The authors of this paper do not have any commercial associations that might pose a conflict of interest in connection with this manuscript.
Eduard Vieta, MD, PhD, Institute of Neurosciences, University Hospital Clinic of Barcelona, Villarroel 170, 08036 Barcelona, Spain. Fax: +34 93 2275477; e-mail: firstname.lastname@example.org
Introduction: Few studies have examined the clinical, neuropsychological and pharmacological factors involved in the functional outcome of bipolar disorder despite the gap between clinical and functional recovery.
Methods: A sample of 77 euthymic bipolar patients were included in the study. Using an a priori definition of low versus good functional outcome, based on the psychosocial items of the Global Assessment of Functioning (GAF, DSM-IV), and taking also into account their occupational adaptation, the patients were divided into two groups: good or low occupational functioning. Patients with high (n = 46) and low (n = 31) functioning were compared on several clinical, neuropsychological and pharmacological variables and the two patient groups were contrasted with healthy controls (n = 35) on cognitive performance.
Results: High- and low-functioning groups did not differ with respect to clinical variables. However, bipolar patients in general showed poorer cognitive performance than healthy controls. This was most evident in low-functioning patients and in particular on verbal memory and executive function measures.
Conclusions: Low-functioning patients were cognitively more impaired than highly functioning patients on verbal recall and executive functions. The variable that best predicted psychosocial functioning in bipolar patients was verbal memory.
Bipolar disorder has traditionally been associated with a better outcome than schizophrenia because of a presumed absence of cognitive impairment and seemingly normal functioning between episodes (1). Thus, generally little attention has been given to psychosocial outcomes in patients with bipolar disorder (2). However, in contrast to early studies (1–3), recent studies point to a significant degree of psychosocial dysfunction even when patients are euthymic (4–10). Functioning is a complex concept since it involves the capacity to work, study, live independently and engage in recreation and romantic life (2). Functional recovery has been described as the ability to achieve the level of functioning prior to the most recent episode (6). A 2-year follow-up study on hospitalized patients with a first manic episode with psychotic symptoms found that most patients (97%) achieved syndromal recovery at 24-month follow-up, whereas only 37% achieved functional recovery (6). These findings suggest that there is a gap between clinical and functional outcome in bipolar disorder, which emerges at the very outset of the illness. Hence, bipolar disorder can negatively impact the individual, reducing health-related quality of life and functioning, including employment and work productivity.
The complex nature of bipolar illness may complicate the measurement of impairment (11); however, several factors have been associated with poor functional outcome, such as comorbid substance abuse (12), the side-effects of medication (2), prior history of psychotic symptoms (5, 6), low premorbid functioning (13) persistent subsyndromal fluctuations (14–17), the number of prior episodes (5, 18), prior admissions (19, 20), younger age of onset (6) and persistent cognitive dysfunction (5, 15, 16, 21–24). However, the factors associated with functional outcome vary between studies and have been difficult to replicate because of varied approaches and the use of differing instruments to assess functional outcome. In this regard, cognitive function has not been routinely examined in the studies of psychosocial outcome in patients with bipolar disorder because clinical variables have been thought to impact function more than cognition (25).
The main aim of the present study was to identify factors, among clinical, neuropsychological and pharmacological variables, which could be associated with the functional outcome of bipolar patients. Our hypothesis was that cognitive impairment, and especially verbal memory dysfunctions, would predict the low psychosocial functioning in patients and do so better than clinical variables. This hypothesis was based on previous findings from the scarce literature investigating the relationship between cognition and functioning. Several authors found a correlation between verbal learning and memory measures and functioning in bipolar patients (15, 16, 21–23).
In contrast to previous research (15, 16), the current study is directly focused on the psychosocial outcome of bipolar patients as the primary objective, involves a larger sample of euthymic bipolar patients that form part of a longitudinal study on clinical course, neuropsychological functioning and functional outcome within our Barcelona Bipolar Disorders Program.
