VERBAL LEARNING

stop-users Differences in clinical and functional follow-up period using linear mixed effects (lme) models for of measures Patients who stopped using substances within the first two years after diagnosis had outcomes similar to those who had never used with less symptoms than episodic or persistent users. Both episodic and persistent users had lower rates of symptom remission than non-users, and persistent users also had more negative symptoms than those who stopped using. Discussion: Our findings emerge from one of very few long-term longitudinal studies examining substance use cessation in FEP with 10-year follow- up. The results convey hope that the detrimental effects of substance abuse on mental health may be significantly reversed if one stops the abuse in time. This can help patients who struggle with addiction with their motiva-tion to embrace abstinence. Background: Long-term use of cannabis has long been associated with changes in cognition, including memory and learning, particularly verbal learning in man. However, evidence regarding the neurobiological under-pinnings of impairments in memory following long-term cannabis use has not been consistent. Furthermore, to our knowledge none of the studies published to date have specifically investigated whether brain function differed between cannabis users and non-users while learning new information as estimated over repeated trials. Therefore, we aimed to investigate this. Methods: Twenty-one predominantly cannabis users (CU) who started using cannabis during adolescence and 21 healthy non-using controls (NU), completed a block design verbal paired associates learning task whilst undergoing functional Magnetic Resonance Imaging. The task required participants to learn and recall a set of word-pairs over 4 repeated trials. We examined the interaction between repetition and group (CU vs NU) on brain activation during encoding and recall condition using non- parametric repeated measures analysis of variance. Results: There was no significant difference in total recall score between CU and NU. However, there was a significant effect of repetition (p<0.001) on recall score, suggesting that there was a significant improvement in recall score over repeated trials across the two groups of participants. Furthermore, there was a significant interaction between repetition and group on recall score such that the change in recall score over repeated trials significantly differed (p =0.032) between the CU and NU groups. This was associated with a significant interaction (p =0.009) between group and repetition on activation in the midbrain bilaterally, extending to the, para-hippocampus, caudate and cingulate gyrus during the encoding condition. There was greater engagement of these regions in CU than in NU over repeated encoding trials. Discussion: These results suggest that verbal learning is slower and more effortful requiring greater engagement of critical brain areas involved in learning in cannabis users compared to non-users. Background: Nicotine use is higher among patients with schizophrenia (50–98%) than in general population (25–30%). This association can reflect a non-specific liability to substance use or specific effects of tobacco on symptoms severity or side effects. Studies about nicotine use and schizo- phrenia symptoms dimensions are controversial. Some of them showed a relation between severe nicotine use and higher positive symptoms and oth- ers presented a correlation between lower negative symptoms and nicotine use. That is why we aimed to verify whether nicotine use is associated with symptoms dimensions in patients with schizophrenia. Methods: Two hundred and seven outpatients were enrolled from the Programa de Esquizofrenia da Universidade Federal de São Paulo (PROESQ/ UNIFESP). Schizophrenia diagnosis was confirmed by Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I). Dimensional psychopathology was assessed with Positive and Negative Syndrome Scale (PANSS) and Fagerstrom Test for Nicotine Dependence. The PANSS items were grouped in five dimensions: positive, negative, disorganized/cognitive, mood/ depression and excitement/hostility. The total score of Fagerstrom Test for Nicotine Dependence was the index used for severity in nicotine dependence. We used Wilcoxon-mann- whitney test to compare the means of PANSS dimensions between nicotine users versus non nicotine use. Results: The patients mean age was 36.75 (SD 10.648), 69.1% were male, 48.3% reported lifetime tobacco use and 34.3% reported current tobacco use. Lower scores on negative dimension were associated with nicotine use (W = 5642.5, p-value = 0.046, effect size = 0.446). All p-values were corrected by Bonferroni test. Tests that evaluated the relationship between nicotine use and the total PANSS score or other dimensions were not statistically significant. Discussion: This study shows that nicotine use impacts negative symptoms of schizophrenia. Increase in hepatic metabolism leading to low antipsychotic blood levels has been previously documented in patients with schizophrenia. Thus, the observed a psychotic disorder and were also present in the control group. Discussion: We found that psychotic symptoms were not limited to patients with a specific ICD-10 diagnosis and were present in a wide range of ICD-10 disorders. These findings highlight the utility of detailed NLP-derived symptom data to better characterise psychotic disorders. Background: Separation of individuals into schizophrenia and bipolar diagnoses has long been questioned, with some suggesting that the classification impairs the understanding of etiology, the accuracy of prognoses, and treatment selection. In this study, we employed unbiased