Identifying neuropsychological phenotypes in multiple sclerosis using latent profile analysis.

INTRODUCTION

Cognitive impairment is a disabling feature of Multiple Sclerosis (MS) with a negative impact on quality of life of people with MS (PwMS), especially in youngs [ 1 ]. Cognitive impairment in people with MS is a multifaceted condition, involving different cognitive domains such as attention, verbal memory or visuospatial abilities. Demographic and radiological features underpinning such variability are not completely understood yet [ 2 , 3 ]. The difficulty in assessing the pathological contributors to cognitive impairment could be partly due to the difficulty in defining classes of patients sharing similar cognitive profile.

Besides cognitive impairment, people with MS also have neurobehavioral issues, with fatigue being reported in up to 92%, depression in up to 42% and anxiety in up to 40% [ 4 ]. As for cognitive impairment, also for neurobehavioral symptoms there is no clear‐cut single pathological causative mechanism. Lesion burden, brain atrophy, lesion locations, brain network derangement, and deficits in neurotransmitters‐dependent cortical activity were all associated with neurobehavioral symptoms [ 5 ].

To further complicate the picture, neurobehavioral symptoms are variably associated with cognitive deficits [ 6 ] and, hence, the precise classification of cognitive and psychological impairment is of utmost importance when attempting to define the biology beyond the clinical burden.

Recently, efforts have been made to identify cognitive phenotypes in PwMS using latent profile analysis (LPA), a data‐driven statistical approach allowing to identify clusters of sub‐groups following similar longitudinal trends for specific and preselected variables. First attempts resulted in classification of cognitive phenotypes according to specific cognitive domains affected according to scores obtained at an extensive cognitive assessment [ 7 , 8 ]. These studies provided clinical and radiological features associated with each group, speculating on the dynamic interplay between pathological changes and the appearance of cognitive impairment. However, neurobehavioral features were not included in the LPA and, hence, were only explored as post hoc analysis. Differently, Podda et al. included also neurobehavioral features (i.e., anxiety and depression) in the definition of cognitive profiles [ 9 ]. Unfortunately, after profile identification they were not able to assess pathological substrates underpinning profiles since no MRI data were included nor it was included fatigue as neurobehavioral symptoms.

Here, we applied LPA to cognitive and neurobehavioral data to identify neuropsychological profiles in MS. We also aim to validate these profiles towards clinical and radiological features, with the goal to understand factors driving the observed clinical heterogeneity.

METHODS

Study design and population

This was a single centre cross‐sectional study including retrospective clinical and cognitive assessments stored in a local clinical database at MS Clinical Care and Research Centre of the Federico II University Hospital of Naples, Italy from November 2016 to March 2023, satisfying the following inclusion criteria: (1) MS diagnosis according to the 2017 McDonald criteria [ 10 ] with either relapsing or progressive phenotype [ 11 ]; (2) no history of significant medical illnesses, fever or substance abuse in the 30 day before sample collection; (3) no other major systemic, psychiatric or neurological diseases; (4) no relapse or corticosteroid treatment in the 30 day before sample collection; (5) no concomitant medication possibly affecting cognitive or psychological function (i.e., antidepressant, antipsychotic, benzodiazepine). All enrolled subjects were screened for the availability, in the radiological database of our institution, of a brain MRI performed on the same day of the clinical evaluation, as part of on‐going non‐interventional non‐profit monocentric observational studies (QUASIMS, THURMS and NAPLISH) not designed to test safety or efficacy of specific disease modifying treatment (DMT).

Standard protocol approvals, registrations, and patient consents

Approval was received from the local ethical committee. All subjects gave written informed consent prior to study participation. The study was performed in accordance with good clinical practices and the Declaration of Helsinki.

