Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Language production impairments in patients with a first episode of psychosis

  • Giulia Gargano,

    Roles Data curation, Writing – original draft, Writing – review & editing

    Affiliation Department of Pathophysiology and Transplantation, University of Milan, Milan, Italy

  • Elisabetta Caletti,

    Roles Data curation, Writing – original draft, Writing – review & editing

    Affiliation Department of Neurosciences and Mental Health, Fondazione IRCCS Ca’ Granda-Ospedale Maggiore Policlinico, Milan, Italy

  • Cinzia Perlini,

    Roles Data curation, Supervision, Writing – review & editing

    Affiliations Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, Verona Hospital Trust–Azienda Ospedaliera Universitaria Integrata Verona–AOUI, Verona, Italy

  • Nunzio Turtulici,

    Roles Data curation, Formal analysis, Writing – original draft

    Affiliation Department of Pathophysiology and Transplantation, University of Milan, Milan, Italy

  • Marcella Bellani,

    Roles Data curation, Supervision, Writing – review & editing

    Affiliations Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, Verona Hospital Trust–Azienda Ospedaliera Universitaria Integrata Verona–AOUI, Verona, Italy

  • Carolina Bonivento,

    Roles Data curation, Writing – review & editing

    Affiliation IRCCS “E.Medea” Polo Friuli Venezia Giulia, San Vito al Tagliamento, PN, Italy

  • Marco Garzitto,

    Roles Data curation, Resources

    Affiliation Department of Languages and Literatures, Communication, Education and Society, University of Udine, Udine, Italy

  • Francesca Marzia Siri,

    Roles Data curation, Resources

    Affiliation Department of Neurosciences and Mental Health, Fondazione IRCCS Ca’ Granda-Ospedale Maggiore Policlinico, Milan, Italy

  • Chiara Longo,

    Roles Data curation, Resources

    Affiliation Department of Neurosciences and Mental Health, Fondazione IRCCS Ca’ Granda-Ospedale Maggiore Policlinico, Milan, Italy

  • Chiara Bonetto,

    Roles Data curation, Writing – original draft

    Affiliation Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy

  • Doriana Cristofalo,

    Roles Data curation, Resources

    Affiliation Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy

  • Paolo Scocco,

    Roles Data curation, Investigation

    Affiliation Department of Mental Health, Azienda ULSS 16, Padua, Italy

  • Enrico Semrov,

    Roles Data curation, Investigation

    Affiliation Department of Mental Health, Reggio Emilia, Italy

  • Antonio Preti,

    Roles Data curation, Investigation

    Affiliation Department of Mental Health, Niguarda Ca’ Granda Hospital, Milan, Italy

  • Lorenza Lazzarotto,

    Roles Data curation, Investigation

    Affiliation Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy

  • Francesco Gardellin,

    Roles Data curation, Investigation

    Affiliation Department of Mental Health, Azienda Ulss 8 Berica, Vicenza, Italy

  • Antonio Lasalvia,

    Roles Funding acquisition, Project administration, Supervision, Writing – review & editing

    Affiliations Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, Verona Hospital Trust–Azienda Ospedaliera Universitaria Integrata Verona–AOUI, Verona, Italy

  • Mirella Ruggeri,

    Roles Funding acquisition, Project administration, Supervision, Writing – review & editing

    Affiliations Department of Neuroscience, Biomedicine and Movement Sciences, University of Verona, Verona, Italy, Verona Hospital Trust–Azienda Ospedaliera Universitaria Integrata Verona–AOUI, Verona, Italy

  • Andrea Marini,

    Roles Conceptualization, Data curation, Methodology, Writing – review & editing

    Affiliation Department of Languages and Literatures, Communication, Education and Society, University of Udine, Udine, Italy

  • Paolo Brambilla ,

    Roles Conceptualization, Data curation, Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing

    paolo.brambilla1@unimi.it

    Affiliations Department of Pathophysiology and Transplantation, University of Milan, Milan, Italy, Department of Neurosciences and Mental Health, Fondazione IRCCS Ca’ Granda-Ospedale Maggiore Policlinico, Milan, Italy

  •  [ ... ],
  • GET UP Group

    Membership of the GET UP Group is provided in the Acknowledgments.

  • [ view all ]
  • [ view less ]

Abstract

Language production has often been described as impaired in psychiatric diseases such as in psychosis. Nevertheless, little is known about the characteristics of linguistic difficulties and their relation with other cognitive domains in patients with a first episode of psychosis (FEP), either affective or non-affective. To deepen our comprehension of linguistic profile in FEP, 133 patients with FEP (95 non-affective, FEP-NA; 38 affective, FEP-A) and 133 healthy controls (HC) were assessed with a narrative discourse task. Speech samples were systematically analyzed with a well-established multilevel procedure investigating both micro- (lexicon, morphology, syntax) and macro-linguistic (discourse coherence, pragmatics) levels of linguistic processing. Executive functioning and IQ were also evaluated. Both linguistic and neuropsychological measures were secondarily implemented with a machine learning approach in order to explore their predictive accuracy in classifying participants as FEP or HC. Compared to HC, FEP patients showed language production difficulty at both micro- and macro-linguistic levels. As for the former, FEP produced shorter and simpler sentences and fewer words per minute, along with a reduced number of lexical fillers, compared to HC. At the macro-linguistic level, FEP performance was impaired in local coherence, which was paired with a higher percentage of utterances with semantic errors. Linguistic measures were not correlated with any neuropsychological variables. No significant differences emerged between FEP-NA and FEP-A (p≥0.02, after Bonferroni correction). Machine learning analysis showed an accuracy of group prediction of 76.36% using language features only, with semantic variables being the most impactful. Such a percentage was enhanced when paired with clinical and neuropsychological variables. Results confirm the presence of language production deficits already at the first episode of the illness, being such impairment not related to other cognitive domains. The high accuracy obtained by the linguistic set of features in classifying groups support the use of machine learning methods in neuroscience investigations.

Introduction

Language is one of the most complex human skills, resulting from the dynamic integration of processes that require the combined functioning of several brain areas. Given the fundamental importance that language has in characterizing human beings and as a primary source for clinicians to assess both strengths and weaknesses in patients, its study in mental illness, and especially in psychosis, is particularly interesting. Patients with psychosis usually have difficulties in selecting contextually appropriate words, and often fill their speech with irrelevant pieces of information and derailments [1]. Impairments are generally observed in a range of psychotic disorders, including schizophrenia and bipolar disorder [2, 3]. In the prodromal phase of the illness, some language production deficits are even considered to be specific predictors of schizophrenia subtypes, and their presence influences the prognosis [46]. A fascinating theory by Crow has even hypothesized a central role of language in the development of schizophrenia [7]. According to such hypothesis, psychotic symptoms may be related to an altered hemispheric lateralization [8], with language lateralization during language production tasks being significantly reduced in psychotic patients even at the onset of the symptoms [9, 10]. Language processing rests on the integration between different systems (e.g., those processing phonological, morphological, syntactic, semantic or pragmatic information, see [11, 12], for reviews), so their systematic investigation can represent a useful tool to clarify the nature of the linguistic deficits observed in psychosis and to guide the understanding of the pathogenic trajectories of the disease from first episodes to the chronic condition. While the current literature on language impairments is quite large for psychotic patients, also including several papers by our group [1, 1217], with chronic patients showing alterations in both language comprehension and production [18], much less emphasis has been given to the first episode so far. In previous studies, linguistic impairments have been observed in patients with a first episode of psychosis (FEP) at the receptive level [1517, 19, 20]. By contrast, the investigation of the linguistic productive level has been less investigated in the first phases of the illness. For example, some studies on patients with a clinical high risk (CHR) to develop psychosis show that language production variables, particularly at the syntactic and semantic level of discourse, can be predictive of psychosis onset [21, 22]. Despite that, very few studies evaluate speech production in FEP in a detailed and systematic way. Existing studies usually focus on relating the patients’ outcomes and prognosis to various clinically objectifiable deficits, including some in the language department [23]. Also, most of the available studies on linguistic production skills in FEP use clinical scales to assess language, thus producing confusing results as they do not always take into account the difference between thought and language variables, therefore missing out on important areas of language production such as discourse, cohesion, naming abstraction and semiotics [24, 25]. Ayer and colleagues [26], for example, used the Thought and Language Index (TLI) [24] to evaluate formal thought disorders (FTD) in FEP patients, showing poverty of speech, perseveration and peculiar word use as significant factors in differentiating FEP from HC. Other studies combine clinical findings and a natural language processing (NLP) approach, consisting in a language analysis made by an artificial intelligence which is programmed to understand text and spoken words in a similar way to human beings [27, 28]. Silva et al, 2021 [29], for example, found that people with a first episode of schizophrenia (FES) have a linguistic style characterized by reduced analytic thinking, with less categorical linguistic style (that is with less formal and hierarchical patterns) than healthy controls (HC), but using the same proportion of function (including articles, prepositions, conjunctions, non-referential adverbs, negations, and auxiliary verbs) and content words (words that hold meaning on their own). Disorganization as a clinical item also seems to be related to an aberrant use of connectives (specifically increased temporal but decreased use of causal connectives) [30].