statistical techniques to identify subgroups of individuals with chronic illness using a large array of variables commonly evaluated at the bedside. We then validated the resulting groups by investigating age of onset, schizophrenia polygenic risk scores (PRS), and functional outcomes at a 1-year follow-up period. Our hypothesis was that transdiagnostic subgroups would be strati- fied based on illness onset whereby individuals with earlier onset would have higher genetic risk loading and poorer functional outcomes. Methods: Participants were selected from a longitudinal, naturalistic, multi- site project (PsyCourse) designed to investigate psychiatric illness course and outcomes. A total of 329 participants (age(SD)=45.7(12.6); 54% female; years of illness duration(SD) = 13.7(10.3)) with a DSM-IV diagnosis of schizophrenia, schizoaffective disorder, or bipolar disorder were assessed from 17 centers at baseline and 1-year follow-up periods. A clinical battery measuring sociodemographic, illness history, symptoms, cognition, and personality ques-tionnaires (199 variables)

and Roskilde (Denmark).As a result, DUP in the early detection area was reduced from 26 weeks median to 4 weeks median.All patients with First Episode Psychosis in the early detection area have been followed for ten years, and a twenty-year follow-up is to take place shortly.Symptom and function advantages of early detection and DUP reduction have been demonstrated as being significant throughout the follow-up period.Social and functional outcome have been increasingly emphasized as being key parameters, as these contribute to both quality of life and to financial costs in society.Substance use is common in first-episode psychosis (FEP) and has been linked to poorer outcomes with more severe psychopathology and higher relapse rates.Early substance discontinuation appears to improve symptoms and function.However, studies vary widely in their methodology, and few have examined patients longitudinally, making it difficult to draw conclusions for practice and treatment.Methods: We aimed to investigate the relationship between substance use and early abstinence and the long-term course of illness in a representative sample of FEP patients.Out of 301 included patients, 266 could be divided into four groups based on substance use patterns during the first two years of treatment: persistent users, episodic users, stop-users and non-users.Differences in clinical and functional during the follow-up period were assessed using linear mixed effects (lme) models for the analysis of repeated measures data.Results: Patients who stopped using substances within the first two years after diagnosis had outcomes similar to those who had never used with less symptoms than episodic or persistent users.Both episodic and persistent users had lower rates of symptom remission than non-users, and persistent users also had more negative symptoms than those who stopped using.Discussion: Our findings emerge from one of very few long-term longitudinal studies examining substance use cessation in FEP with 10-year followup.The results convey hope that the detrimental effects of substance abuse on mental health may be significantly reversed if one stops the abuse in time.This can help patients who struggle with addiction with their motivation to embrace abstinence.

T99. LONG-TERM CANNABIS USE ASSOCIATED WITH ALTERED FUNCTIONING DURING VERBAL LEARNING
Grace Blest-Hopley* ,1 , Aisling O'Neill 2 , Robin Wilson 3 , Vincent Giampietro 4 , Sagnik Bhattacharyya 3 1 King's College London, Institute of Psychiatry, Psychology, & Neuroscience Background: Long-term use of cannabis has long been associated with changes in cognition, including memory and learning, particularly verbal learning in man.However, evidence regarding the neurobiological underpinnings of impairments in memory following long-term cannabis use has not been consistent.Furthermore, to our knowledge none of the studies published to date have specifically investigated whether brain function differed between cannabis users and non-users while learning new information as estimated over repeated trials.Therefore, we aimed to investigate this.Methods: Twenty-one predominantly cannabis users (CU) who started using cannabis during adolescence and 21 healthy non-using controls (NU), completed a block design verbal paired associates learning task whilst undergoing functional Magnetic Resonance Imaging.The task required participants to learn and recall a set of word-pairs over 4 repeated trials.We examined the interaction between repetition and group (CU vs NU) on brain activation during encoding and recall condition using nonparametric repeated measures analysis of variance.Results: There was no significant difference in total recall score between CU and NU.However, there was a significant effect of repetition (p<0.001) on recall score, suggesting that there was a significant improvement in recall score over repeated trials across the two groups of participants.Furthermore, there was a significant interaction between repetition and group on recall score such that the change in recall score over repeated trials significantly differed (p =0.032) between the CU and NU groups.This was associated with a significant interaction (p =0.009) between group and repetition on activation in the midbrain bilaterally, extending to the, parahippocampus, caudate and cingulate gyrus during the encoding condition.There was greater engagement of these regions in CU than in NU over repeated encoding trials.Discussion: These results suggest that verbal learning is slower and more effortful requiring greater engagement of critical brain areas involved in learning in cannabis users compared to non-users.

