Disentangling longitudinal dynamics in schizophrenia using longitudinal normative modelling
Poster No:
1863
Submission Type:
Abstract Submission
Authors:
Barbora Rehak Buckova1, charlotte fraza2, Antonín Škoch3, Marián Kolenič4, Christian Beckmann5, Filip Španiel6, Jaroslav Hlinka7, Andre Marquand8
Institutions:
1Radboud UMC, Nijmegen, Netherlands, 2Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Netherlands, 3National Institute of Mental Healt of the Czech republic, Klecany, Czech Republic, 4National Institute of Mental Health of the Czech Republic, Klecany, Czech Republic, 5Donders Institute for Brain, Cognition, and Behavior, Radboud University Nijmegen, Nijmegen, NL, Nijmegen, Netherlands, 6National Institue of Mental Health of the Czech Republic, Klecany, Czech Republic, 7Institute of Computer Science of the Czech Academy of Sciences, Prague, Czech Republic, 8Radboud University Nijmegen, Nijmegen, Gelderland
First Author:
Co-Author(s):
Christian Beckmann
Donders Institute for Brain, Cognition, and Behavior, Radboud University Nijmegen, Nijmegen, NL
Nijmegen, Netherlands
Donders Institute for Brain, Cognition, and Behavior, Radboud University Nijmegen, Nijmegen, NL
Nijmegen, Netherlands
Jaroslav Hlinka
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Introduction:
Despite the raising interest in the analysis of longitudinal dynamics in healthy brain development as well as the course of neuronal and psychiatric diseases, there has not been a substantial progress in the development of methods designed for longitudinal data analysis. We decided to address this issue by introducing a dynamical element into the state-of-the-art normative models creating the first of its kind framework for longitudinal analysis with normative models. We showcase this method on the analysis of two longitudinal samples of schizophrenia subjects.
We chose the schizophrenia sample as despite the intensive research into this area, we are still lacking reliable objective neuroimaging biomarkers, which would enable effective outcome prediction and targeted treatment of this condition, in which over 25% of patients do not respond to traditional medication. Thus, an insight into the longitudinal changes is timely.
We chose the schizophrenia sample as despite the intensive research into this area, we are still lacking reliable objective neuroimaging biomarkers, which would enable effective outcome prediction and targeted treatment of this condition, in which over 25% of patients do not respond to traditional medication. Thus, an insight into the longitudinal changes is timely.
Methods:
We present a method of longitudinal normative modelling which uses the power of pre-trained cross-sectional normative models and combines it with an estimate of longitudinal change to introduce a dynamical element. Using freely available normative models pretrained on over 58,000 samples (Rutherford 2022) and a control cohort for estimating healthy change (Rehak Buckova 2023), we derive a metric known as the z-diff score. This score, the longitudinal analogue to a standard cross-sectionally computed z-score, quantifies the probability of the observed change being statistically significant.
We applied this method to two longitudinal datasets of schizophrenia subjects. The first dataset involved a one-year follow-up of 98 young adults with first episode psychosis from the National Institute of Mental Health in the Czech Republic (Rehak Buckova 2023) (Fig. 1A), and the other from the NUSDAST dataset (Wang 2013), featuring 48 adults with schizophrenia and a two-year follow-up (Fig. 1A).
We applied this method to two longitudinal datasets of schizophrenia subjects. The first dataset involved a one-year follow-up of 98 young adults with first episode psychosis from the National Institute of Mental Health in the Czech Republic (Rehak Buckova 2023) (Fig. 1A), and the other from the NUSDAST dataset (Wang 2013), featuring 48 adults with schizophrenia and a two-year follow-up (Fig. 1A).
Results:
Cross-sectionally, both datasets showed comparable patterns of decreased z-scores in cortical thickness across multiple grey matter regions. This effect was more pronounced in patients in the early stages of psychosis, aligning with recent studies indicating a lag in neurodevelopmental trajectories in schizophrenia followed by a period of normalisation (Fig. 1 B)(Sheffield 2018, Milan 2016).
Longitudinally, distinct dynamics emerged in each dataset. While the first episode psychosis sample exhibited higher cortical thickening compared to the general population, the NUSDAST cohort showed cortical thinning in time, although the changes were not statistically significant after correcting for multiple comparisons (Fig. 1C). It is also possible that the size and direction of the changes observed in NUSDAST sample might have been affected by the composition of controls, which were not matched to the patients in terms of age, making it more challenging to reliably estimate the changes. Specifically, adapting the cross-sectional normative models using healthy data with a young and narrow age range makes it difficult to get good estimates of healthy variation across the lifespan, which in turn impacts on your ability to detect disorder-related effects.
Longitudinally, distinct dynamics emerged in each dataset. While the first episode psychosis sample exhibited higher cortical thickening compared to the general population, the NUSDAST cohort showed cortical thinning in time, although the changes were not statistically significant after correcting for multiple comparisons (Fig. 1C). It is also possible that the size and direction of the changes observed in NUSDAST sample might have been affected by the composition of controls, which were not matched to the patients in terms of age, making it more challenging to reliably estimate the changes. Specifically, adapting the cross-sectional normative models using healthy data with a young and narrow age range makes it difficult to get good estimates of healthy variation across the lifespan, which in turn impacts on your ability to detect disorder-related effects.
Conclusions:
In conclusion, our framework for longitudinal analysis, addresses the critical gap in methods for studying longitudinal brain development. Applying this method to schizophrenia datasets revealed distinct longitudinal patterns in the changes of cortical thickness. The first episode psychosis sample showed increased thickness, while the NUSDAST cohort demonstrated thinning, though not statistically significant after correction for multiple comparisons. Our findings stress the importance of age-matched controls in for reliable longitudinal analyses and offer insights into neurodevelopmental processes in schizophrenia.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 2
Modeling and Analysis Methods:
Bayesian Modeling
Methods Development 1
Neuroinformatics and Data Sharing:
Workflows
Keywords:
Data analysis
Open-Source Software
Schizophrenia
Statistical Methods
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
Rehak Buckova, B. (2023), "Using normative models pre-trained on cross-sectional data to evaluate longitudinal changes in neuroimaging data." bioRxiv.
Rutherford, S. (2022), "Charting brain growth and aging at high spatial precision", elife, no. 11, pp. e72904.
Sheffield, J. M. (2018), "Cognitive deficits in psychotic disorders: a lifespan perspective", Neuropsychology review, vol. 28, pp. 509-533.
Wang, L. (2013), "Northwestern University schizophrenia data and software tool (NUSDAST)", Frontiers in neuroinformatics, vol. 7.