Despite recent findings that have proven new insight on linguistic production deficits in FEP, an overall comprehension of language production impairments in this population is still missing. For this reason, the first aim of the present study is to investigate linguistic production abilities in a large sample of FEP, according to a well-established and systematic methodology that has been specifically designed to evaluate language production in all its sub-domains and variables [31]. In particular, we focused on both micro- (lexicon, morphology, syntax) and macro-linguistic (discourse coherence, pragmatics) dimensions of language [31].

Another interesting issue in clinical research is represented by the investigation of similarities and differences in patients with affective and non-affective psychosis. In previous studies by our group, we detected linguistic deficits in both affective and non-affective chronic patients at both receptive [13, 1517] and productive levels [14], with more severe and generalized impairment in patients with schizophrenia than in those with bipolar disorder. In particular, in narrative production, participants with schizophrenia had slight problems in speech rate (that is number of word/unit time) and deficits at both local and global discourse coherence, whereas patients with bipolar disorder showed reduced mean length of utterance, compared with healthy participants [14]. Other studies show that patients with bipolar disorder usually display poverty of speech and content, as well as circumstantiality and self-reference. These findings, however, usually concern patients in the depressive state of the illness, aside from circumstantiality, which can be portrayed by patients in the manic phase as well [32]. Manic patients also usually are known to make clang associations (those associations based on similarities of sounds). An interesting study by Docherty et al [33] comparing schizophrenia and bipolar disorder patients shows that schizophrenic patients produce utterances with a relevantly lower syntactic complexity than patients with mania, even at an early stage of illness. Despite some type of language production impairment has been observed in both affective and non-affective chronic patients, specific deficits have not been clearly identified in a conclusive manner in FEP. A study by Kravariti et al (2009) showed FEP patients with positive history of mania to have an isolated, selective deficit in semantic verbal fluency [34], while another [23] identified verbosity as the only FTD category which could discriminate between FEP patients with a mania or schizophrenia diagnosis. Again, however, these results are derived from analysis on clinical scales that do not take into account the full complexity of natural language. Given such a premise, the second aim of the present study is to compare affective and non-affective FEP patients in an effort to compare their language production profiles with the aim of improving our comprehension of these pathologies [35]. A further intriguing issue concerns the relationship between language and other cognitive skills. In chronic patients with psychosis different cognitive abilities are altered (e.g., language, memory, attention) [3638]. As these complex functions strongly affect one another, it is not clear yet whether language deficits are to be considered at least partly independent of other cognitive dysfunctions [e.g., 3941]. The potential relation between different cognitive difficulties may be easier to evaluate in persons with FEP, where dysfunctions are generally quite moderate and the effects of medication and chronicity are not prevalent.

In sum, the present study aimed to compare language production abilities in a large cohort of patients with FEP compared to a group of HC, by considering both affective (FEP-A) and non-affective (FEP-NA) and taking into account the potential interrelations between different cognitive impairments. In addition to classical statistical analyses, we also adopted a machine learning (ML) approach in order to test the predictive values of the linguistic variables in discriminating between FEP vs HC and FEP-NA vs FEP-A. Specifically, we hypothesized that a) patients with FEP would show deficits in linguistic performance compared to HCs; b) patients with FEP-NA would display different profiles compared to FEP-A; c) some language difficulties would be to some extent independent from other cognitive impairments while others (e.g., language planning and discourse organization) would be more related to other cognitive skills (e.g., executive functions), and d) linguistic production variables can discriminate between FEP and HC and, possibly, between FEP-NA and FEP-A.

Materials and methods

Participants

Three-hundred thirty-nine Italian speaking adults with no history of drug or alcohol abuse and with normal or corrected to normal vision and hearing took part in this study. They formed an experimental and a control group. The experimental group was formed by 206 outpatients with FEP that were recruited from 117 public community mental health centers (CMHC) located in the north of Italy in the frame of the study ‘Genetics Endophenotypes and Treatment: Understanding early Psychosis’ (the GET UP study; [42]). The GET UP inclusion criteria were: age 18–54 years, residence in the catchment regions of the CMHCs and first lifetime contact, presence of at least one of the following symptoms: hallucinations, delusions, qualitative speech disorder, qualitative psychomotor disorder, bizarre, or grossly inappropriate behavior, or two of the following: loss of interest, initiative, and drive; social withdrawal; episodic severe excitement; purposeless destructiveness; overwhelming fear; or marked self-neglect. Exclusion criteria were: antipsychotic treatment (>3 months) prescribed for an identical or similar mental disorder; mental disorders caused by a general medical condition; moderate or severe mental disability evaluated by a clinical functional assessment; and psychiatric diagnosis other than International Classification of Diseases (ICD)-10 [43] for psychosis. The specific ICD-10 codes for psychosis were assigned at 9 months. Diagnoses were made by using the Item Group Checklist (IGC) of the Schedule for Clinical Assessment in Neuropsychiatry (SCAN) [44] and were confirmed by the clinical consensus of two staff psychiatrists. Participants’ social, occupational, and psychological functioning were assessed with the Global Assessment of Functioning (GAF) scale [45], while FEP positive and negative symptoms were assessed by means of the Positive and Negative Syndrome Scale (PANSS) [46]. Patients were also assessed for the presence or absence of affective symptoms by means of the Hamilton Depression Rating Scale (HDRS) [47] and the Bech-Rafaelsen Mania Rating Scale (BRMRS) [48]. In addition, FEP mean duration (in days) of untreated psychosis (DUP), defined as the time from onset of first psychotic symptom (as reported by patients) to first contact with public mental health services, was recorded.

A group of 133 healthy participants was also recruited by word of mouth in the frame of other studies at the AOUI of Verona, Italy. Participants in the control group had no DSM-IV Axis I disorders, determined using a brief modified version of the Structured Clinical Interview for DSM-IV–Non-Patient Version, no history of psychiatric disorder among first-degree relatives and alcohol or substance misuse, no current major medical illness. Since a core part of the present study involves machine learning (ML) analyses, we randomly sampled the entire set of data of FEP and HC in order to obtain a final dataset of 133 FEP and 133 HC (Table 1). All the data, analyses and results described from here onward will be referred to this final dataset. According to the ICD-10 criteria, 38 out of 133 patients were classified as having affective psychosis, FEP-A (F30.2, F31.2, F31.5, F31.6, F32.3, F33.3), while 95 received a diagnosis of non-affective psychosis, FEP-NA (F20-F29) (see S1 Table for a description of FEP-NA and FEP-A).

thumbnail
Table 1. Sociodemographic and clinical data of the sample of FEP and HC.

https://doi.org/10.1371/journal.pone.0272873.t001

The study was approved by the Ethics Committee of the AOUI of Verona (Prot. N. 20406/CE; N. 1877; N. 1338; N. 2290) and, only for patients, from the local Ethic Committee of each site of recruitment. All participants gave signed informed consent, after an explanation of all issues involved in participation in the research. The entire procedure and the informed consent form were drawn up according to the declaration of Helsinki [49] and approved by the Ethics Committee, which did not require further procedures to assess the capacity to consent.

Linguistic production measures

The participants’ narrative production skills were assessed by adopting a multilevel and systematic procedure already published by our group [13, 14, 31, 50]. Patients with FEP and HC were asked to provide a verbal description of a drawing representing the story of a boy trying to reach a bird’s nest (The Nest Story; [51]). No time limit was given, and the evaluator was instructed to refrain from intervening. The narrative productions were recorded and later transcribed verbatim with the inclusion of phonological fillers, pauses, and false starts by two independent coders who had been previously trained. The transcripts were compared for ensuring intercoder reliability. The transcripts underwent an accurate manual linguistic analysis by the same coders, focusing on both micro- and macro-linguistic levels of processing, which refer respectively to the intra-phrasal (lexicon, morphology, syntax) and inter-phrasal (discourse coherence and pragmatics) levels of language processing [31]. An in-depth description of linguistic variables and their meaning is detailed in the S1 File.

Neuropsychological assessment

All participants were administered tasks assessing global cognitive and executive functioning. In particular, verbal intelligence quotient (V-IQ) was assessed by administering the Test di Intelligenza Breve (TIB; [52]). Working memory and the ability to process multiple units of information by short term memory were assessed with the n-Back task (adapted from Kirchner, [53]) and the Span of Apprehension task, SOA (adapted from Asarnow and MacCrimmon, [54]), respectively. For a detailed description of the cognitive tasks, please refer to the S2 File.

