Fernandes, B. S., Hidalgo-Mazzei, D., order. Journal of Affective Disorders, 140, 113–124. Garriga, M., … Grande, I. (2017). Treatment of neurocognitive symptoms Lima, F., Rabelo-da-Pontea, F. D., Bückera, J., Czepielewskia, L., Hasse-Sousaa, in unipolar depression: a systematic review and future perspectives. M., Telescaa, R., … Rosa, A. R. (2019). Identifying cognitive subgroups in Journal of Affective Disorders, 221, 205–221. bipolar disorder: A cluster analysis. Journal of Affective Disorders, 246, Semkovska, M., Quinlivan, L., O’Grady, T., Johnson, R., Collins, A., O’Connor, 252–261. J., … Gload, T. (2019). Cognitive function following a major depressive McAllister-Williams, R. H., Bones, K., Goodwin, G. M., Harrison, J., Katona, episode: a systematic review and meta-analysis. The Lancet Psychiatry, 6, C., Rasmussen, J. S., … Young, A. H. (2017). Analysing UK clinicians’ 851–861. understanding of cognitive symptoms in major depression: a survey of pri- Serra-Blasco, M., de Vita, S., Rodríguez, M. R., de Diego-Adeliño, J., mary care physicians and psychiatrists. Journal of Affective Disorders, 207, Puigdemont, D., Martín-Blanco, A., … Portella, M. J. (2015). Cognitive 346–352. functioning after deep brain stimulation in subcallosal cingulate gyrus for McDermott, L. M., & Ebmeier, K. P. (2009). A meta-analysis of depression treatment-resistant depression: an exploratory study. Psychiatry Research, severity and cognitive function. Journal of Affective Disorders, 119, 1–8. 225, 341–346. Miskowiak, K. W., Burdick, K. E., Martinez-Aran, A., Bonnin, C. M., Bowie, C. Serra-Blasco, M., & Lam, R. W. (2019). Clinical and functional characteristics R., Carvalho, A. F., … Vieta, E. (2017). Methodological recommendations of cognitive dysfunction in major depressive disorder. In C. J. Harmer & B. for cognition trials in bipolar disorder by the International Society for T. Baune (Eds.), Cognitive dimensions of Major depressive disorder (pp. 45–58). Bipolar Disorders Targeting Cognition Task Force. Bipolar Disorders, 19, 1–13. Oxford: Oxford University Press. Mur, M., Portella, M. J., Martínez-Arán, A., Pifarré, J., & Vieta, E. (2007). Serra-Blasco, M., Torres, I. J., Vicent-Gil, M., Goldberg, X., Navarra-Ventura, Persistent neuropsychological deficit in euthymic bipolar patients: executive G., Aguilar, E., … Cardoner, N. (2019). Discrepancy between objective and function as a core deficit. Journal of Clinical Psychiatry, 68, 1078–1086. subjective cognition in major depressive disorder. European Murrough, J. W., Burdick, K. E., Levitch, C. F., Perez, A. M., Brallier, J. W., Neuropsychopharmacology, 29, 46–56. Chang, L. C., … Iosifescu, D. V. (2015). Neurocognitive effects of Solé, B., Jiménez, E., Torrent, C., del Mar Bonnin, C., Torres, I., Reinares, M., Ketamine and association with antidepressant response in individuals … Martínez-Arán, A. (2016). Cognitive variability in bipolar II disorder: with treatment-resistant depression: a randomized controlled trial. who is cognitively impaired and who is preserved. Bipolar Disorder, 18, Neuropsychopharmacology, 40, 1084–1090. 288–299. Norusis, M. (2011). IBM SPSS statistics 19 procedures companion. Reading, Tombaugh, T. N. (2004). Trail making test A and B: Normative data stratified MA: Addison. by age and education. Archives of Clinical Neuropsychology, 19, 203–214. Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the num- Venezia, R. G., Gorlyn, M., Burke, A. K., Oquendo, M. A., Mann, J. J., & Keilp, ber of classes in latent class analysis and growth mixture modeling: A J. G. (2018). The impact of cognitive reserve on neurocognitive performance Monte Carlo simulation study. Structural Equation Modeling: A in major depressive disorder. Psychiatry Research, 270, 211–218. Multidisciplinary Journal, 14, 535–569. Vicent-Gil, M., Keymer-Gausset, A., Serra-Blasco, M., Carceller-Sindreu, M., Opdebeeck, C., Matthews, F. E., Wu, Y.-T., Woods, R. T., Brayne, C., & Clare, de Diego-Adeliño, J., Trujols, J., … Portella, M. J. (2018). Cognitive predic- L. (2017). Cognitive reserve as a moderator of the negative association tors of illness course at 12 months after first-episode of depression. between mood and cognition: Evidence from a population-representative European Neuropsychopharmacology, 28, 529–537. cohort. Psychological Medicine, 48, 1–11. Wechsler, D. (2008). Weschler adult intelligence scale (WAIS-IV) (4th ed.). San Peña-Casanova, J., Quinones-Ubeda, S., Gramunt-Fombuena, N., Quintana- Antonio, Texas: Pearson. Aparicio, M., Aguilar, M., Badenes, D., … Blesa, R. (2009). Spanish multi- Zuckerman, H., Pan, Z., Park, C., Brietzke, E., Musial, N., Shariq, A. S., … center normative studies (NEURONORMA project): norms for verbal McIntyre, R. S. (2018). Recognition and treatment of cognitive dysfunction fluency tests. Archives of Clinical Neuropsychology, 24, 395–411. in major depressive disorder. Frontiers in Psychiatry, 9, 655.