T100. NICOTINE USE IMPACTS NEGATIVE SYMPTOMS SEVERITY IN SCHIZOPHRENIA
Hianna Oliveira* ,1 , Luccas Coutinho 1 , Cinthia Higuchi 1 , Cristiano Noto 1 , Rodrigo Bressan 1 , Ary Gadelha 1 1 Universidade Federal de São Paulo Background: Nicotine use is higher among patients with schizophrenia (50-98%) than in general population (25-30%).This association can reflect a non-specific liability to substance use or specific effects of tobacco on symptoms severity or side effects.Studies about nicotine use and schizophrenia symptoms dimensions are controversial.Some of them showed a relation between severe nicotine use and higher positive symptoms and others presented a correlation between lower negative symptoms and nicotine use.That is why we aimed to verify whether nicotine use is associated with symptoms dimensions in patients with schizophrenia.Methods: Two hundred and seven outpatients were enrolled from the Programa de Esquizofrenia da Universidade Federal de São Paulo (PROESQ/ UNIFESP).Schizophrenia diagnosis was confirmed by Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I).Dimensional psychopathology was assessed with Positive and Negative Syndrome Scale (PANSS) and Fagerstrom Test for Nicotine Dependence.The PANSS items were grouped in five dimensions: positive, negative, disorganized/cognitive, mood/ depression and excitement/hostility.The total score of Fagerstrom Test for Nicotine Dependence was the index used for severity in nicotine dependence.We used Wilcoxon-mann-whitney test to compare the means of PANSS dimensions between nicotine users versus non nicotine use.Results: The patients mean age was 36.75 (SD 10.648), 69.1% were male, 48.3% reported lifetime tobacco use and 34.3% reported current tobacco use.Lower scores on negative dimension were associated with nicotine use (W = 5642.5,p-value = 0.046, effect size = 0.446).All p-values were corrected by Bonferroni test.Tests that evaluated the relationship between nicotine use and the total PANSS score or other dimensions were not statistically significant.Discussion: This study shows that nicotine use impacts negative symptoms of schizophrenia.Increase in hepatic metabolism leading to low antipsychotic blood levels has been previously documented in patients with schizophrenia.Thus, the observed results can either indicate effect on primary negative symptoms or indirect effects through reduced D2 blockade caused by lower antipsychotic levels.Future quantitative analyzes and Longitudinal studies may better inform on direction of the association between nicotine dependence and negative symptoms in schizophrenia.However, the clinical classification of psychotic disorders has remained largely unchanged and is based on criterion-based diagnostic systems (such as ICD-10 and DSM-5) which do not necessarily reflect their underlying aetiology and pathophysiology.A more refined characterisation of clinical phenotype could help to improve our understanding of these disorders.Clinical data are increasingly recorded in the form of electronic health records (EHRs).Automated information extraction methods such as natural language processing (NLP) offer the opportunity to quickly extract and analyse large volumes of clinical data from EHRs.We sought to characterise the range of presenting symptoms in a large sample of patients with psychotic disorders using NLP.Methods: Dataset: South London and Maudsley NHS Trust (SLaM) Biomedical Research Centre (BRC) Case Register comprising pseudonymised EHRs of over 270,000 people.Clinical sample: 18,761 patients with an ICD-10 diagnosis of a psychotic disorders (F20, F25 or F31) and a control group of 57,999 patients with a non-psychotic disorder diagnosis (mood/affective/personality disorders without psychotic symptoms).Data collection: The NLP software package TextHunter was used.All sentences containing keywords relevant to the following symptom categories were analysed using a support vector machine learning (SVM) approach: positive symptoms, negative symptoms, disorganisation, mania and catatonia.Data on 46 symptoms were obtained with 37,211 instances annotated to contribute training and gold standard data for machine learning.2,950 instances were independently annotated to determine inter-annotator agreement.
Outcomes: prevalence of psychotic symptoms and their association with ICD-10 diagnosis.Results: A good degree of inter-annotator agreement was achieved (Cohen's κ: 0.83).Machine learning NLP achieved a mean precision (positive predictive value) of 83% and recall (sensitivity) of 78%.Among patients with psychotic disorders, the most frequently documented symptoms were paranoia, disturbed sleep and hallucinations.Psychotic symptoms were not limited to patients with an ICD-10 diagnosis of a psychotic disorder and were also present in the control group.Discussion: We found that psychotic symptoms were not limited to patients with a specific ICD-10 diagnosis and were present in a wide range of ICD-10 disorders.These findings highlight the utility of detailed NLP-derived symptom data to better characterise psychotic disorders.