Statistical analyses

Analyses were performed using the SciPy library, version 1.8.0, for Python 3.10 (https://scipy.org). Continuous socio-demographic, clinical, neuropsychological and linguistic variables were entered into different series of t-tests for independent samples, comparing FEP and HCs. If the main effect was statistically significant (i.e. the comparison between FEP and HC gave a p≤0.05), post-hoc t-tests further explored the differences between FEP-NA and FEP-A, FEP-NA vs HC and FEP-A vs HC. For the post-hoc comparisons, the Bonferroni correction was applied adjusting the alpha level according to the number of comparisons to a value alpha = 0.02. The differences between groups in the distribution of males and females were analyzed by means of the Chi2 Fisher’s exact test. Pearson’s correlations were performed separately for the group of FEP and the group of HC in order to investigate the relationship between the linguistic and neuropsychological variables which resulted statistically significant in the t-tests (alpha level corrected for multiple comparisons).

Machine learning (ML) analyses

A machine learning (ML) algorithm including independent variables (socio-demographic, neuropsychological and linguistic measures) was used to classify the two diagnostic groups (FEP and HC). The eXtreme Gradient Boosting algorithm was used. This is an improved version of decision trees first proposed by Chen & Carlos Guestrin [55] which uses a gradient optimization technique to better fit several decision trees to the training data, which then contribute to give a global more reliable response. To prevent overfitting of the models and more accurate results, we performed a train-test split of the dataset, the former used for training the model, the latter to measure accuracy of the trained model. We applied the common train/test splitting ratio of 80%/20%. We used the Julia language, version 1.6, and its ScikitLearn.jl package, a part of the ScikitLearn package available for Python mainly for data preprocessing and plotting; we also implied the XGBoost port of the XGBoost C++ algorithm implementation, version 1.4.0 [https://xgboost.readthedocs.io], for Python, version 3.10, called from the Julia environment. We tuned the main XGBoost parameter, max_depth, a hyperparameter setting the maximum depth of the trees in the model, with a grid search based on cross validation scores to select the best value. Maximum depth of the trees clearly improves model complexity so that its tuning is necessary to balance and set the model between underfitting and overfitting. Due to the expected potential variations of the predictive models’ accuracy, we repeatedly simulated fitting experiments to address these variations. For this purpose, the model has been fitted several times, randomly splitting at each run the dataset in train and test cases for 100 runs. We first fitted and tested our model choosing as predictors language production features only. Then, in order to improve the model’s discriminative power, we fitted and tested further models by adding language production variables to models trained on sociodemographic and neuropsychological data only. As a reference threshold, we list the "dummy" mode classifier accuracy, which always predicts the most represented class. Finally, in the model containing only the linguistic measures, we extracted feature importance in order to establish which linguistic production variables are the most useful in differentiating FEP and HCs and their relative contribution to the model’s accuracy.

Results

Multilevel assessment of narrative language

Patients with FEP as a group obtained worse performance than HC in measures assessing both the microlinguistic and macrolinguistic domains.

At micro-linguistic level, FEP and HC reported significantly worse Speech Rate, Mean Length of Utterances, Phonological Paraphasias, Lexical Fillers and Syntactic Completeness (all p<0.05) (Table 2).

Post-hoc comparisons showed that FEP-NA did significantly worse than HC for Speech Rate, Mean Length of Utterances, and Syntactic Completeness (all p<0.02), while the difference between FEP-NA and HC for their production of Lexical Fillers did not survive the Bonferroni correction (p>0.02). At a macro-linguistic level, FEP did more Local Coherence Errors and produced more Utterances with Semantic Errors than HC (all p<0.05). Post-hoc comparisons showed that only the difference between FEP-NA and HC for the production of Utterances with Semantic Errors and the difference between FEP-A and HC for the Local Coherence Errors (missing) passed the Bonferroni correction threshold (p<0.02), suggesting that both FEP-NA and FEP-A contributed equally in the variability of the measures of Local Coherence Errors (ambiguous) and the production of Utterances with Semantic Errors (p>0.02). See Tables 35 for details.

Results of neuropsychological assessment

All subgroups of patients with FEP had lower Verbal IQ (V-IQ) than HC (all p<0.05). Post-hoc comparisons showed that both FEP-NA and FEP-A had a lower V-IQ than HC (all p<0.02) and that FEP-NA and FEP-A did not differ between each other (p>0.02). Patients with FEP generally reported worse n-Back sensitivity scores than HC in all the difficulty conditions (0-Back, 1-Back, 2-Back, 3-Back, all p<0.05). The Post-hoc comparisons revealed a significant difference between FEP-NA and HC and FEP-A and HC in the 1-Back, 2-Back and 3-Back conditions (all p<0.02). Instead, the Post-hoc difference between FEP-NA and HC and FEP-A and HC did not pass the Bonferroni correction in the 0-Back condition (p>0.02). FEP-NA and FEP-A did not differ significantly between each other in any condition (all p>0.02). As for n-Back specificity, patients with FEP generally reported worse specificity scores than HC in the conditions 0-Back and 1-Back (p<0.05), while they did not differ in the 2-Back and 3-Back conditions (p>0.05). The Post-hoc comparisons revealed a significant difference between FEP-NA and HC in the 1-Back condition (p<0.02). All the other Post-hoc comparisons (i.e. FEP-A v HC; FEP-NA v FEP-A) were not significant (p>0.02). Concerning SOA sensitivity, FEPs’ patients did significantly worse than HC at both conditions, with 3 and 12 letters (all p<0.05). The Post-hoc comparisons between FEP-NAF and HC, as well as between FEP-NA and HC, highlighted significant differences (p<0.02). Instead, the comparison between FEP-NA and FEO-A did not prompt any significant difference (p>0.02). The analysis of SOA specificity scores showed a significant main difference between FEP and HC in the condition with 3 letters (p<0.05). The Post-hoc comparisons showed that FEP-NA had significantly poorer Specificity scores (3 letters condition) than HC (p<0.02). All the other comparisons did not give any significant result (all p>0.02). FEPs and HCs groups’ mean (+/- SD) neurocognitive scores are summarized in Table 6. Comparison between non-affective and affective FEP patients are detailed in S2 Table.

thumbnail
Table 6. Cognitive and neuropsychological data in FEP and HC.

https://doi.org/10.1371/journal.pone.0272873.t006

Correlation analysis results

We performed correlation analysis between the variables which resulted to be statistically significant in previous (Bonferroni corrected) t-tests on linguistic assessment. In the group of FEP, local coherence errors (both ambiguous and missing scores) negatively correlated with verbal IQ (p<0.05). In the Group of HC, local coherence errors (ambiguous) correlate negatively with SOA12 sensitivity (p<0.05). After correction (alpha level = 0.05 divided by the number of correlations), no correlations survived, neither in the group of FEP nor in the group of HC. Results are detailed as S3 Table.

ML results

Our model trained only on language production variables reached an accuracy of 76.36% in discriminating between FEP and HC. Fig 1 shows the relative importance of each linguistic variable in distinguishing FEP from HC in this model (XGBoost’s result feature gain). Specifically, the three variables with the highest predictive power are semantic shifts, followed by lexical informative units and utterances with semantic errors (Fig 1). Various metrics (precision, recall, f1-score and accuracy) predicting group condition obtained on several sets of variables including and/or combining linguistic data with sociodemographic (gender, age, educational level) and/or neuropsychological (SOA, n-Back, IQ verbal) measures are summarized in Table 7. ML analysis was not able to discriminate between FEP-NA and FEP-A, possibly due to the small and not balanced samples.

thumbnail
Table 7. XGBoost models precision, recall, f1-score and accuracy metrics by different predictive variables sets.

https://doi.org/10.1371/journal.pone.0272873.t007

Discussion

This study focused on the identification of potential language production difficulties in psychotic patients as early as in their first episode of disease. To the best of our knowledge, this is the first study assessing patients with FEP language production skills using a multilevel procedure for the analysis of narrative production abilities through a standardized procedure that aims at minimizing to the bone the subjective bias of scale raters thus reaching the highest reproducibility possible.

