| Objective:Cognitive symptoms and acute phase treatment response in schizophrenia are two critical clinical issues,both of which are associated with specific brain imaging features,and cognitive symptoms themselves may have a predictive role in acute phase treatment response.The purpose of this study was to investigate the prediction of acute phase treatment response in patients with schizophrenia,yet the psychometric structure of cognitive function in schizophrenia is still an unresolved issue.Therefore,in this study,we firstly investigated the statistical performance of different psychometric structures of cognitive function and their predictive ability for treatment response,and then we investigated the multimodal brain network characteristics of different psychometric structures.Finally,we explored the predictive abilities of multimodal brain networks combined with cognitive performance on treatment response.Methods:(1)Study I included 733 subjects from several projects,consisting of647 patients with schizophrenia in the acute phase and 86 healthy controls.All subjects completed the MCCB neurocognitive assessment,and patients with schizophrenia also completed the PANSS assessment.Categorical models were fitted to the patients’neurocognitive functioning using a latent profile model,and longitudinal stability was validated using a latent transition analysis.Dimensional models were fitted to patients’neurocognitive functioning using confirmatory factor analysis and validated for stability using longitudinal measurement invariance analysis.In addition,the association between baseline cognitive performance and PANSS reduction ratio was explored in patients who had PANSS assessed after acute phase treatment.For the optimal categorical model,differences in reduction ratio were compared across the categorical sample;for the optimal dimensional model,correlations between different cognitive factors and reduction ratio were calculated.(2)StudyⅡincluded 239 subjects from two centers,including 168patients with schizophrenia in the acute phase and 71 healthy controls.MCCB was assessed in all subjects,and magnetic resonance imaging(MRI)data were acquired in three modalities:functional MRI,structural MRI,and diffusion MRI.We constructed three brain networks,functional connectivity network(FCN),morphological connectivity network(MCN),and structural connectivity network(SCN),using MRI data from the three modalities,respectively.For the category model of cognition,a one-way ANOVA was used to compare differences in multimodal brain network connections across categories and healthy controls(NBS multiple comparison correction).For the dimensional model of cognition,multimodal brain network connections associated with the cognitive composite score and each cognitive dimension score were calculated using LASSO multivariate regression.(3)In StudyⅢ,MRI data were collected at baseline for 69 patients with schizophrenia from two centers in three modalities,and PANSS was assessed at baseline and after acute phase treatment.correlations between multimodal brain network connections and overall treatment response,as well as correlations with treatment response for different symptom factors,were calculated using LASSO multivariate regression.Finally,the individualized predictive performance of this correlation was examined using SVR analysis and tested whether combining baseline cognitive test scores improved predictive performance.Results:(1)Latent profile analysis supported the three-category model(full sample BIC=39132.637,ssa BIC=39024.688,LMRTp<0.0001,BLRTp<0.0001),which allowed the classification of patients with acute phase schizophrenia into three different categories based on cognitive performance:severe impairment group,moderate impairment group and mild impairment group.While confirmatory factor analysis showed that the cognitive performance of patients with acute phase schizophrenia was best with the Lam three-factor model(χ~2=71.809,TLI=0.924,CFI=0.954,RMSEA=0.071,SRMR=0.037,BIC=39009.628 for the whole sample),i.e.,processing speed/attention factor,executive function factor,and learning/memory factors.The analysis also showed that both models were longitudinally stable.For the three different categories of patients,there were differences in the reduction ratio of their PANSS scale total score,positive symptom factor and general symptom factor(F=3.704,p=0.026;F=3.784,p=0.024;F=4.936,p=0.008,respectively),mainly between the severe impairment group and the mild impairment group.For the different cognitive dimensions,after correction for FDR,the processing speed/attention factor was significantly correlated with the reduction ratio in PANSS total score,positive symptoms factor and general symptoms factor(r=0.156,p=0.029;r=0.149,p=0.029;r=0.159,p=0.029,respectively),the learning/memory factor was significantly correlated with the reduction ratio in PANSS total score(r=0.130,p=0.050),and the neurocognitive composite score was also significantly correlated with the reduction ratio in PANSS total score,positive symptom factor and general symptom factor(r=0.148,p=0.029;r=0.143,p=0.031;r=0.151,p=0.029,respectively).