Patients participating in this study were enrolled at the Bipolar Disorders Program of the University Hospital Clinic of Barcelona. All patients met DSM-IV criteria for bipolar I or II disorder and were euthymic. The clinical state of the patients was determined by a psychiatrist responsible for the follow-up of bipolar patients in the Bipolar Disorders Program using DSM-IV criteria, and the Structured Clinical Interview for Diagnostic Symptoms (SCID, DSM-IV), the Spanish version of the Hamilton Depression Rating Scale (HDRS, 17-item) (26, 27) and the Young Mania Rating Scale (YMRS) (28, 29). Euthymia was defined as YMRS ≤ 6 and HDRS ≤ 8, during monthly visits over a 6-month period. After screening for inclusion, 77 bipolar patients and 35 healthy controls were enrolled. The reasons for not entering the study were history of head injury or LOC, neurological illness, substance dependence (in the last year), mental retardation, significant medical illness, electroconvulsive therapy in the last year and subsyndromal fluctuations. A total of 112 were included initially, but 31 met one or more than one exclusion criteria and were not admitted into the present study. The most common exclusion criterion was the presence of persistent subsyndromal symptoms along the prior 6 months, and 22 patients were excluded for this reason. Therefore, 81 bipolar patients met the criteria for remission; of these 77 gave written informed consent to participate in the study after procedures had been fully explained. On the contrary, in 92% of patients, medication was kept constant in the weeks before testing. Thirty-five healthy comparison subjects without psychiatric or neurologic history were also recruited via advertisement and from a known pool of normal volunteers. Controls were screened for Axis I psychiatric disorders using the SCID (DSM-IV), and it was ensured that none had first-degree relatives with bipolar disorder. With respect to the control group, six subjects could not enter the study (two because of history of head injury and four because of anxiety disorders). All subjects gave written informed consent to participate in the study after the procedures had been fully explained. Ethical approval for the study was granted by the Ethics Committee.
Clinical and psychosocial assessment
Clinical variables were collected as part of the Bipolar Disorders Program protocol of the Hospital Clinic of Barcelona. The clinical variables included in this study were the number and type of episodes, duration of the illness (chronicity), age at onset of the illness, number of hospitalizations, suicide attempts, family history of affective disorders, history of psychotic symptoms, and diagnostic type I or II.
Overall functioning status was assessed using the Global Assessment of Functioning scale (GAF, DSM-IV). The original GAF instructions call for rating symptoms or functioning. Since many other measures of mood symptoms were obtained as part of the evaluation, the rater was instructed to use the GAF to measure psychosocial functioning in the month prior to rating, as reported in previous studies (16, 30), because it was expected that after 6 months of sustained clinical remission, the functional recovery would be at the best point to be assessed.
Second, occupational adaptation, as an additional measure of functional outcome, was established as ‘good occupational adaptation’ when patients were working at a good or acceptable level of functioning. ‘Low occupational adaptation’ was determined as not working at all or showing moderate or severe difficulties in their job. This measure references the last 3 years prior to evaluation, a span of time significant enough to not be influenced by the potential impact of long episodes. This information was provided by the patient himself and confirmed by a first-degree relative or partner. For example, if the patient was working but showed difficulties completing even simple tasks, the patient was classified as having low occupational adaptation, whereas if the patient worked more than 75% of the time and showed only mild work impairment, then the patient was classified as having good occupational functioning. This additional measure was useful for confirmation of findings and took part of the structured interview of the protocol (15). The clinical interview, including psychosocial functioning, was conducted by a trained psychiatrist, and the neuropsychological evaluation was carried out by a trained neuropsychologist, blind to the results of the clinical and psychosocial assessments.
The patient group was divided into two groups with respect to psychosocial functioning as assessed through the GAF. A score of 60 was a priori considered as cut-off to distinguish patients with good and low psychosocial functioning. As per DSM-IV, a score higher or equal to 60 represents some mild difficulty in social, occupational or academic activities or satisfactory activity. However, in general, the patient works quite well and has significant interpersonal relationships. Scores below 60 indicate moderate to severe impairment in functioning. The use of this cut-off was previously recommended by earlier studies that used this cut-off to consider poor functioning in severe psychiatric disorders, such as schizophrenia (31–33).
An extensive review of previous literature on this issue guided the choice of neuropsychological tests used in the present study. In order to enhance replication, only tests frequently documented by the neuropsychological literature were employed (34, 35). The neuropsychological battery of tests is fully described elsewhere (16).
• Frontal executive functions: Wisconsin Card Sorting Test (WCST) (37); Stroop Colour-Word Interference test (SCWT) (38); FAS (Controlled Oral Word Association Test) (39). Attention/concentration and mental tracking: Digit subtest (Wechsler Adult Intelligence Scale) (36); Trail Making Test (TMT) (40).