First, we evaluated whether patients with FEP would show deficits in linguistic performance compared to HC. Results showed that FEP patients have significant deficits in language production at the micro-level of speech, which includes speech measures making up for the intra-phrasal level of discourse construction. In particular, FEP patients had a productive style that was characterized by a “lower speech rate” and “shorter utterances” compared to HC, meaning that they verbalized fewer well-formed words per minute and constructed sentences which were composed of fewer words. Patients also showed a tendency to produce “more phonological paraphasias” than HC, meaning that the generally shorter sentences they produced also contain some errors, although the absolute values of paraphasias are very low in both HC and FEP to draw conclusive results. By contrast, impairments in speech rate and length of discourse are more relevant in absolute quantitative terms (119 vs 134 words per minute and 6 vs 7 word per sentence in FEP and HC, respectively). This is in line with other literature findings on psychotic patients, with poverty of speech being one of the most consistent findings in these patients, not only in chronic populations, but also at the onset of disease [5, 6, 18, 26]. Narratives of people with FEP also have a lower percentage of syntactic completeness, meaning that their discourses are not only shorter but also less complex with respect to controls. Moreover, our analysis shows a reduced use of lexical fillers by FEP compared to healthy subjects. Lexical fillers are meaningless words that we usually use during everyday colloquial speech and that interrupt the flow of a sentence by filling silent moments between ordinary (non-filler) words or sentences. For example, ‘kind of’ or ‘sort of’ or ‘y’know’ are common lexical fillers in the English language. Interestingly, fillers during speech occur when a person is engaged in verbal working memory and word retrieval tasks, being such mental operations anatomically sustained by large-scale brain networks encompassing associative cortices [56]. The fact that FEP uses fewer fillers than controls may suggest a difficulty in accessing verbal memory to retrieve well-formed words, which could at least partly explain the poverty of speech. Overall, our results on FEP show a similar pattern of micro-linguistic impairments to that observed in chronic patients with schizophrenia in a previous paper by our group using an analogue linguistic analysis [14]. Similarity refers not only to the types of linguistic abilities which are most impaired, but also to absolute values; in fact, the severity of the impairment does not change much between onset of disease and the chronic stages. Such similarity suggests that deficits at the micro-linguistic productive level would be constant through the course of illness, although more data should be collected on a direct comparison between the two populations to deepen this issue. In regards to macro-linguistic abilities, patients showed a slightly more ambiguous choice of those elements of a discourse which grant cohesion in sentences, i.e. by using words with unclear referents (‘he’ instead of ‘she’). The final effect for the listener is a reduction of the local coherence of the narration. Moreover, patients showed a tendency to construct utterances with more semantic errors than controls. However, while these differences were in fact statistically significant between FEP and HC, the absolute number of these types of error in both groups was little, especially when looking at the local coherence items. However, it is also meaningful that such impairments in the macro-linguistic level of linguistic processing are in line with previous literature findings in chronic [14] and FEP patients, both producing a more disorganized speech with aberrant use of connectives [30] and choosing more peculiar words than controls [26]. As for the second aim of our study, we subsequently performed post-hoc comparisons between FEP-A vs FEP-NA, FEP-NA vs HC and FEP-A vs HC. We considered language production to possibly be differently altered in affective and non-affective psychotic patients, given that the clinical presentation of these two types of patients is often significantly different and we already have some evidence from previous literature [21, 27, 57]. Interestingly, the analysis confirmed significant impairments in speech rate and mean length of utterances in both groups of affective and non-affective patients when compared to HC, strengthening the thesis that all psychotic patients show to some extent an impairment in same productive aspects of phrasal construction (speech rate and mean length of utterances), while other results were less consistent. When FEP-NA and FEP-A were put in direct comparisons, however, there was no significant difference between the two populations on language production. Nonetheless, while no difference was strong enough to be significant, patients in the FEP-NA group did display a relevantly lower percentage of syntactic completeness than those in the FEP-A group, which more or less performed as well as the HC group on this item. When put beside the results by Docherty et al [33], which found schizophrenic patients to produce utterances with a significantly lower syntactic complexity than patients with mania even at an early stage of illness, findings may point towards the hypothesis that patients in the schizophrenia-spectrum, even at onset of disease, show difficulties in the syntactic level of language production. Nevertheless, more data is needed to further evaluate this matter.

We also hypothesized that language production abilities of our patients would be to some extent independent from other cognitive impairments, while some others (e.g., language planning and discourse organization) would be more related to cognitive skills (e.g., executive functions). Patients with FEP did indeed perform worse than controls in our neuropsychological tasks (verbal-IQ, n-back, SOA), in line with current literature [37, 5866]. However, in our sample, the language deficits observed in FEP and HC did not appear to be related to working memory and to the ability to process contextual information, after correction for multiple comparisons. This might point toward the idea that language production impairment at the onset of the illness appears to not be subdued to impairments in other cognitive abilities, but it is perhaps independent (at least the variables that we have explored and whose main effect in t-tests are statistically significant). More research should be conducted on this matter in order to confirm these results and reach a greater understanding of their causes.

Finally, the last aim of our study was to evaluate whether language production measures could be used to build a predictive model discriminating between FEP and HC. We therefore trained several ML models using different sets of features, also including clinical, socio-demographic and neuropsychological measures. Firstly, prediction accuracy of the model using language production variables only reaches 76.36%. This is quite interesting, when considering that the dummy classifier usually has a prediction accuracy of around 60%. Among the linguistic variables, those with the highest accuracy in predicting groups (FEP vs HC) were semantic shifts (it occurred when the concept in the interrupted preceding utterance was not resumed in the following sentence), lexical informative units (words that were not only well-formed from a phonological point of view, but also grammatically and pragmatically accurate) and utterances with semantic errors (Fig 1). Overall, such result points at a difficulty in the group of FEP in accessing semantically appropriate and accurately formed words and maintaining discourse coherence, which is in line with the results already described with reference to the first aim of the study (in particular, the reduced use of lexical fillers and the impairment in local coherence). Secondly, our ML results showed that GAF alone can predict the groups of FEP and HC with an accuracy of 97.90%. Also, neuropsychological measures have a predictive power of 99%. At a first glance, such results do not seem in favor of linguistic data, but some crucial issues should be considered: 1) although it is very easy and brief to administer, the Global Assessment of Functioning Scale score is (as the name suggests) a global measure, not specifically focused on psychosis but rather on the level of severity of symptoms and/or functioning characterizing several psychopathological conditions (ranging from personality disorders to anxiety, psychosis etc). The limited specificity of this scale and the impossibility to clearly distinguish between symptoms and functioning (although related) limit the informative power of this scale and its score; 2) the set of neuropsychological variables used in our ML analyses included SOA, n-back, and verbal-IQ. When combined together, their predictive power reached 99%, but such a result implies having a pc available (we used computerized versions of SOA and n-back), the presence of a neuropsychologist as part of the clinical staff and an assessment of around 1 hour. If, by contrast, we consider the single neuropsychological variable/instrument, the predictive power decreases to 76.67% for the SOA and 79.76% in the case of n-back, being these values comparable to the 76.36% reached by our set of linguistic data. Finally, Verbal IQ is 85.31% predictive with respect to the groups (FEP and HC) but, like GAF, it still represents a general measure. Based on such premises, we think that the use of language in the assessment of FEP and in classifying groups represents an extremely useful tool. Specifically, we think that linguistic deficits represent a core dimension of psychosis, being present both at the first stages and in the chronic phase of the illness and covering a large range of linguistic dimensions (both receptive and productive), as showed by a series of publications by our group [1317] and elsewhere [i.e. 27]. In the present study, we used a very brief and simple task which consisted in the description by participants of a series of vignettes for a total of 10–30 seconds. Despite the simplicity and brevity of the task, the linguistic variables we used performed as well as the n-back in predicting groups. We also tried to build a predictive model that could discriminate between FEP-NA and FEP-A patients on all the same models performed to discriminate between FEP and HC, with no conclusive results. The attempt to predict the groups in this case did not indeed perform better than the dummy classifier, possibly due to the small number of subjects and the excessive imbalance in the size of the two groups (95 vs 38 subjects, respectively). Taken together, these results support the hypothesis that language production skills are impaired in patients with FEP at both micro- and macro-linguistic levels, being such deficits not related to other cognitive domains in our sample. Furthermore, semantic deficits were the most predictive of the group of FEP vs HC in the ML analyses. Importantly, the use of a narrative production task and of a multilevel procedure for the analysis of narrative discourse production allowed us to assess language in an ecological setting. This also allowed us to avoid the results’ biases highlighted by Barch and Berenbaum [67], with tasks with fewer directions yielding more negative thought disorders and tasks with vague topics yielding more positive thought disorders. Furthermore, studies such as the current investigation point toward the auspicial transition from a clinical practice based on clinical observation alone to the much more reliable “measured-based care” [68], with the adoption of a systematic analysis of language and which classifies patients on objective features using ML. On the other hand, our study also has some limitations. Firstly, because the patients were recruited within a larger project, they were all outpatients. This means that patients who had a psychotic episode at its worst were not evaluated in this study. Indeed, patients’ GAF scores were only moderately impaired. Our results should not then be considered definitive and representative of the whole population of patients with FEP. Also, the absence of significant differences between language production impairments in FEP-A and FEP-NA as shown by statistical analysis is not at all to be seen as conclusive, and further research should be done on the matter. As for ML analysis, the difference in sample sizes between FEP-A and FEP-NA (38 vs. 95) unfortunately did not allow us to obtain conclusive results. Balancing the datasets for FEP-A and FEP-NA was not possible because of the reduced sample size, as it would have produced unreliable results. As a limitation of our approach, it has also to be mentioned that we applied a time-consuming linguistic analysis method, which limits its application in clinical context. Despite that, we still support the importance of including such assessment in the first phases of psychosis. In particular, the recent advance of automatic methods for the analysis of natural samples of speech can tremendously reduce this limit (see [27] for an extensive review of such methods). Finally, our analysis can represent a first step for the construction of new linguistic tools based on the most predictive linguistic variables as shown by our statistical and ML analyses.