T102. AN INVESTIGATION OF SCHIZOPHRENIA-BIPOLAR SUBGROUPS WITH GENETIC AND PROGNOSTIC VALIDATION
Dominic Dwyer* ,1 , Janos Kalman 1 , Monika Budde 1 , Urs Heilbronner 1 , Anne Ruef 1 , Heike Anderson-Schmidt 2 , Katrin Gade 4 , Nikola Mueller 5 , Ivan Kondofersky 3 , Sergi Papiol 1 , Peter Falkai 1 , Thomas G. Schulze 2 , Nikolaos Koutsouleris 1 1 Ludwig Maximilian University; 2 University Medical Centre Göttingen; 3 Institute of Computational Biology Background: Separation of individuals into schizophrenia and bipolar diagnoses has long been questioned, with some suggesting that the classification impairs the understanding of etiology, the accuracy of prognoses, and treatment selection.In this study, we employed unbiased statistical techniques to identify subgroups of individuals with chronic illness using a large array of variables commonly evaluated at the bedside.We then validated the resulting groups by investigating age of onset, schizophrenia polygenic risk scores (PRS), and functional outcomes at a 1-year follow-up period.Our hypothesis was that transdiagnostic subgroups would be stratified based on illness onset whereby individuals with earlier onset would have higher genetic risk loading and poorer functional outcomes.Methods: Participants were selected from a longitudinal, naturalistic, multisite project (PsyCourse) designed to investigate psychiatric illness course and outcomes.A total of 329 participants (age(SD)=45.7(12.6);54% female; years of illness duration(SD) = 13.7(10.3))with a DSM-IV diagnosis of schizophrenia, schizoaffective disorder, or bipolar disorder were assessed from 17 centers at baseline and 1-year follow-up periods.A clinical battery measuring sociodemographic, illness history, symptoms, cognition, and personality questionnaires (199 variables) was used to subgroup individuals.A non-negative factor analytic consensus clustering MATLAB toolbox was created based on previous methodological work in oncology.PRS were generated using widely used strategies, and differences between resulting subgroups were investigated with MANCOVA controlling for ancestry effects.Differences in functional outcomes were investigated with repeated measures ANOVA.Results: A 4-subgroup solution was robustly defined as the optimal solution using resampling techniques and cluster validity indices.Diagnoses were mixed in two subgroups, but predominantly bipolar or schizophrenia in the other two.All subgroups had equal illness durations (p>0.05),but the age of onset showed a decreasing trend with the earliest age being linked to two subgroups: a mixed bipolar-schizophrenia group with intermediate levels of general functioning and in a schizophrenia group with low levels of functioning (p<0.001).PRS scores were significantly increased in the early-onset, mixed bipolar-schizophrenia subgroup (p=0.007,uncorrected) and in the schizophrenia group (p=0.025,uncorrected).Prognoses differed between the four groups (p=0.003), with the greatest increases in functional outcomes in a late-onset mixed diagnostic subgroup (p=0.006) and in the schizophrenia group (p=0.002).Discussion: Four subgroups were detected and our hypothesis was supported by a relationship between earlier illness onset and higher schizophrenia genetic risk loading.While one of the subgroups with an earlier onset mostly consisted of individuals with schizophrenia, the other subgroup was diagnostically mixed.Our results tentatively suggest that transdiagnostic clustering may identify subgroups that could be effectively used to understand etiology and prognoses.Future research will investigate the possibility of differential treatment effects in these subgroups.

T103. ODIP (OUTIL DE DIAGNOSTIC INFORMATISÉ DES PSYCHOSES / PSYCHOSIS COMPUTERIZED DIAGNOSTIC TOOL): A NEW, SIMPLE METHOD FOR GENERATING DSM DIAGNOSES FOR PSYCHOTIC DISORDERS
Jean-Romain Richard* ,1 , Baptiste Pignon 1 , Franck Schurhoff 1 , Andrei Szoke 1 1 INSERM Background: OPCRIT was designed as a powerful tool to diagnose psychotic and affective psychoses.It has been frequently used in international psychiatric research.However, with 90 items it is time-consuming to complete and the diagnoses provided include many which are no longer used.Furthermore, this application is no updated for certain operating systems or psychiatric classifications.For these reasons, we have developed, a similar but much simpler tool focused on DSM classification of affective and non-affective psychoses.Methods: ODIP is based on the DSM-IV psychotic disorders classification, focusing on psychotic disorders (affective and non-affective).We identified 13 criteria that allow for the distinction between affective disorders with psychotic features (Bipolar or Depressive episode), schizophrenia, schizophreniform, schizoaffective, delusional, brief or non-specified psychotic disorders.We also designed a form to collect data on these 13 items.To assess how ODIP performs we tested it against the more complete OPCRIT and discordances in diagnosis were compared with the clinical diagnosis or, in a subsample of patients, with a research diagnosis.This was done in a total sample of 464 patients with a first episode of psychosis.First, we observed that only 34 out of 90 OPCRIT items are required to obtain a coherent DSM-IV diagnosis and that we could complete the items automatically using an algorithm based on the ODIP form.