• Verbal learning and memory: California Verbal Learning Test (CVLT) (41).
The three groups [patients with good psychosocial functioning (GAF ≥ 60), patients with low psychosocial functioning (GAF < 60) and healthy controls] were compared on clinical and sociodemographic variables using ANOVA and chi-square test, as appropriate. Scatterplots were performed in order to confirm the appropriateness of the a priori selected cut-off point of the GAF. MANOVA analysis was performed to show the overall differences in neuropsychological tests between groups. Since multiple dependent variables were used, a prior protective MANCOVA analysis was performed with estimated premorbid intelligence as covariate and group as a main factor. Residual subsyndromal depressive symptoms, as assessed using the HDRS, were thought to have a confounding effect. When controlling for low levels of symptoms in affected patients, the results did not significantly vary; hence, this variable was not included as a covariate. Since neuropsychological tests are naturally correlated, this procedure was considered better than Bonferroni inequality correction that would increase type II error. Group differences were tested using one-way ANOVA, followed by Tukey post hoc comparison procedure, when significant main effects were present. Homogeneity of variance was tested with Levene statistics.
In bipolar patients, to identify those variables that would be good predictors of functional outcome, we used a hierarchical regression model. Two regression models were calculated. In the first model, the variables were selected by empirical criteria. Those clinical and neuropsychological variables that significantly correlated with the GAF scores, through Pearson correlations with p < 0.01, were introduced in the model. These variables were entered using a stepwise method. In the second regression model, the selection of variables was theoretical, based on findings from literature. Data analyses were performed using SPSS 10.0 statistical package.
As shown in Table 1, the three groups (high functioning, low functioning and healthy controls) did not differ with respect to sex, age and educational level, but they differed regarding estimated premorbid IQ that was lower in the patient groups. The inspection of the scatterplots confirmed the acceptability of the cut-off established in the literature, as shown in Fig. 1.
Table 1. Demographic, clinical and pharmacological variables
Demographic and clinical variables
Low functioning (n = 31)
High functioning (n = 46)
Control group (n = 35)
Educational level (years)
Estimated premorbid IQ
Age at onset
Hamilton Depression Rating Scale
Young Mania Rating Scale
Number of medications
Prior psychotic symptoms
Bipolar type I
Family history of affective disorder
In regards to clinical variables, we did not find statistical differences between patients with high and low functioning, as shown in Table 1. The patient groups statistically differed from the control group in HDRS scores. No significant differences were found with respect to medication regimens (lithium, carbamazepine, valproate, antidepressants and antipsychotics) between the patient groups, as shown in Table 1. Only two patients were drug-free. Most patients (31/40) were receiving atypical antipsychotics, and no differences were observed between groups (χ2 = 0.94; p = 0.33). Nevertheless, those patients with poorer psychosocial functioning were treated with a higher number of medications compared with patients with better functioning (Table 1).
With regard to neuropsychological variables, results are shown in Table 2. MANCOVA yielded Pillai's F = 2.073, df = 30,182, p = 0.002 for the main effect, indicating that there were overall differences in neuropsychological performance between groups.
Table 2. Neuropsychological performance in bipolar patients with high and low psychosocial functioning
A. Low functioning
B. High functioning
C. Control group
Tukey post hoc tests
CVLT = California Verbal Learning Test; SCWT = Stroop Colour-Word Interference Test; TMT = Trail Making Test; WCST = Wisconsin Card Sorting Test; FAS = Controlled Oral Word Association Test; WAIS = Weschler Adult Intelligence Scale.
Frontal executive function
Attention/concentration and mental tracking
Subtest digits (WAIS)
Verbal learning and memory
List A (total)
Free short recall
Cued short recall
Free delayed recall
Cued delayed recall
For 12 of 15 comparisons, the differences reached statistical significance (p < 0.05). In general, the two patient groups performed poorly on most neuropsychological measures as compared with healthy controls especially in those measures related to semantic verbal fluency (animal naming), verbal learning and memory (CVLT learning task and cued short and delayed recall) and attention and working memory (trail A and Digits task). Moreover, patients with lower psychosocial functioning showed even more impaired performance on CVLT free short and long delay recall and recognition task as well as on the Stroop interference and trail B scores.