Supporting information

S1 Table. Sociodemographic and clinical data of the sample of FEP-A and FEP-NA.

FEP-A, First Episode Psychosis–Affective; FEP-NA, First Episode Psychosis–Non-Affective; GAF, Global Assessment of Functioning; PANSS, Positive and Negative Syndrome Scale, General psychopathology subscale; HAM-D, Hamilton’s Depression Rating Scale; BRMRS, Bech-Rafaelsen Mania Rating Scale; DUP, Duration of untreated psychosis.

https://doi.org/10.1371/journal.pone.0272873.s003

(DOCX)

S2 Table. Cognitive and neuropsychological data in FEP-A and FEP-A.

FEP-A, First Episode Psychosis–Affective; FEP-NA, First Episode Psychosis–Non-Affective; IQ, Intelligence Quotient; TIB, Brief Intelligence Test.

https://doi.org/10.1371/journal.pone.0272873.s004

(DOCX)

S3 Table. Correlations between linguistic and neuropsychological variables.

https://doi.org/10.1371/journal.pone.0272873.s005

(DOCX)

Acknowledgments

THE GET UP GROUP. GET UP—Genetics, Endophenotypes, Treatment: Understanding Early Psychosis.

National Coordinator: Professor Mirella Ruggeri (Verona) Email: mirella.ruggeri@univr.it.

Leading Project: PIANO (Psychosis: Early Intervention and Assessment of Needs and Outcome).

Scientific Coordinator: Mirella Ruggeri (Verona).

Leading administrative institution: Azienda Ospedaliera Universitaria Integrata Verona, Regione Veneto.

Coordinating center: Maria Elena Bertani, Sarah Bissoli, Chiara Bonetto, Doriana Cristofalo, Katia De Santi, Antonio Lasalvia, Silvia Lunardi, Valentina Negretto, Sara Poli, Sarah Tosato, Maria Grazia Zamboni, Mario Ballarin.

Project: TRUMPET (TRaining and Understanding of Service Models for Psychosis Early Treatment).

Scientific coordinator: Giovanni De Girolamo (Bologna and Brescia).

Leading administrative institution: Agenzia Sanitaria e Sociale Regionale, Regione Emilia Romagna.

Coordinating center: Angelo Fioritti, Giovanni Neri, Francesca Pileggi, Paola Rucci.

Project: GUITAR (Genetic data Utilization and Implementation of Targeted Drug Administration in the Clinical Routine).

Scientific coordinator: Massimo Gennarelli (Brescia).

Leading administrative institution: IRCCS Centro S.Giovanni di Dio Fatebenefratelli, Brescia.

Coordinating center: Luisella Bocchio Chiavetto, Catia Scasselatti, Roberta Zanardini.

Project: CONTRABASS Cognitive Neuroendophenotypes for Treatment and Rehabilitation of Psychoses: Brain Imaging, Inflammation and Stress.

Scientific coordinator: Paolo Brambilla (Udine and Verona).

Leading administrative institution: Azienda Ospedaliera Universitaria Integrata, Verona, Regione Veneto.

Coordinating center: Marcella Bellani, Alessandra Bertoldo, Veronica Marinelli, Valentina Negretto, Cinzia Perlini, Gianluca Rambaldelli.

Enrolment and treatment research units.

Research unit Western Veneto.

Coordinator: Antonio Lasalvia (Verona).

Leading administrative institution: Azienda Ospedaliera Universitaria Integrata, Verona.

Coordinating center: Mariaelena Bertani, Sarah Bissoli, Lorenza Lazzarotto.

Participating MHCs: TAU Arm: Ulss 3 (Bassano), Ulss 4 Alto Vicentino (Thiene), Ulss 5 Montecchio (CSM ‘Centro/Sud’), Ulss 6 Vicenza (Secondo CSM), Ulss 18 Rovigo (Rovigo), Ulss 20 Verona (II° Servizio), Ulss 22 Bussolengo (Isola della Scala). Experimental Arm: Ulss 5 Montecchio (Nord), Ulss 6 Vicenza (Primo CSM; CSM Noventa), Ulss 18 Rovigo (Badia), Ulss 19 Adria (Adria), Ulss 20 Verona (I° Servizio; III° Servizio; IV° Servizio CSM ‘La Filanda’),Ulss 21 Legnago (CSM ‘Il Tulipano’; CSM ‘il Girasole’).

MHC reference contacts: Sonia Bardella, Francesco Gardellin, Dario Lamonaca, Antonio Lasalvia, Marco Lunardon, Renato Magnabosco, Marilena Martucci, Stylianos Nicolau, Francesco Nifosì, Michele Pavanati, Massimo Rossi, Carlo Piazza, Gabriella Piccione, Alessandra Sala, Annalisa Sale, Benedetta Stefani, Spyridon Zotos.

CBT staff: Mirko Balbo, Ileana Boggian, Enrico Ceccato, Rosa Dall’Agnola, Francesco Gardellin, Barbara Girotto, Claudia Goss, Dario Lamonaca, Antonio Lasalvia, Roberta Leoni, Alessia Mai, Annalisa Pasqualini, Michele Pavanati, Carlo Piazza, Gabriella Piccione, Stefano Roccato, Alberto Rossi, Annalisa Sale, Stefania Strizzolo, Spyridon Zotos, Anna Urbani.

Family intervention staff: Flavia Aldi, Barbara Bianchi, Paola Cappellari, Raffaello Conti, Laura De Battisti, Ermanna Lazzarin, Silvia Merlin, Giuseppe Migliorini, Tecla Pozzan, Lucio Sarto, Stefania Visonà.

Case management staff: Andrea Brazzoli, Antonella Campi, Roberta Carmagnani, Sabrina Giambelli, Annalisa Gianella, Lino Lunardi, Davide Madaghiele, Paola Maestrelli, Lidia Paiola, Elisa Posteri, Loretta Viola, Valentina Zamberlan, Marta Zenari.

Staff for biological sample processing and support for brain imaging procedures: Sarah Tosato, Martina Zanoni, Giovanni Bonadonna, Mariacristina Bonomo.Research unit Eastern Veneto. Coordinator: Paolo Santonastaso (Padova).

Leading administrative institution: University of Padova.

Coordinating center: Carla Cremonese, Paolo Scocco, Angela Veronese.

Participating MHCs: TAU Arm: Ulss 8 (Castelfranco), Ulss 9 (Treviso Nord; Oderzo), Ulss 10 (San Donà di Piave), Ulss 12 (Venezia; Mestre sud), Ulss 13 (Dolo), Ulss 14 (Piove di Sacco), Ulss 15 (Cittadella), Ulss 16 (II° Servizio), Ulss 17 (Este; Montagnana). Experimental Arm: Ulss 8 (Montebelluna; Valdobbiadene), Ulss 9 (Treviso; Mogliano Veneto), Ulss 10 (PortogReserach Unitaro), Ulss 12 (Mestre Centro), Ulss 13 (Mirano), Ulss 14 (Chioggia I°; Cavarzere), Ulss 15 (Camposanpiero), Ulss 16 (I° Srvizio; III° Servizio), Ulss 17 (Monselice; Conselve).

MHC reference contacts: Patrizia Anderle, Andrea Angelozzi, Isabelle Amalric Gabriella Baron, Enrico Bruttomesso Fabio Candeago, Franco Castelli, Maria Chieco, Carla Cremonese, Enrico Di Costanzo, Mario Derossi, Michele Doriguzzi, Osvaldo Galvano, Marcello Lattanzi, Roberto Lezzi, Marisa Marcato, Alessandro Marcolin, Franco Marini, Manlio Matranga, Donato Scalabrin, Maria Zucchetto, Flavio Zadro.

CBT staff: Giovanni Austoni, Maria Bianco, Francesca Bordino, Filippo Dario, Alessandro De Risio, Aldo Gatto, Simona Granà, Emanuele Favero, Anna Franceschini, Silvia Friederici, Vanna Marangon, Michela Pascolo, Luana Ramon, Paolo Scocco, Angela Veronese, Stefania Zambolin, Rossana Riolo.