Clinical and neuropsychometric assessment

For each enrolled participant we collected clinical and demographic data (i.e., age, sex, disease duration, disease course [relapsing–remitting, primary progressive and secondary progressive with relapsing–remitting MS patients classified as either early or late RRMS according to disease duration lower or higher than 5 years], education, on‐going DMT categorised into high‐efficacy monoclonal antibodies [anti‐CD52, Anti‐CD20, and natalizumab], oral therapies [cladribine, fumarate, S1P inhibitors, and teriflunomide], and injectables [peg‐interferon beta, interferon beta and glatiramer acetate]) as for clinical practice. Each subject underwent a clinical examination, including the assessment of physical disability through the Expanded Disability Status Scale (EDSS), cognitive and neurobehavioral assessment. Cognitive function was assessed through the BICAMS battery [ 12 ]. BICAMS battery includes the assessment of attention, information processing speed and working memory using the Symbol Digit Modalities Test (SDMT) [ 13 ], and the assessment of verbal and spatial memory using the California Verbal Learning Test‐II (CVLT II) and the Brief Visuo‐spatial Memory Test–Revised (BVMT‐R) [ 14 ], respectively. Scores were corrected for age, gender and education following Italian normative values, and corrected scores <35 were considered as impaired [ 15 ]. We calculated the total cerebral functional score (CFS) as the sum of impaired cognitive domains. Briefly, patients with CFS = 0 present with normal scores for the three cognitive domains explores while CFS scores from 1 to 3 correspond to deficit in 1–3 cognitive tests. Each patient also completed the Beck Depression Inventory (BDI‐II) [ 16 ] to assess depressive symptoms, the Modified Fatigue Impact Scale (MFIS) [ 17 ] providing scores for cognitive, physical and psychosocial fatigue, the Pittsburgh Sleep Quality Index (PSQI) [ 18 ] assessing sleep quality and disturbances, and Beck Anxiety Inventory (BAI) [ 19 ] measuring the severity of anxiety.

MRI acquisition and analysis

All MRI studies, that included a 3D T1‐weighted sequence for volumetric analyses and a 3D Fluid Attenuated Inversion Recovery (FLAIR) sequence for the quantification of white matter (WM) lesion volume (LV), were acquired on the same 3 Tesla scanner (Magnetom Trio, Siemens Healthineers). We enrolled 181 patients from three on‐going non‐interventional non‐profit monocentric observational studies 62 patients from protocol 1; 82 patients from protocol 2 and 37 patients from protocol 3. Details of the acquisition protocol are provided in Supplementary eMethods. WM lesions were automatically segmented on FLAIR images using Lesion Segmentation Tool (LST) 3.0.0 ( www.statistical‐modelling.de/lst.html ). Corresponding masks were used to (i) fill lesions in T1‐weighted images with a distribution of signal‐intensity values equivalent to those in normal‐appearing WM and (ii) compute WM LV. Filled T1‐weighted images were used in all subsequent volumetric analyses. Normalized brain, grey matter (GM) and WM volumes were quantified using SIENAX (FMRIB center, Oxford, UK). The normalized neocortical volume was quantified by applying a standard space mask from FSL to the individual GM probability map from SIENAX. Estimation of deep GM (DGM) structures volume (ie, the thalamus, caudate, pallidum, putamen, amygdala and hippocampus) was performed using FIRST (FMRIB center, Oxford, UK). All DGM volumes were normalized by multiplying the raw volumes by the SIENAX scaling factor and used in subsequent analyses as the sum between the two sides.

Statistical analysis

Statistical analyses were performed using the Stata/MP software (version15.03; StataCorp LP, College Station, TX). Demographic, clinical and radiological features of study subjects are presented as means, medians or proportions as appropriate. All demographic, clinical, laboratory and radiological variables were checked for normality using the Shapiro–Wilk normality test. In order to unveil latent neuropsychological profiles we employed a LPA. As it was already performed to identify latent cognitive profiles in MS [ 7 ], LPA is a model‐based method for clustering individuals in unobserved groups [ 20 ]. Differently from latent class analysis only allowing group to be modelled according to categorical variables, LPA is a more flexible mixture‐modelling approach allowing the inclusion of continuous variables estimating their joint variation as a function of a finite and mutually exclusive and exhaustive number of unobserved components, using a latent categorical profile without a priori estimation of group classification and a data‐driven approach [ 20 ]. In our study we would envisage latent classes as neuropsychological profiles in MS patients. LPA requires the selection of the model with a specified number of classes that better fit the joint distribution of included variables. To perform such an analysis, we run models from 1 to 9 classes including SDMT, CVLT II, BVMT‐R, BDI II, cognitive MFIS, physical MFIS, psychosocial MFIS as interest variables and age and gender as covariates. We recorded for each model the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC) and we drew an Elbow‐plot of BIC and AIC for the multivariate LPA. After selecting the best‐fitting model, we classified patients into one of the neuropsychological profiles on the basis of their class probabilities estimated directly from the model. Neuropsychological profiles were named according to the mean score for each test. For the cognitive battery, we will classify scores falling above the 2nd quartile as unaffected, scores falling between 1st and 2nd quartile as mildly affected and scores falling below the 1st quartile as severely affected. Similarly, we classified scores at the MFIS and BDI scales falling below the 2nd quartile as unaffected, scores falling between 2nd and 3rd quartile as mildly affected and scores falling above the 3rd quartile as severely affected. In order to disentangle also the possible role of physical disability, we performed the LPA analysis with models from 1 to 9 classes including CFS, EDSS, BDI II, cognitive MFIS, physical MFIS, psychosocial MFIS as interest variables and age and gender as covariates.

