Stratification and prediction of remission in first-episode psychosis patients OPTIMISE
Emanuela Martinuzzi,Susana Barbosa,Douglas Daoudlarian,Wafa Bel Haj Ali,Cyprien Gilet,Lionel Fillatre,Olfa Khalfallah,Réjane Troudet,Stéphane Jamain,Guillaume Fond,Iris Sommer,Stefan Leucht,Paola Dazzan,Philip McGuire,Celso Arango,Covadonga M. Diaz-Caneja,Wolfgang Fleischhacker,Dan Rujescu,Birte Glenthøj,Inge Winter,René Sylvain Kahn,Robert Yolken,Shon Lewis,Richard Drake,Laetitia Davidovic,Marion Leboyer,Nicolas Glaichenhaus
2019-01-01
Abstract:Early response to first-line antipsychotic treatments is strongly associated with positive long-term symptomatic and functional outcome in psychosis. Unfortunately, attempts to identify reliable predictors of treatment response in firstepisode psychosis (FEP) patients have not yet been successful. One reason for this could be that FEP patients are highly heterogeneous in terms of symptom expression and underlying disease biological mechanisms, thereby impeding the identification of one-size-fits-all predictors of treatment response. We have used a clustering approach to stratify 325 FEP patients into four clinical subtypes, termed C1A, C1B, C2A and C2B, based on their symptoms assessed using the Positive and Negative Syndrome Scale (PANSS) scale. Compared to C1B, C2A and C2B patients, those from the C1A subtype exhibited the most severe symptoms and were the most at risk of being non-remitters when treated with the second-generation antipsychotic drug amisulpride. Before treatment, C1A patients exhibited higher serum levels of several pro-inflammatory cytokines and inflammation-associated biomarkers therefore validating our stratification approach on external biological measures. Most importantly, in the C1A subtype, but not others, lower serum levels of interleukin (IL)-15, higher serum levels of C-X-C motif chemokine 12 (CXCL12), previous exposure to cytomegalovirus (CMV), use of recreational drugs and being younger were all associated with higher odds of being non-remitters 4 weeks after treatment. The predictive value of this model was good (mean area under the curve (AUC)= 0.73 ± 0.10), and its specificity and sensitivity were 45 ± 0.09% and 83 ± 0.03%, respectively. Further validation and replication of these results in clinical trials would pave the way for the development of a blood-based assisted clinical decision support system in psychosis.