Family intervention staff: Antonella Buffon, Carla Cremonese, Elena Di Bortolo, Silvia Friederici, Stefania Fortin, Marisa Marcato, Francesco Matarrese, Simona Mogni, Novella Codemo, Alessio Russi, Alessandra Silvestro, Elena Turella, Paola Viel, Anna Dominoni.

Case management staff: Lorenzo Andreose, Mario Boemio, Loretta Bressan, Arianna Cabbia, Elisabetta Canesso, Romina Cian, Claudia Dal Piccol, Maria Manuela Dalla Pasqua, Anna Di Prisco, Lorena Mantellato, Monica Luison, Sandra Morgante, Mirna Santi, Moreno Sacillotto, Mauro Scabbio, Patrizia Sponga, MLuisa Sguotto, Flavia Stach, MGrazia Vettorato, Giorgio Martinello, Francesca Dassiè, Stefano Marino, Linda Cibiniel, Ilenia Masetto, Marisa Marcato.

Staff for biological sample processing and support for brain imaging procedures: Oscar Cabianca, Amalia Valente, Livio Caberlotto, Alberto Passoni, Patrizia Flumian, Luigino Daniel, Massimo Gion, Saverio Stanziale, Flora Alborino, Vladimiro Bortolozzo, Lucio Bacelle, Leonarda Bicciato, Daniela Basso, Filippo Navaglia, Fabio Manoni, Mauro Ercolin.

Research unit Emilia.

Coordinators: Giovanni Neri (Modena), Franco Giubilini (Parma).

Leading administrative institution: Azienda ULSS, Parma.

Coordinating center: Massimiliano Imbesi, Emanuela Leuci, Fausto Mazzi, Enrico Semrov.

Participating MHCs: TAU Arm: Piacenza (Castel S.Giovanni), Parma (Parma Est; Sud Est; Valli Taro e Ceno), Reggio Emilia (CastelNovo nei Monti; Montecchio), Modena (Mirandola; Polo Ovest; Sassuolo; Pavullo). Experimental Arm: Piacenza (Piacenza; Fiorenzuola), Parma (Nord; Ovest; Fidenza), Reggio Emilia (Correggio; Guastalla; Reggio Emilia III; Reggio Emilia; Scandiano), Modena (Carpi; Polo Est; Vignola).

MHC reference contacts: Silvio Anelli, Mario Amore, Laura Bigi, Welsch Britta, Giovanna Barazzoni Anna, Uobes Bonatti, Maria Borziani, Stefano Crosato, Isabella Fabris, Raffaele Galluccio, Margherita Galeotti, Mauro Gozzi, Vanna Greco, Emanuele Guagnini, Stefania Pagani, Silvio Maccherozzi, Raffaello Malvasi, Francesco Marchi, Ermanno Melato, Elena Mazzucchi, Franco Marzullo, Pietro Pellegrini, Nicoletta Petrolini, Paolo Volta.

CBT staff: Silvio Anelli, Franca Bonara, Elisabetta Brusamonti, Roberto Croci, Ivana Flamia, Francesca Fontana, Romina Losi, Fausto Mazzi, Roberto Marchioro, Stefania Pagani, Luigi Raffaini, Luca Ruju, Antonio Saginario, MGrazia Tondelli, Donatella Marrama.

Family intervention staff: Lucia Bernardelli, Federica Bonacini, Annaluisa Florindo, Marina Merli, Patrizia Nappo, Lorena Sola, Ornella Tondelli, Matteo Tonna, MTeresa Torre, Morena Tosatti, Gloria Venturelli, Daria Zampolla.

Case management staff: Antonia Bernardi, Cinzia Cavalli, Lorena Cigala, Cinzia Ciraudo, Antonia Di Bari, Lorena Ferri, Fabiana Gombi, Sonia Leurini, Elena Mandatelli, Stefano Maccaferri, Mara Oroboncoide, Barbara Pisa, Cristina Ricci.

Staff for biological sample processing and support for brain imaging procedures: Enrica Poggi, Mara Oroboncoide, Corrado Zurlini, Monica Malpeli, Rossana Colla, Elvira Teodori, Luigi Vecchia, Rocco D’Andrea, Tommaso Trenti, Paola Paolini, Fausto Mazzi, Paolo Carpeggiani.

Research unit Romagna.

Coordinators: Francesca Pileggi (Bologna), Daniela Ghigi (Rimini).

Leading administrative institution: Azienda ULSS, Rimini.

Coordinating center: Mariateresa Gagliostro, Michela Pratelli, Paola Rucci.

Participating MHCs: TAU Arm: Bologna (Zanolini; Scalo; Casalecchio; Vergato; San Giovanni), Ferrara (CSA Ferrara; SIPI Ferrara Sud; Codigoro; Portomaggiore), Ravenna (Ravenna; Fenza), Forlì (Forlì), Cesena (Cesena), Rimini (Riccione). Experimental Arm: Bologna (Mazzacorati; Tiarini, Nani; S. Lazzaro; Budrio; San Giorgio), Imola (UOT_Imola), Ferarra (Copparo; Ferrara Nord; Cento), Ravenna (Lugo), Cesena (Rubicone), Rimini (Rimini).

MHC reference contacts: Antonio Antonelli, Luana Battistini, Francesca Bellini, Eva Bonini, Caterina Bruschi Rossella Capelli, Cinzia DiDomizio, Chiara Drei, Giuseppe Fucci, Alessandra Gualandi, Maria Rosaria Grazia, AnnaM. Losi, Federica Mazzanti Paola Mazzoni, Daniela Marangoni, Giuseppe Monna, Marco Morselli, Alessandro Oggioni, Silvio Oprandi, Walter Paganelli, Morena Passerini, Maria Piscitelli, Gregorio Reggiani, Gabriella Rossi, Federica Salvatori, Simona Trasforini, Carlo Uslenghi, Simona Veggetti>.

CBT staff: Giovanna Bartolucci, Rosita Baruffa, Francesca Bellini, Raffaella Bertelli, Lidia Borghi, Patrizia Ciavarella, Cinzia DiDomizio, Giuseppe Monna, Alessandro Oggioni, Elisabetta Paltrinieri, Francesco Rizzardi, Piera Serra, Damiano Suzzi, Uslenghi Carlo, Maria Piscitelli.

Family intervention staff: Paolo Arienti, Fabio Aureli, Rosita Avanzi, Vincenzo Callegari, Alessandra Corsino, Paolo Host, Rossella Michetti, Michela Pratelli,Francesco Rizzo, Paola Simoncelli, Elena Soldati, Eraldo Succi.

Case management staff: Massimo Bertozzi, Elisa Canetti, Luca Cavicchioli, Elisa Ceccarelli, Stefano Cenni, Glenda Marzola, Vanessa Gallina, Carla Leoni, Andrea Olivieri, Elena Piccolo, Sabrina Ravagli, Rosaria Russo, Daniele Tedeschini.

Staff for biological sample processing and support for brain imaging procedures: Marina Verenini, Walter Abram, Veronica Granata, Alessandro Curcio, Giovanni Guerra, Samuela Granini, Lara Natali, Enrica Montanari, Fulvia Pasi, Umbertina Ventura, Stefania Valenti, Masi Francesca, Rossano Farneti, Paolo Ravagli, Romina Floris, Otello Maroncelli, Gianbattista Volpones, Donatella Casali.

Research unit Firenze.

Coordinator: Maurizio Miceli (Firenze).

Leading administrative institution: Azienda Sanitaria di Firenze.

Coordinating center: Maurizio Miceli.

Participating MHCs: TAU Arm: MOM SMA 5; MOM SMA 8; MOM SMA 11; MOM SMA 12. Experimental Arm: MOM SMA 3; MOM SMA 7; MOM SMA 9; MOM SMA 10.

MHC reference contacts: Andrea Bencini, Massimo Cellini, Luca De Biase, Leonardo Barbara, Liedl Charles, Maurizio Miceli, Cristina Pratesi, Andrea Tanini.

CBT staff: Massimo Cellini, Maurizio Miceli, Riccardo Loparrino, Cristina Pratesi, Cinzia Ulivelli.

Family intervention staff: Cristina Cussoto, Nico Dei, Enrico Fumanti, Manuela Pantani, Gregorio Zeloni.

Case management staff: Rossella Bellini, Roberta Cellesi, Nadia Dorigo, Patrizia Gullì, Luisa Ialeggio, Maria Pisanu.

Staff for biological sample processing and support for brain imaging procedures: Graziella Rinaldi, Angela Konze.

Research unit Milano Niguarda.

Coordinator: Angelo Cocchi (Milano).

Leading administrative institution: Azienda Ospedaliera Ospedale Niguarda Ca’ Granda, Milano.

Coordinating center: Anna Meneghelli.

Participating MHCs: TAU Arm: corso Plebisciti; via Mario Bianco. Experimental Arm: via Cherasco e via Livigno; via Litta Modignani.