Between‐group comparisons of demographic and clinical features were performed using linear regression models and chi‐squared or nonparametric tests as appropriate. Between‐group comparisons of radiological features were performed using age‐, gender‐ and scanner‐adjusted linear regression models. Post‐regression proper contrasts were evaluated and Bonferroni post‐hoc correction for multiple comparisons was applied. For each comparison estimated mean difference (EMD) and 95% confidence intervals (95% CI) are reported.

A p value <0.05 was considered statistically significant. Results are presented with 95% confidence interval (95% CI) or p values.

RESULTS

Clinical measures at baseline

We enrolled 600 MS patients. We included 413 female (68.8%) and 187 male (31.2%) PwMS with a mean age of 42.3 ± 12.8 years, a median disease duration of 9.4 (0–43.1) years and a median EDSS of 2.5 (0–7.5). Most of the patients were late relapsing–remitting (54.7%), although we also enrolled early relapsing–remitting (27.8%), secondary progressive (12.8%) and primary progressive (4.7%) patients. Demographic and clinical data from subjects enrolled in the study are summarized in Table  1 . Cognitive and neuropsychological features are depicted in Table  2 .

TABLE 1 Demographic and clinical features of patients with multiple sclerosis (MS). Number of subjects 600 Sex Male, N (%) 187 (31.2) Female, N (%) 413 (68.8) Age, mean (SD) (years) 42.3 (12.8) Education, mean (SD) 12.9 (3.7) EDSS, median (Range) 2.5 (0–7.5) Age at onset, median (Range) (years) 29.7 (10–70.27) Disease duration, median (Range) (years) 9.4 (0–43.1) MS course Early relapsing–remitting, N (%) 167 (27.8) Late relapsing–remitting, N (%) 328 (54.7) Secondary progressive, N (%) 77 (12.8) Primary‐progressive, N (%) 28 (4.7) Paediatric onset, N (%) 50 (8.3) Ongoing DMT No DMT, N (%) 13 (2.2) Injectables (BRACE), N (%) 113 (18.8) Oral therapies, N (%) 244 (40.6) Teriflunomide, N (%) 25 (4.2) Fumarate, N (%) 61 (10.2) S1P inhibitors, N (%) 119 (19.7) Cladribine, N (%) 39 (6.5) High‐efficacy monoclonal antibodies, N (%) 230 (38.4) Natalizumab, N (%) 166 (27.7) Anti‐CD20, N (%) 46 (7.7) Anti‐CD52, N (%) 18 (3) Abbreviations: BRACE, betaferon, rebif, avonex (plegridy), copaxone, extavia; DMT, disease modifying treatment; EDSS, expanded disability status scale; MS, multiple sclerosis; N, Number; S1P, sphingosine 1‐phosphate; SD, standard deviation.

Demographic and clinical features of patients with multiple sclerosis (MS).

Abbreviations: BRACE, betaferon, rebif, avonex (plegridy), copaxone, extavia; DMT, disease modifying treatment; EDSS, expanded disability status scale; MS, multiple sclerosis; N, Number; S1P, sphingosine 1‐phosphate; SD, standard deviation.

TABLE 2 Corrected cognitive scores and neurobehavioral features in multiple sclerosis (MS) patients. Mean (SD) 1st Quartile 2nd Quartile 3rd Quartile Deficit, N (%) SDMT 44.0 (12.6) 35 44.2 52.3 149 (24.8) CVLT 42.5 (14.4) 31.2 40.6 51.6 208 (34.7) BVMT 42.4 (12.6) 33.1 42.5 51.3 177 (29.5) BDI‐II 7.6 (10.1) 0 3 13 – MFIS cognitive 8.6 (10.2) 0 4 16 – MFIS physical 10.6 (11.1) 0 6.5 19.75 – MFIS psychosocial 1.8 (2.4) 0 0 3 – MFIS total 21.1 (22.1) 0 14 38.8 – 0 1 2 3 CFS score, N (%) 299 (49.8) 147 (24.5) 75 (12.5) 79 (13.2) Abbreviations: BDI‐II, Beck depression inventory; BVMT, brief visuo‐spatial memory test; CFS, cerebral functional score.CVLT, california verbal learning test; MFIS, modified fatigue impact scale; MS, multiple sclerosis; N, number; SD, standard deviation; SDMT, symbol digit modalities test.