Effect sizes are also shown in Table 2 (Cohen's d values). The values obtained indicate that effect sizes were, in general, medium or large (42). Larger effect sizes were observed for the low functioning group when compared with controls. Two regression models were calculated, the former selecting the variables following empirical criteria and the latter theoretically.
Pearson correlations were also used in order to establish which clinical and neuropsychological variables correlated with psychosocial functioning, as measured using the GAF. The variables that were found to correlate with psychosocial outcome were HDRS scores (R = 0.29; p = 0.009), number of medications (R = −0.33; p = 0.004), trail B (R = −0.30; p = 0.007), FAS (R = 0.27; p = 0.02) and all CVLT measures: learning task (R = 0.37; p = 0.001), free short recall (R = 0.31; p = 0.006), cued short recall (R = 0.31; p = 0.006), free delayed recall (R = 0.40; p < 0.001), cued delayed recall (R = 0.36; p = 0.001) and recognition hits (R = 0.27; p = 0.02). No correlations were found between psychosocial outcome and age, chronicity, age of onset, number of hospitalizations or suicide attempts and number or type of episodes. Therefore, those variables that correlated with psychosocial outcome (p < 0.01), as mentioned above, were introduced in the linear regression model using stepwise method. Taking into account the potential impact of subsyndromal symptoms on functioning, this variable was included at a first block as well as the number of medications that also significantly correlated with the GAF scores. Once the effect of these variables was controlled, at a second block, neuropsychological measures that correlated with the GAF scores were included. The variables that better predicted psychosocial functioning in euthymic bipolar patients were CVLT free delayed recall (t = 3.51, β = 0.365, df = 73, p = 0.001) and the number of medications (t = −2.51, β = −0.261, df = 73, p = 0.014). Overall, the model reached significance (F = 11.29, df = 2,73, p < 0.001), with an adjusted r2 = 0.22. Verbal recall was the strongest independent predictor of psychosocial functioning, explaining 16% of variance (p < 0.001). Rerunning of the regression model included clinical variables reported in the literature as potential predictors of functional outcome, specifically, subclinical symptoms, chronicity, number of episodes and number of hospitalizations, which were included in a first block, and the cognitive variables were introduced in a second block. Since the number of medications only accounted for the 4% of the variance, it was taken out of the model. The variable that best predicted functioning was verbal delayed recall (t = 4.21, β = 0.484, df = 60, p < 0.001), which accounted for the 21% of the variance of psychosocial functioning. The other factors did not enter the model. An interesting finding was that very low levels of depressive symptoms only accounted for the 7% of the variance (t = −1.34, β = −0.154, df = 60, p = 0.19). This model accounted for the 28% of the variance of psychosocial functioning (F = 12.59, df = 2,60, p < 0.001).
Occupational adaptation, our secondary measure of functional outcome, was analysed among the patient groups. Neuropsychological performance between patients showing good and low occupational adaptation was analysed by using ANOVA. Results were confirmed when patients with good and low occupational functioning were compared (Fig. 2). Better neuropsychological performance was found in the ‘good occupational adaptation’ group compared with the ‘low occupational adaptation’ group, showing similar results than those obtained using the GAF.
The association of cognitive factors and psychosocial outcomes
The present study reports that euthymic bipolar patients with high functioning do not differ from those with low functioning with respect to clinical variables. However, they differ in terms of medication (low functioning patients received higher number of medications) and cognitive function (low-functioning patients performed worse on verbal memory and executive function measures).
Euthymic bipolar patients performed poorer on new learning as well as on recall than the comparison group. Both groups obtained poorer performance on attentional measures and semantic verbal fluency. Moreover, patients with low functioning were most impaired on measures related to inhibitory control (Stroop interference task) and working memory (trail B), which are associated with frontal executive functions. Low functioning patients showed worse performance on CVLT free delayed recall and recognition measures compared with highly functioning and control groups. These findings suggest that those patients with lower psychosocial functioning showed more generalized verbal memory deficits, even in simpler tasks, indicating difficulties in encoding and retrieval of verbal information. Verbal memory impairment has been suggested as a good predictor of work impairment in euthymic bipolar patients (15, 16).