MHC reference contacts: Maria Frova, Emiliano Monzani, Alberto Zanobio, Marina Malagoli, Roberto Pagani.

CBT staff: Simona Barbera, Carla Morganti, Emiliano Monzani, Elisabetta Sarzi Amadè.

Family intervention staff: Virginia Brambilla, Anita Montanari.

Case management staff: Giori Caterina, Carmelo Lopez.

Staff for biological sample processing and support for brain imaging procedures: Alessandro Marocchi, Andrea Moletta, Maurizio Sberna, M. Teresa Cascio.

Research unit Milano S. Paolo.

Coordinator: Silvio Scarone (Milano). Leading administrative institution: Azienda ULSS San Paolo, Milano.

References

  1. 1. Marini A, Spoletini I, Rubino IA, et al. The language of schizophrenia: an analysis of micro and macrolinguistic abilities and their neuropsychological correlates. Schizophr Res. 2008;105(1–3):144–155. pmid:18768300
  2. 2. Marengo JT, Harrow M. Schizophrenic thought disorder at follow-up. A persistent or episodic course? Arch Gen Psychiatry. 1987 Jul;44(7):651–9. pmid:3606331.
  3. 3. Marengo JT, Harrow M. Longitudinal courses of thought disorder in schizophrenia and schizoaffective disorder. Schizophr Bull. 1997;23(2):273–85. pmid:9165637.
  4. 4. Spalletta G, Tomaiuolo F, Marino V, Bonaviri G, Trequattrini A, Caltagirone C. Chronic schizophrenia as a brain misconnection syndrome: a white matter voxel-based morphometry study. Schizophr Res. 2003 Nov 1;64(1):15–23. pmid:14511797.
  5. 5. Gourzis P, Katrivanou A, Beratis S. Symptomatology of the initial prodromal phase in schizophrenia. Schizophr Bull. 2002;28(3):415–29. pmid:12645674.
  6. 6. Barajas A, Pelaez T, González O, Usall J, Iniesta R, Arteaga M, et al. Predictive capacity of prodromal symptoms in first-episode psychosis of recent onset. Early Interv Psychiatry. 2019 Jun;13(3):414–424. Epub 2017 Nov 8. pmid:29116670.
  7. 7. Crow TJ. Is schizophrenia the price that Homo sapiens pays for language? Schizophr Res. 1997 Dec 19;28(2–3):127–41. pmid:9468348.
  8. 8. Crow TJ. Schizophrenia as the price that homo sapiens pays for language: a resolution of the central paradox in the origin of the species. Brain Res Brain Res Rev. 2000 Mar;31(2–3):118–29. pmid:10719140.
  9. 9. Bleich-Cohen M, Hendler T, Kotler M, Strous RD. Reduced language lateralization in first-episode schizophrenia: an fMRI index of functional asymmetry. Psychiatry Res. 2009 Feb 28;171(2):82–93. Epub 2009 Jan 29. pmid:19185468.
  10. 10. van Veelen NM, Vink M, Ramsey NF, Sommer IE, van Buuren M, Hoogendam JM, et al. Reduced language lateralization in first-episode medication-naive schizophrenia. Schizophr Res. 2011 Apr;127(1–3):195–201. Epub 2011 Jan 14. pmid:21237617.
  11. 11. Covington MA, He C, Brown C, Naçi L, McClain JT, Fjordbak BS, et al. Schizophrenia and the structure of language: the linguist’s view. Schizophr Res. 2005 Sep 1;77(1):85–98. Epub 2005 Apr 2. pmid:16005388.
  12. 12. Bellani M, Perlini C, Brambilla P. Language disturbances in schizophrenia. Epidemiol Psichiatr Soc. 2009 Oct-Dec;18(4):314–7. pmid:20170045.
  13. 13. Tavano A, Sponda S, Fabbro F, Perlini C, Rambaldelli G, Ferro A, et al. Specific linguistic and pragmatic deficits in Italian patients with schizophrenia. Schizophr Res. 2008 Jul;102(1–3):53–62. Epub 2008 Apr 7. pmid:18396387.
  14. 14. Perlini C, Marini A, Garzitto M, Isola M, Cerruti S, Marinelli V, et al. Linguistic production and syntactic comprehension in schizophrenia and bipolar disorder. Acta Psychiatr Scand. 2012 Nov;126(5):363–76. Epub 2012 Apr 17. pmid:22509998.
  15. 15. Perlini C, Bellani M, Finos L, Lasalvia A, Bonetto C, Scocco P, et al. Non literal language comprehension in a large sample of first episode psychosis patients in adulthood. Psychiatry Res. 2018 Feb;260:78–89. Epub 2017 Nov 10. pmid:29175503.
  16. 16. Caletti E, Delvecchio G, Andreella A, Finos L, Perlini C, Tavano A, et al. Prosody abilities in a large sample of affective and non-affective first episode psychosis patients. Compr Psychiatry. 2018 Oct;86:31–38. Epub 2018 Jul 26. pmid:30056363.
  17. 17. Delvecchio G, Caletti E, Perlini C, Siri FM, Andreella A, Finos L, et al. Altered syntactic abilities in first episode patients: An inner phenomenon characterizing psychosis. Eur Psychiatry. 2019 Sep;61:119–126. Epub 2019 Aug 20. pmid:31442739.
  18. 18. de Boer JN, Brederoo SG, Voppel AE, Sommer IEC. Anomalies in language as a biomarker for schizophrenia. Curr Opin Psychiatry. 2020 May;33(3):212–218. pmid:32049766.
  19. 19. Blanchard MM, Jacobson S, Clarke MC, Connor D, Kelleher I, Garavan, H, et al. Language, motor and speed of processing deficits in adolescents with subclinical psychotic symptoms. Schizophrenia research. 2010; 123(1), 71–76 pmid:20580205
  20. 20. Pawełczyk A, Kotlicka-Antczak M, Łojek E, Pawełczyk T. Preliminary study of higher-order language and extralinguistic impairments in individuals with high clinical risk of psychosis and first episode of schizophrenia. Early Intervention in Psychiatry. 2019; 13(3), 369–378. pmid:28857488
  21. 21. Bedi G, Carrillo F, Cecchi GA, Slezak DF, Sigman M, Mota NB, et al. Automated analysis of free speech predicts psychosis onset in high-risk youths. NPJ Schizophr. 2015 Aug 26;1:15030. pmid:27336038; PMCID: PMC4849456.
  22. 22. Corcoran CM, Carrillo F, Fernández-Slezak D, Bedi G, Klim C, Javitt DC, et al. Prediction of psychosis across protocols and risk cohorts using automated language analysis. World Psychiatry. 2018 Feb;17(1):67–75. pmid:29352548; PMCID: PMC5775133.
  23. 23. Roche E, Lyne J, O’Donoghue B, Segurado R, Behan C, Renwick L, et al. The prognostic value of formal thought disorder following first episode psychosis. Schizophrenia Research. 2016; 178(1–3), 29–34. pmid:27639419
  24. 24. Liddle PF, Ngan ET, Caissie SL, Anderson CM, Bates AT, Quested DJ, et al. Thought and Language Index: an instrument for assessing thought and language in schizophrenia. The British Journal of Psychiatry. 2002; 181(4), 326–330. pmid:12356660
  25. 25. Kircher T, Krug A, Stratmann M, Ghazi S, Schales C, Frauenheim M, et al. A rating scale for the assessment of objective and subjective formal Thought and Language Disorder (TALD). Schizophrenia research. 2014; 160(1–3), 216–221. pmid:25458572
  26. 26. Ayer A, Yalınçetin B, Aydınlı E, Sevilmiş Ş, Ulaş H, Binbay T, et al. Formal thought disorder in first-episode psychosis. Compr Psychiatry. 2016 Oct;70:209–15. Epub 2016 Aug 9. pmid:27565775.
  27. 27. Corcoran CM, Cecchi GA. Using Language Processing and Speech Analysis for the Identification of Psychosis and Other Disorders. Biol Psychiatry Cogn Neurosci Neuroimaging. 2020 Aug;5(8):770–779. Epub 2020 Jun 14. pmid:32771179; PMCID: PMC7430500.
  28. 28. Irving J, Patel R, Oliver D, Colling C, Pritchard M, Broadbent M, et al. Using natural language processing on electronic health records to enhance detection and prediction of psychosis risk. Schizophrenia bulletin, 2021; 47(2), 405–414. pmid:33025017. PMCID: PMC7965059
  29. 29. Silva A, Limongi R, MacKinley M, Palaniyappan L. Small words that matter: linguistic style and conceptual disorganization in untreated first-episode Schizophrenia. Schizophrenia bulletin open. 2021; 2(1), sgab010. pmid:33937775 PMCID: PMC8072135
  30. 30. Mackinley M, Chan J, Ke H, Dempster K, Palaniyappan L. Linguistic determinants of formal thought disorder in first episode psychosis. Early intervention in psychiatry. 2021; 15(2), 344–351. pmid:32129010