Corrected cognitive scores and neurobehavioral features in multiple sclerosis (MS) patients.

Abbreviations: BDI‐II, Beck depression inventory; BVMT, brief visuo‐spatial memory test; CFS, cerebral functional score.CVLT, california verbal learning test; MFIS, modified fatigue impact scale; MS, multiple sclerosis; N, number; SD, standard deviation; SDMT, symbol digit modalities test.

Phenotype classification

Using LPA and the elbow‐plot to highlight the kinking point identifying the optimal class number, we observed that both the 2‐class model and the 5‐class model fit the best our data (Figure  S1 ). Since the 2‐class model actually divided patients with normal neuropsychological phenotype from patients with both cognitive impairment and neurobehavioral features without providing any further insight into possible sub‐classification, we selected the 5‐class model (Clinico‐demographic details of the 2‐class model are presented in Table  S1 ). In order to also correct for physical disability, we also included EDSS in the LPA model and we obtained a similar result with a 3‐class model being selected (see Figure  S2 ) with the 3 classes representing the patients with normal neuropsychological phenotype, patients with mild cognitive impairment and neurobehavioral features and patients with severe cognitive impairment and neurobehavioral features (see Table  S2 ). Finally, we have also performed an LPA analysis using the short form of the BDI‐II in order to disentangle possible overlap between somatic and psychological depressive symptoms providing a 4‐classes model (see Data  S1 ).

The selected model depicted five neuropsychological phenotypes: (1) Normal neuropsychological phenotype including 145 patients (24.2%) with normal cognitive and neurobehavioral scores; (2) Isolated cognitive impairment including 183 patients (30.5%) showing reduced SDMT score, reduced CVLT score and reduced BVMT score vs normal neuropsychological phenotype; (3) Isolated neurobehavioral impairment including 76 patients (12.7%) showing increased BDI‐II score, increased cognitive MFIS score, increased physical MFIS score, increased psychosocial MFIS score and an increased total MFIS score vs normal neuropsychological phenotype; (4) Mild neuropsychological impairment including 119 patients (19.8%) showing reduced SDMT score, reduced CVLT score, reduced BVMT score, increased BDI‐II score, increased cognitive MFIS score, increased physical MFIS score, increased psychosocial MFIS score and an increased total MFIS score vs normal neuropsychological phenotype; (5) severe neuropsychological impairment including 72 patients (12%) showing reduced SDMT score, reduced CVLT score, reduced BVMT score, increased BDI‐II score, increased cognitive MFIS score, increased physical MFIS score, increased psychosocial MFIS score and an increased total MFIS score vs patients with normal neuropsychological phenotype. Cognitive and psychological scores for each neuropsychological phenotype are reported in Table  3 .