(2)Two subnetworks of the FCN differed between cognitive categories and healthy controls between groups,with the first subnetwork dominated by the left superior parietal gyrus and bilateral middle temporal gyrus regions(p=0.0092,NBS corrected)and the second subnetwork dominated by bilateral temporal pole and parahippocampal gyrus regions(p=0.0074,NBS corrected).Post hoc analysis of both subnetworks revealed a significant decrease in connection strength mainly in the severe and moderate injury groups compared to healthy controls.LASSO regression showed that connections associated with the neurocognitive composite score were mainly present in the MCN(r=0.645,p<0.001;r=-0.547,p<0.001 for positive and negative correlated connections,respectively),with positive correlated connections involving the frontal-striatal loop and negative correlated connections involving regions such as the nucleus accumbens and precuneus;connections were also associated in the FCN(r=0.416,p<0.001;r=-0.265,p<0.001 for positive and negative correlated connections,respectively).In addition,the brain network connections associated with each of the three cognitive factors were different.Connections related to the processing speed/attention factor were found in the FCN and MCN(r=0.332,p<0.001;r=-0.290,p<0.001 for FCN positive and negative correlated connections,respectively;r=0.694,p<0.001;r=-0.565,p<0.001 for MCN positive and negative correlated connections,respectively),and most of these connections were similar to those of the neurocognitive composite score connections of the representations were similar;The executive function factor was positively correlated with the bilateral temporal pole-frontal connections in the FCN(r=0.384,p<0.001)and negatively correlated with the connections centered in the precuneus and dorsal inferior frontal regions in the MCN(r=-0.591,p<0.001);the learning/memory factor was only correlated with the FCN(r=0.450,p<0.001;r=-0.298,p<0.001 for positive and negative correlated connections,respectively),with positive correlations involving the hippocampal-superior parietal gyrus-centered subnetwork and negative correlations involving mainly the rectus gyrus.(3)LASSO regression showed that the MCN was associated with the reduction ratio of PANSS total scores(r=0.620,p<0.001;r=-0.716,p<0.001,for positive and negative correlated connections,respectively),with positively correlated connections located in the frontal-temporal and negatively correlated connections in the occipito-frontal-parietal loops.We also found that specific connections in the SCN were able to predict responses in positive symptoms(r=0.300,p=0.012;r=-0.564,p<0.001 for positive and negative correlated connections,respectively)and specific connections in the FCN were able to predict responses in general symptoms(r=0.415,p=0.012;r=-0.450,p<0.001).Further SVR analysis showed that these connections associated with treatment response also showed better performance in predicting response in individuals.Specifically,11 MCN connections predicted reduction ratio of PANSS score(r=0.820,p<0.001,MSE=0.0063,p<0.001),22 SCN connections predicted reduction ratio of positive symptom factor(r=0.867,p<0.001,MSE=0.0114,p<0.001),and 23 FCNs predicted reduction ratio of general symptom factor(r=0.739,p<0.001,MSE=0.0121,p<0.001).Combining 8cognitive test scores only weakly improved the predictive performance of the SCN for reduction ratio of positive symptom factor(r=0.885,p<0.001,MSE=0.0105,p<0.001).Conclusion:(1)The latent structure of cognitive functioning in acute schizophrenia can be represented by a categorical and a dimensional view.The category model divides the patient population into three subgroups with different levels of impairment,while the dimensional model downscales eight neurocognitive tests into three factors.The cognitive structures of both models are reproducible and longitudinally stable,and are predictive of treatment response over the acute period.(2)Multimodal brain network connectivity can characterize cognitive function in acute phase schizophrenia.Functional networks within the default network and between the default and dorsal attention networks are important for maintaining normal cognitive function in patients;morphological networks in the frontal-striatal loop correlate with the overall level of neurocognitive function.In addition,different dimensions of cognitive function have their own specific brain network basis.(3)Morphological connectivity of the extensive cortex is associated with treatment response,especially the imbalanced pattern of frontal connections.In addition,different brain network types predict different aspects of treatment response.The predictive role of brain networks for treatment response can be generalized to the individual level. |
PDF Full Text Request