  31. 31. Marini A, Boewe A, Caltagirone C, Carlomagno S. Age-related differences in the production of textual descriptions. J Psycholinguist Res. 2005 Sep;34(5):439–63. pmid:16177935.
  32. 32. McKenna PJ, Oh TM. Schizophrenic speech: Making sense of bathroots and ponds that fall in doorways. Cambridge University Press; 2005.
  33. 33. Docherty NM, DeRosa M, Andreasen NC. Communication disturbances in schizophrenia and mania. Archives of General Psychiatry. 1996; 53(4), 358–364. pmid:8634014
  34. 34. Kravariti E, Reichenberg A, Morgan K, Dazzan P, Morgan C, Zanelli JW, et al. Selective deficits in semantic verbal fluency in patients with a first affective episode with psychotic symptoms and a positive history of mania. Bipolar Disord. 2009 May;11(3):323–9. pmid:19419389.
  35. 35. Lott PR, Guggenbühl S, Schneeberger A, Pulver AE, Stassen HH. Linguistic analysis of the speech output of schizophrenic, bipolar, and depressive patients. Psychopathology. 2002 Jul-Aug;35(4):220–7. pmid:12239438.
  36. 36. Heinrichs RW. The primacy of cognition in schizophrenia. American Psychologist. 2005; 60(3), 229. pmid:15796677
  37. 37. Sheffield JM, Karcher NR, Barch DM. Cognitive Deficits in Psychotic Disorders: A Lifespan Perspective. Neuropsychol Rev. 2018 Dec;28(4):509–533. Epub 2018 Oct 20. pmid:30343458; PMCID: PMC6475621.
  38. 38. McCleery A., & Nuechterlein K. H. Cognitive impairment in psychotic illness: prevalence, profile of impairment, developmental course, and treatment considerations. Dialogues in clinical neuroscience; 2022.
  39. 39. Tracy JI, Glosser G, DellaPietra L. W-14. A cognitive/linguistic model of single-word production abnormalities in schizophrenia: Data from two case reports Brain and Cognition. 1996;30:311–315. PMCID: PMC6829172
  40. 40. Daffner KR, Searl MM. The dysexecutive syndromes. Handb Clin Neurol. 2008;88:249–67. pmid:18631695.
  41. 41. Radanovic M, Sousa RT, Valiengo L, Gattaz WF, Forlenza OV. Formal Thought Disorder and language impairment in schizophrenia. Arq Neuropsiquiatr. 2013 Jan;71(1):55–60. Epub 2012 Dec 18. pmid:23249974.
  42. 42. Ruggeri M, Bonetto C, Lasalvia A, De Girolamo G, Fioritti A, Rucci P, et al. A multi-element psychosocial intervention for early psychosis (GET UP PIANO TRIAL) conducted in a catchment area of 10 million inhabitants: study protocol for a pragmatic cluster randomized controlled trial. Trials. 2012 May 30;13:73. pmid:22647399; PMCID: PMC3464965.
  43. 43. World Health Organization. The ICD-10 classification of mental and behavioural disorders: clinical descriptions and diagnostic guidelines. World Health Organization; 1992.
  44. 44. Wing JK, Babor T, Brugha T, Burke J, Cooper JE, Geil R, et al. SCAN: Schedules for Clinical Assessment in Neuropsychiatry. Arch Gen Psychiatry. 1990; 47: 589±59. pmid:2190539
  45. 45. American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 4th ed., text rev.; 2000.
  46. 46. Kay SR, Fiszbein A, Opler LA. The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophr Bull. 1987;13(2):261–76. pmid:3616518.
  47. 47. Hamilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry. 1960;23:56–62. pmid:14399272 PMCID: PMC495331
  48. 48. Bech P, Rafaelsen OJ, Kramp P, Bolwig TG. The mania rating scale: scale construction and inter-observer agreement. Neuropharmacology. 1978 Jun;17(6):430–1. pmid:673161
  49. 49. World Medical Association. World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. Jama. 2013; 310(20), 2191–2194. pmid:24141714
  50. 50. Marini A, Andreetta S, del Tin S, Carlomagno S. A multi-level approach to the analysis of narrative language in aphasia. Aphasiology. 2011;25(11):1372–1392.
  51. 51. Paradis M., & Libben G. The assessment of bilingual aphasia. Lawrence Erlbaum Associates, Inc. 1987.
  52. 52. Colombo L, Sartori G, Brivio C. Stima del quoziente intellettivo tramite l’applicazione del TIB (Test Breve di Intelligenza). Giornale italiano di psicologia 2002;3:613–638.
  53. 53. Kirckner WK. Age differences in short-term retention of rapidly changing information. J Exp Psychol. 1958 Apr;55(4):352–8. pmid:13539317.
  54. 54. Asarnow RF, MacCrimmon DJ. Span of apprehension deficits during the postpsychotic stages of schizophrenia. A replication and extension. Arch Gen Psychiatry. 1981 Sep;38(9):1006–11. pmid:6116484.
  55. 55. Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining:2016. https://doi.org/10.1145/2939672.2939785
  56. 56. Sugiura A, Alqatan Z, Nakai Y, Kambara T, Silverstein BH, Asano E. Neural dynamics during the vocalization of ‘uh’or ‘um’. Scientific reports. 2020;10(1), 1–8. pmid:32686761 PMCID: PMC7371885
  57. 57. Solovay MR, Shenton ME, Holzman PS. Comparative studies of thought disorders. I. Mania and schizophrenia. Arch Gen Psychiatry. 1987 Jan;44(1):13–20. pmid:3800579.
  58. 58. Andreasen NC. Thought, language, and communication disorders. I. Clinical assessment, definition of terms, and evaluation of their reliability. Arch Gen Psychiatry. 1979 Nov;36(12):1315–21. pmid:496551
  59. 59. Gooding DC, Ott SL, Roberts SA, Erlenmeyer-Kimling L. Thought disorder in mid-childhood as a predictor of adulthood diagnostic outcome: findings from the New York High-Risk Project. Psychol Med. 2013 May;43(5):1003–12. Epub 2012 Aug 30. pmid:22932128.
  60. 60. Mesholam-Gately RI, Giuliano AJ, Goff KP, Faraone SV, Seidman LJ. Neurocognition in first-episode schizophrenia: a meta-analytic review. Neuropsychology. 2009 May;23(3):315–36. pmid:19413446.
  61. 61. Aas M, Dazzan P, Mondelli V, Melle I, Murray RM, Pariante CM. A systematic review of cognitive function in first-episode psychosis, including a discussion on childhood trauma, stress, and inflammation. Front Psychiatry. 2014 Jan 8;4:182. pmid:24409157; PMCID: PMC3884147.
  62. 62. Bagner DM, Melinder MR, Barch DM. Language comprehension and working memory language comprehension and working memory deficits in patients with schizophrenia. Schizophr Res. 2003 Apr 1;60(2–3):299–309. pmid:12591591.
  63. 63. Hitczenko K, Mittal VA, Goldrick M. Understanding Language Abnormalities and Associated Clinical Markers in Psychosis: The Promise of Computational Methods. Schizophr Bull. 2021 Mar 16;47(2):344–362. pmid:33205155.
  64. 64. Reilly JL, Harris MS, Khine TT, Keshavan MS, Sweeney JA. Antipsychotic drugs exacerbate impairment on a working memory task in first-episode schizophrenia. Biol Psychiatry. 2007 Oct 1;62(7):818–21. Epub 2007 Feb 14. pmid:17300756.
  65. 65. Laurent A, Biloa-Tang M, Bougerol T, Duly D, Anchisi AM, Bosson JL, et al. Executive/attentional performance and measures of schizotypy in patients with schizophrenia and in their nonpsychotic first-degree relatives. Schizophr Res. 2000 Dec 15;46(2–3):269–83. pmid:11120438.
  66. 66. Asarnow RF, Granholm E, Sherman T. Span of apprehension in schizophrenia. In Steinhauer S. R., Gruzelier J. H., & Zubin J. (Eds.), Handbook of schizophrenia, Vol. 5. Neuropsychology, psychophysiology, and information processing (p.335–370). Elsevier Science; 1991.
  67. 67. Barch DM, Berenbaum H. The effect of language production manipulations on negative thought disorder and discourse coherence disturbances in schizophrenia. Psychiatry Res. 1997 Jul 4;71(2):115–27. pmid:9255856.
  68. 68. Insel TR. Digital Phenotyping: Technology for a New Science of Behavior. JAMA. 2017 Oct 3;318(13):1215–1216. pmid:28973224.