TABLE 3 Corrected cognitive scores and neurobehavioral features in the 5 neuropsychological phenotypes. Normal neuropsychological phenotype Isolated cognitive impairment
p value *
Mild neuropsychological impairment
p value *
Isolated neurobehavioral impairment
p value *
Severe neuropsychological impairment
p value *
SDMT, mean (SD) 53.9 (25.6) 40.2 (15.8) <0.001 39.1 (17) <0.001 52.3 (33.8) 1.0 33.8 (16.6) <0.001 Deficit, N (%) 5 (3.5) 60 (32) <0.001 44 (37) <0.001 1 (1.3) 1.0 45 (58.4) <0.001 CVLT, mean (SD) 58.5 (34.7) 35.1 (13.3) <0.001 36.5 (15.1) <0.001 49.8 (20.5) <0.001 32 (14.7) <0.001 Deficit, N (%) 2 (1.4) 92 (50.3) <0.001 56 (47.1) <0.001 6 (7.9) <0.001 52 (67.5) <0.001 BVMT, mean (SD) 53 (28.7) 38 (16.3) <0.001 35.8 (14.4) <0.001 52.3 (28.1) 1.0 33.5 (14.5) <0.001 Deficit, N (%) 4 (3) 71 (39) <0.001 56 (47) <0.001 2 (3) 1.0 44 (57) <0.001 CFS score 0, N (%) 135 (93.1) 49 (26.8) <0.001 38 (31.9) <0.001 68 (89.5) 1.0 9 (11.7) <0.001 1, N (%) 9 (6.2) 77 (42.1) 33 (27.7) 7 (9.2) 21 (27.3) 2, N (%) 1 (0.7) 30 (16.4) 22 (18.5) 1 (1.3) 21 (27.3) 3, N (%) 0 (0) 27 (14.8) 26 (21.9) 0 (0) 26 (22.8) Ongoing DMT No DMT, N (%) 2 (1.4) 2 (1.1) 1.0 6 (5) 0.44 2 (2.6) 1.0 1 (1.3) 1.0 Injectables, N (%) 21 (14.4) 35 (19.1) 30 (25.2) 14 (18.4) 13 (16.9) Oral therapies, N (%) 61 (42.1) 67 (36.6) 52 (43.7) 31 (40.8) 33 (42.9) High‐efficacy monoclonal antibodies, N (%) 61 (42.1) 79 (43.2) 31 (26.1) 29 (38.2) 30 (38.9) BDI‐II, mean (SD) 2.8 (0) 2.1 (0) 1.0 9.2 (0) <0.001 12.8 (0) <0.001 22.2 (0) <0.001 MFIS Cognitive, mean (SD) 2.1 (0) 1.2 (0) 1.0 11.8 (0) <0.001 17.8 (1) <0.001 24.7 (0) <0.001 MFIS Physical, mean (SD) 2.4 (0) 1.6 (0) 1.0 15.8 (4) <0.001 22.8 (4) <0.001 27.2 (13) <0.001 MFIS Psychosocial, mean (SD) 0.2 (0) 0.1 (0) 1.0 2.2 (0) <0.001 4.3 (0) <0.001 5.9 (0) <0.001 MFIS total, mean (SD) 4.7 (0) 2.9 (0) 0.42 30.2 (13) <0.001 44.8 (24) <0.001 57.9 (32) <0.001 Abbreviations: BDI‐II, Beck depression inventory; BVMT, brief Visuo‐spatial Memory Test; CFS, cerebral functional score.CVLT, california verbal learning test; MFIS, modified fatigue impact Scale; MS, Multiple sclerosis; N, number; SD, standard deviation; SDMT, symbol digit modalities test. * Vs patients with normal neuropsychological phenotype, Linear regression.

Corrected cognitive scores and neurobehavioral features in the 5 neuropsychological phenotypes.

Abbreviations: BDI‐II, Beck depression inventory; BVMT, brief Visuo‐spatial Memory Test; CFS, cerebral functional score.CVLT, california verbal learning test; MFIS, modified fatigue impact Scale; MS, Multiple sclerosis; N, number; SD, standard deviation; SDMT, symbol digit modalities test.

Vs patients with normal neuropsychological phenotype, Linear regression.

Clinical‐demographic features in neuropsychological phenotypes

Clinical‐demographic differences among the 5 neuropsychological phenotypes are depicted in Figure  1 and Table  S3 . Differently from disease duration, which was not different among neuropsychological classes, all other classes were older and had an older age at onset compared to patients with the normal neuropsychological phenotype. The exception was patients with isolated neurobehavioral impairment, who did not differ in age compared to patients with the normal neuropsychological phenotype. Compared with all other neuropsychological classes, patients with Severe neuropsychological impairment showed higher EDSS, higher PSQI and a higher BAI, except when compared with Mild neuropsychological impairment class. Compared with patients in both the Normal neuropsychological phenotype and Isolated cognitive impairment class, patients in the Mild neuropsychological impairment class and in the Isolated neurobehavioral class showed higher EDSS, higher BAI and a higher PSQI. Patients in the Isolated cognitive impairment class showed an increased EDSS compared with patients in the Normal neuropsychological phenotype. Finally, patients in the Normal neuropsychological phenotype and Isolated cognitive impairment are more likely in relapsing stages and with paediatric onset, whilst patients with progressive phenotypes were almost entirely allocated in the Mild neuropsychological impairment and Severe neuropsychological impairment without showing peculiar differences between secondary and primary progressive phenotype. Treatments distribution was not different among the five neuropsychological phenotypes.

Clinical‐demographic features of neuropsychological profiles. Violin plots with distribution of age (a), age at onset (b), disease duration (c), EDSS (d), BAI (g) and PSQI (h) for each neuropsychological profile. Histograms with relative frequencies of clinical disease courses (e) and paediatric onset (f) for each neuropsychological profile. * p values<0.05; ** p values<0.001.

MRI features in neuropsychological phenotypes

MRI data were available for 181 subjects (Normal neuropsychological phenotype: 30; Isolated cognitive Impairment: 27; Mild neuropsychological impairment: 54; Isolated neurobehavioral impairment: 31; Severe neuropsychological impairment: 39). Radiological differences among the 5 neuropsychological phenotypes are depicted in Figure  2 .