In the present study, the verbal recovery of information was the variable that best predicted the psychosocial outcome of euthymic bipolar patients. Difficulty remembering long-term information may represent a serious problem for bipolar patients in their occupational functioning as well as in their interpersonal relationships. In general, patients complain about having difficulty in remembering names and conversations both in the distant past and more recently. These results were confirmed using the GAF scores as the main measure as well as the occupational adaptation of patients during the last 3 years prior to inclusion in the present study. Other neuropsychological variables that seemed to correlate with psychosocial functioning, such as verbal fluency and trail B, have been reported as a predictor of outcome (24, 43), but only verbal memory appeared in linear regression analysis. Probably one of the major findings that come out of the regression model is that attention and concentration seem not to be responsible for poor psychosocial functioning.
Recent studies have found verbal memory impairment to be similar in remitted bipolar patients and schizophrenic patients but that executive functions are more impaired in schizophrenics, with bipolar patients showing intermediate performance between schizophrenic and healthy subjects (44). Dixon et al. (45) have observed a shared pattern of executive dysfunctions in manic, depressed and euthymic patients, with respect to strategic thinking, inhibitory control and response initiation, independently of mood state, and it is now apparent that all three phases of the illness demonstrate cognitive deficits (15, 46). However, overall deficits on executive function measures were most evident in manic patients (45). Other recent reports such as a 3-year follow-up study by Balanza-Martinez et al. (47) suggest persistent executive function impairment not only in schizophrenic but also in bipolar patients over time. Interestingly, in this longitudinal study, bipolar patients obtained an even worse endpoint performance on the Stroop interference task compared with schizophrenic patients. Findings from the present study with respect to executive and verbal memory functioning implicate the prefrontal cortex and temporolimbic structures, especially ventromedial areas as well as the amygdala and the hippocampus. In a recent review, verbal memory and executive function have also been proposed as potential cognitive endophenotypes for bipolar disorder (48). On the contrary, the scarcity of studies on healthy first-degree relatives may limit consideration of these cognitive domains as endophenotypes, although delayed verbal recall impairment has been reported in healthy first-degree relatives of bipolar probands (49).
Regarding other research on functional outcome of bipolar disorder, in a study comparing bipolar I and schizophrenic stable outpatients found that bipolar patients did differ from schizophrenic only in 5 of 41 measures related to social functioning, quality of life and cognitive dysfunctions (23). Nevertheless, Ferrier et al. (50) did not find differences between patients with good and poor outcome with regard to neurocognitive functioning, probably because of their own definition of outcome, based on the clinical course of the illness.
The association of clinical factors with psychosocial outcomes and cognition
Clinical variables that had been found to influence psychosocial outcome in bipolar disorder in other studies, such as the number of episodes, age of onset or duration of the illness, did not correlate with GAF scores. Subclinical symptoms and especially low levels of subdepressive symptoms were also controlled in the present study. The patients had very low levels of depressive and manic symptoms. When results were covaried for HDRS scores, the findings did not substantially change. The only neuropsychological variable in which the patient groups did not differ from healthy controls was the Stroop Interference score and recognition hits from the CVLT, as found in a previous study (16). It is likely that low levels of subclinical symptoms may influence some aspects of executive functions, such as inhibitory control and recognition memory which is more impaired when encoding processing fails. On the contrary, significant correlations were found between HDRS and GAF scores, as observed in previous studies (15, 16, 51). However, when this variable was introduced in the linear regression analysis, it failed to enter in the equation. Subsyndromal depressive symptoms may interfere with functioning and quality of life (17). However, although inter-episode symptoms may contribute to poor outcome (52), functional outcome is not appreciably better in studies in which patients have very low levels of residual symptomatology (6, 19, 53). Some studies have suggested a relationship between cognitive impairment and psychosocial functioning (15, 16, 22, 24) and therefore after controlling the effects of clinical and subclinical symptoms in our study, verbal memory impairment was still found to influence social and occupational functioning in bipolar patients (15).
Other clinical variables, such as the history of psychotic symptoms, did not influence psychosocial outcome in our bipolar sample and such discrepancies have been found in other studies, for instance, with respect to the impact of psychotic symptoms on outcome (8, 10, 19, 50, 54).
The association of medication with psychosocial outcomes and cognition
No significant differences on neuropsychological performance were found between patients taking lithium, antidepressants or antipsychotics compared with patients who did not take those medications.