Radiological features of neuropsychological profiles. Violin plots with distribution of T2 lesion (a), total brain (b), white matter (c), total grey matter (d), deep grey matter (e), cortical grey matter (f), putamen (g), thalamus (h), pallidum (i), caudate nucleus (l), amygdala (m), hippocampus (n) volume for each neuropsychological profile. * p values<0.05; ** p values<0.001.

Patients with Severe neuropsychological impairment showed lower cortical GM volume (EMD [95% CI]: −27.3 [−53.1 to −1.5], p  = 0.03), lower thalamic volume (EMD [95% CI]: −1.6 [−2.7 to −0.4], p  = 0.002) and lower pallidum nucleus (EMD [95% CI]: −0.3 [−0.7 to −0.001], p  = 0.05) volume vs patients with Normal neuropsychological phenotype. They also showed lower brain volume (EMD [95% CI]: −59.7 [−112.1 to −7.2], p  = 0.02), GM volume (EMD [95% CI]: −30.4 [−58.3 to −2.4], p  = 0.02), cortical GM volume (EMD [95% CI]: −25.2 [−50.2 to −0.1], p  = 0.04), thalamic volume (EMD [95% CI]: −1.8 [−2.9 to −0.7], p  < 0.001), pallidum (EMD [95% CI]: −0.4 [−0.8 to −0.1], p  = 0.001), putamen (EMD [95% CI]: −0.8 [−1.58 to −0.02], p  = 0.04) and amygdala (EMD [95% CI]: −0.34 [−0.68 to −0.002], p  = 0.05) volume vs Isolated neurobehavioral impairment and lower thalamic (EMD [95% CI]: −1.5 [−2.7 to −0.2], p  = 0.008) and pallidum nucleus (EMD [95% CI]: −0.38 [−0.7 to −0.04], p  = 0.02) volume vs Isolated cognitive impairment. Finally, patients in the Mild neuropsychological impairment class showed lower deep GM (EMD [95% CI]: −7.9 [−15.6 to −0.3], p  = 0.04) and thalamic (EMD [95% CI]: −1.1 [−2.1to −0.02], p  = 0.04) volume compared with patients in the Isolated neurobehavioral impairment class.

DISCUSSION

We identified five different neuropsychological profiles of people with MS using a data‐driven approach based on cognitive and neurobehavioral scores. These groups included: patients with Normal neuropsychological profile; patients with Isolated cognitive impairment, showing mild isolated deficit in processing speed, working, verbal and/or spatial memory; patients with Isolated neurobehavioral impairment, showing only increased level of depression and/or fatigue; patients with Mild or Severe neuropsychological impairment, showing co‐occurrence of mild or severe deficit at cognitive and neurobehavioral evaluation. We also validated our phenotypes towards conventional clinical and radiological features, with significant differences as from different pathobiology. Taken together, our phenotypes depict different phases of MS, from normality, to isolated neurobehavioral or cognitive impairment, to mild or severe neuropsychological impairment.

First, we showed that cognitive and neurobehavioral impairment frequently coexists and clusters with different grades of severity. Second, relapsing–remitting and paediatric‐onset patients are more likely in the Normal neuropsychological phenotype and Isolated cognitive impairment class, whilst patients in the Isolated neurobehavioral impairment and/or in neuropsychological impaired classes showed more frequently higher disability and progressive course. Third, patients with neurobehavioral issues or Mild neuropsychological involvement showed poor sleep quality. Finally, patients with any grade of neuropsychological impairment showed lower thalamic volume compared with those with Isolated neurobehavioral impairment.

Using LPA, De Meo et al. described five distinct cognitive phenotypes in people with MS exploring the extent of neurobehavioral burden as a post‐hoc analysis after cognitive profile identification. This was done since they applied a very extensive cognitive battery (the Rao battery) [ 21 ] exploring different cognitive domains. However, in recent years, the BICAMS battery has become the most frequently used battery for cognitive assessment in MS, due to well established validation studies, along with rapidity and ease of administration compared with larger batteries [ 22 ]. BICAMS explores three different cognitive domains: processing speed, visual and verbal memory which are more usually affected in MS. Given the difference in the tools applied to assess cognition in the previous study by De Meo et al., we were not able to cluster patients according to selective cognitive domains impaired. Conversely, we demonstrated clusters of patients showing a continuum of multi‐domain cognitive deficits in terms of severity. Recent evidence suggests that SDMT does not only capture processing speed but also episodic memory, language and sensory processes [ 23 ]. Therefore, SDMT results may cluster with results from CVLT and BVMT not allowing to properly differentiating affected cognitive domains. Accordingly, patients with processing speed/attention deficits in the study from De Meo et al. were classified according to two tests not embedded in the BICAMS: Stroop Color and Word Test (SCWT) and the Paced Auditory Serial Addition Test (PASAT) [ 23 ].

Although SDMT demonstrated comparable validity with slightly better predictive value compared with PASAT [ 24 ] and Stroop Test [ 25 ], we assumed that our scales were not enough informative to define cognitive profile with LPA with specific regards to the processing speed/attention domain.

Another interesting finding from our study was the demonstration that cognitive and neurobehavioral impairment frequently coexists and clusters in different grades of severity. In previous studies it has been already demonstrated that depression impacts on attention, working memory, executive functions, information processing speed, and visuo‐spatial deficits in people with MS. [ 26 ] On the other hand, although various studies failed to find relationship between subjective fatigue and objective cognitive function [ 27 ], the combination of fatigue and depression was demonstrated to have great impact on verbal memory [ 26 ]. Globally, fatigue accounts for decrements in performance over time in tasks that require sustained mental effort [ 28 ]. A small group of patients showed isolated neurobehavioral deficit, consisting in the presence of both depressive symptoms and physical‐cognitive fatigue without significant deficit in cognitive performances. Since depression and fatigue frequently coexist and probably share common mechanisms in patients with MS [ 29 ], our finding was not completely unexpected. This subgroup of patients pointed toward the existence of disease mechanisms related specifically to neurobehavioral aspects of MS, but not with cognition.

Differently from patients with neurobehavioral issues, patients with any grade of cognitive impairment were older and had a higher age at onset compared with the unaffected group, pointing out that aging poses people with MS to a higher risk of cognitive decline but has no impact on neurobehavioral features. In line with this finding, previous studies demonstrated that, after adjusting for physical disability, age was not associated with depression or fatigue [ 30 ]. Moreover, older age at onset has been previously associated with decreased cognitive reserve with reduced capability of functional reorganization in MS [ 31 ], ultimately resulting in the inability of the brain to counteract for pathological damages resulting in cognitive disability. To reinforce this concept, paediatric‐onset MS patients were more frequently unaffected or had mild cognitive impairment suggesting that brain reserve is able to overcome brain damages through plasticity mechanisms [ 32 ]. Notwithstanding the ability of the brain to counteract for accumulating pathology, young MS patients present with a greater cognitive impairment compared with controls. This difference reduces with aging, suggesting that MS patients experience an early cognitive aging [ 33 ]. This finding might represent an accelerated rate of brain atrophy in patients with MS, especially in those regions more closely associated with cognitive functions [ 34 , 35 ].

In subgroups with higher grade of cognitive impairment, patients showed more frequently higher disability and progressive course. Cognitive impairment had a 2‐fold higher prevalence in patients with progressive forms. Association between physical and cognitive disability and progression has been related to GM atrophy and brain volume loss that characterize advanced phases of MS. [ 36 ].

When evaluating anxiety and sleep symptoms in neuropsychological profiles we demonstrated that patients with normal neuropsychological profile and patients with isolated cognitive impairment did not show any sleep‐related issue or anxiety, whilst patients with any grade of neurobehavioral issues showed poor sleep quality. Although the exact mechanisms of the interplay is still to be determined, poor sleep quality correlates with fatigue, depression and cognitive impairment in patients with MS. [ 37 ] It would be intriguing to assess whether poor sleep quality precedes the other symptoms or not, as to highlight a possible causative mechanism.

Looking at MRI volumetric analysis, patients with severe neuropsychological impairment showed lower volume of cortical GM, thalamus and nucleus pallidum when compared with patients with normal neuropsychological phenotype. They also showed lower volume of brain, GM, cortical GM, thalamus, nucleus pallidum, putamen, and amygdala when compared with isolated psychological impairment, and lower volume of thalamus and pallidum nucleus when compared with isolated cognitive impairment. In our analysis, atrophy of thalamus and pallidum was the hallmark of subgroups with major grade of neuropsychological impairment, confirming the central role of these structures in cognition in MS.