An unexpected finding was the influence of the number of drugs on psychosocial outcome. In previous reports, this variable has been found to have a negative impact on cognitive complaints in euthymic bipolar patients (55). In the present study the more medications the patients received, the greater the psychosocial functioning impairment. Interestingly, some studies of bipolar patients on lithium monotherapy compared with patients on lithium plus other medications showed better occupational functioning in the former group, indicating that the former group may differ clinically from those requiring multiple psychotropic medications (56). In this longitudinal open-label study comparing lithium, neuroleptic, and lithium-neuroleptic therapies, significant differences were found among monotherapy treated, combination therapy treated and untreated bipolar patients in overall functioning. The lowest impairment was found among those on lithium monotherapy, followed by untreated individuals, those on neuroleptic monotherapy and those on combination therapy. The impact of treatment on functional outcome has been widely reviewed in Dean et al. (11). It is still unclear whether the number of medications per se had a direct impact on outcome or whether the impaired patients received a higher number of drugs in an attempt to improve their condition. That is, we do not really know the directionality cause–effect between the number of medications and the functional outcome, and probably patients with low functional outcome were more likely to be prescribed a larger number of drugs. A longitudinal study would help to clarify this important issue.
Our study also highlights the limitations of current treatment strategies regarding their capacity to improve the functioning and reduce the disability of bipolar patients. As much as 44% of a sample of euthymic patients followed up at a specialized program for bipolar disorders had a poor or low functional outcome based on the psychosocial items of the GAF. This poor functioning, even in treated euthymic patients, is one of the main factors that explains why bipolar disorders have been ranked seventh amongst the worldwide causes of non-fatal disease burden, as measured in disability-adjusted life years (DALYs) (57). Moreover, it is consistent with recent studies that highlight the modest impact of available interventions on the functional outcomes of a large proportion of bipolar disorders (58), and call for a research agenda that specifically addresses these issues.
The main limitation of the present study is the partly artificial separation of the patients into two groups based on a cut-off GAF score. However, it is probably useful to distinguish patients with no or mild impairment (highly functioning group) on psychosocial outcome from those patients with severe or moderate impairment (low functioning group). In order to avoid this limitation, a multiple regression analysis was performed including the GAF scores as a continuum variable to confirm results found through MANCOVA. A longitudinal follow-up may provide more information about the progression or lack of progression of cognitive dysfunctions and their impact on psychosocial functioning. Probably, further longitudinal studies should include more specific instruments to assess the functional outcome of bipolar patients. The use of measures related to the emotional processing of information would probably be more associated with the functional outcome than basic cognitive measures, such as attention, memory or executive functions (55). Another relevant issue is the differences between patients and controls with respect to medication that could partly explain the differences on neuropsychological performance. However, although cognitive dysfunctions may be related to the effects of medication, they do not seem to be a primary effect of pharmacological treatment.
The present findings suggest that cognitive factors may contribute to psychosocial outcome in bipolar disorder. Cognitive dysfunctions were found not only in low-functioning patients but also in patients with good psychosocial functioning. However, cognitive impairments are most evident in low-functioning patients. Differences relative to cognitive dysfunctions between high- and low-functioning patients seem to be independent of illness severity. Verbal memory dysfunction seems to be a good predictor of psychosocial outcome in euthymic bipolar patients.
Until recently, cognitive impairment has been underestimated in bipolar patients, and our study amongst others suggests that it should be considered more widely with respect to the long-term management of bipolar disorder. This may mean simplifying pharmacological treatment and selecting combination therapies suitable for long-term prophylaxis, taking into account tolerability, side effects and interactions. The current treatments for bipolar patients have limitations regarding functional recovery. Therefore, psychosocial interventions, such as psychoeducation, family intervention and cognitive rehabilitation (15, 16, 59) should be implemented. It is clear that clinical trials are necessary to establish the efficacy of the different psychological interventions as adjunctives to pharmacological treatment in the improvement of the illness course, cognitive dysfunctions and psychosocial outcome in bipolar disorder.
This study was supported in part by grants from the Spanish Ministry of Health, Instituto de Salud Carlos III, Red de Investigacion en Enfermedades Mentales REM-TAP and FIS no P1050206, Fundació Marató de TV3 (2510/01), and the Stanley Medical Research Institute, Bethesda, MD, USA. We also thank the Biostatistical Department of the Hospital Clinic (Erika Sierra and Llorenç Quinto) for statistical support.