The thalamus plays a central role in several cognitive processes like memory and behavioral flexibility [ 38 ]. From previous evidence, maladaptive connectivity of thalamus and thalamic atrophy closely correlated with cognitive impairment in MS, and early deep GM atrophy is not only a relevant correlate, but also a valuable predictor of long‐term cognitive impairment in MS. [ 39 ]

A recent study reported more extensive and severe damage of cortical and deep brain regions apart from normal‐appearing WM and amygdala in severe multi‐domain cognitive impairment compared with single‐domain and mild impairment [ 40 ]. On the other hand, the specific role of pallidum has not been investigated, demonstrating correlation with cognitive processing speed test only for patients with SPMS [ 41 ].

On the contrary, the subgroup with isolated neurobehavioral impairment showed no significant differences in atrophy compared with patients with normal neuropsychological phenotype. Previous studies demonstrated that cognitive impairment is correlated with cortical and subcortical GM atrophy as well as widespread functional reorganization of brain networks, while depression could be related to circumscribed atrophy in cortical areas deputed to emotional behavior [ 42 ]. Similarly, also fatigue has no clear association with GM atrophy whilst it appears more closely related to widespread normal‐appearing WM damage [ 43 ] and functional disconnection between cortical and subcortical structures [ 43 ]; as well as in the deregulation of the monoamine circuits [ 5 , 44 ].

We acknowledge that this study is not without limitations. First, we conducted LPA based on cognitive and neurobehavioral profile assessed through commonly used clinical scores. For cognitive assessment we applied BICAMS score that is widely used due to its ease of administration, but other more complete cognitive batteries, exploring more cognitive domains could be used. Similarly, neurobehavioral profile of patients was assumed only by assessing depression and fatigue. Although depression and fatigue are the neurobehavioral symptoms more negatively impacting on quality of life of people with MS [ 45 ], other issues (e.g. phobias, obsessions, resilience etc.) could be explored to better define behavioral features of patients with MS as well as it might be helpful to also include anxiety and sleep quality in the LPA model. We decided to not include them in the analysis to keep the model as simple as possible. Moreover, BDI‐II questionnaire includes some items referring to somatic symptoms. To this purpose a short form of the BDI was developed [ 46 ]. However, we decided to keep with the BDI‐II as Italian multicentric studies mostly used this questionnaire [ 47 ]. Further studies disentangling the somatic and neurobehavioral impact on depressive symptoms should be performed. In addition, we did not exclude from the LPA analysis patients treated with anticholinergic drugs for sphincter disorders. While it has been demonstrated in the healthy population that treatment with anticholinergic drugs is associated with reduced cognitive performances and a more pronounced brain atrophy [ 48 ], studies in MS exploring association between treatment with anticholinergic drugs for sphincter disorders and cognitive functions are lacking and the association deserves to be further investigated. Finally, the cross‐sectional nature of our study did not allow us to make conclusion on predictive role of these profiles nor on the dynamic of the symptoms' onset. It should be interesting to explore factors that determine the transition between different subgroups and the impact of pharmacological and lifestyle intervention on it.

In conclusion, we showed that LPA based on cognitive and neurobehavioral scores define five profiles that have specific clinical and radiological features. These features provide information on the possible mechanisms underpinning MS‐related symptoms and could help clinicians in addressing patients to specific cognitive or neurobehavioral assessment given the limited time‐frame of the clinical assessment in clinical settings Appendix  A1 .

AUTHOR CONTRIBUTIONS

Antonio Luca Spiezia: Data curation; investigation; conceptualization; writing – review and editing; writing – original draft. Giuseppe Pontillo: Writing – review and editing; data curation; investigation. Fabrizia Falco: Writing – review and editing; investigation; data curation. Martina Eliano: Writing – review and editing; data curation; investigation. Federica Lamagna: Data curation; investigation; writing – review and editing. Antonio Esposito: Writing – review and editing; investigation; data curation. * Cristina Di Monaco1: Writing – review and editing; data curation; investigation. Valerio Nicolella: Investigation; data curation; writing – review and editing. Federica Novarella: Investigation; data curation; writing – review and editing. Marcello Moccia: Data curation; investigation; writing – review and editing; writing – original draft; visualization; methodology. Sirio Cocozza: Investigation; data curation; writing – review and editing; writing – original draft; methodology. Arturo Brunetti: Investigation; data curation; writing – review and editing. Roberta Lanzillo: Investigation; data curation; writing – review and editing. Maria Petracca: Methodology; conceptualization; writing – review and editing; writing – original draft; supervision; investigation; data curation. Vincenzo Brescia Morra: Investigation; data curation; supervision; writing – review and editing; writing – original draft; methodology. Antonio Carotenuto: Methodology; supervision; investigation; writing – review and editing; writing – original draft; visualization; conceptualization; data curation; formal analysis; validation.

Supporting information

Data S1.