Distinct clinical phenotypes and their neuroanatomic correlates in chronic traumatic brain injury

Abstract

Accumulating evidence of heterogeneous long-term outcomes after traumatic brain injury (TBI) has challenged longstanding approaches to TBI outcome classification that are largely based on global functioning. A lack of studies with clinical and biomarker data from individuals living with chronic (>1 year post-injury) TBI has precluded refinement of long-term outcome classification ontology. Multimodal data in well-characterized TBI cohorts are required to understand the clinical phenotypes and biological underpinnings of persistent symptoms in the chronic phase of TBI. The present cross-sectional study leveraged data from 281 participants with chronic complicated mild-to-severe TBI in the Late Effects of Traumatic Brain Injury Study. Our primary objective was to develop and validate clinical phenotypes using data from 41 TBI measures spanning a comprehensive cognitive battery, motor testing and assessments of mood, health and functioning. We performed a 70/30% split of training (n = 195) and validation (n = 86) datasets and performed principal components analysis to reduce the dimensionality of data. We used Hierarchical Clustering on Principal Components with k-means consolidation to identify clusters, or phenotypes, with shared clinical features. Our secondary objective was to investigate differences in brain volume in seven cortical networks across clinical phenotypes in the subset of 168 participants with brain MRI data. We performed multivariable linear regression models adjusted for age, age-squared, sex, scanner, injury chronicity, injury severity and training/validation set. In the training/validation sets, we observed four phenotypes: (i) mixed cognitive and mood/behavioural deficits (11.8%; 15.1% in the training and validation sets, respectively); (ii) predominant cognitive deficits (20.5%; 23.3%); (iii) predominant mood/behavioural deficits (27.7%; 22.1%); and (iv) few deficits across domains (40%; 39.5%). The predominant cognitive deficit phenotype had lower cortical volumes in executive control, dorsal attention, limbic, default mode and visual networks, relative to the phenotype with few deficits. The predominant mood/behavioural deficit phenotype had lower volumes in dorsal attention, limbic and visual networks, compared to the phenotype with few deficits. Contrary to expectation, we did not detect differences in network-specific volumes between the phenotypes with mixed deficits versus few deficits. We identified four clinical phenotypes and their neuroanatomic correlates in a well-characterized cohort of individuals with chronic TBI. Phenotypes defined by symptom clusters, as opposed to global functioning, could inform clinical trial stratification. Individuals with predominant cognitive and mood/behavioural deficits had reduced cortical volumes in specific cortical networks, providing insights into sensitive, though not specific, candidate imaging biomarkers of clinical symptom phenotypes after chronic TBI and potential targets for intervention.

Keywords: machine learning; neuroimaging; phenotyping; traumatic brain injury.

Graphical Abstract

Figure 1

Study flow diagram. We provide details on the derivation of the phenotypic (primary objective) and neuroimaging analytic (secondary objective) samples. LETBI, Late Effects of TBI Study.

Figure 2

Lesion correction steps. The pipeline for our lesion correction method is as follows: (A) obtain manually traced lesion mask; (B) merge manually traced lesion mask with white matter segmentation volume; (C) re-run recon-all in FreeSurfer; and (D) re-apply manually traced lesion mask to whole brain segmentation (ASEG) volume and assign as lesion label (i.e. right lesion and left lesion). ASEG, Automated Segmentation (using FreeSurfer); LUT, Look-up Table (where each number represents a unique labelled structure in the segmentation atlas).

Figure 3

Hierarchical cluster group assignment (training set; n = 195). Hierarchical cluster group assignment in the training set. The results are based on the Hierarchical Clustering on Principal Components (HCPC) statistical methodology. Each dot in the figure corresponds to a unique participant. Their PC1 versus PC2 scores have been plotted in the x–y coordinate space, and their assigned cluster memberships are indicated in the legend. There are no traditional between-group statistical tests and P-values derived for the HCPC procedure.

Figure 4

Heat map characterizing average values of neurobehavioural measures by cluster assignment (training set; n = 195). Heat map characterizing average values of neurobehavioural measures by cluster assignment in the training set. The measures have all been transformed such that darker colours represent greater impairment, and lighter colours represent less impairment. Based on the findings, we have assigned the following qualitative descriptors of each cluster: Cluster 1: mixed deficits; Cluster 2: predominant cognitive deficits; Cluster 3: predominant mood and behavioural deficits; and Cluster 4: relatively few deficits. No between-group statistical tests and P-values were derived for the heat maps. CVLT, California Verbal Learning Test; COWAT, Controlled Oral Word Association Test; WAIS, Wechsler Adult Intelligence Scale; QOL, quality of life; SWLS, Satisfaction with Life; BIS, Barrett Impulsivity Scale; UPDRS, Unified Parkinson's Disease Rating Scale; ACT, Adult Changes in Thought; MIDUS, Midlife in the United States; ASSIST, Alcohol, Smoking and Substance Involvement Screening Test.

Figure 5

Dot plot of model-based estimated marginal mean cortical volume by cluster and cortical network. Each dot in this figure represents the model-based estimated marginal mean cortical volume (mm³) by cluster group (n = 36 mixed; n = 60 cognitive; n = 73 mood/behaviour, n = 112 fewest deficit). The estimated marginal mean is based on results from the linear mixed effects model. Specifically, the least-squares mean value at each level of the cluster, adjusting for model covariates (age, age-squared, sex, scanner type, injury severity, injury chronicity and training set). All volumes were standardized by network to have a mean of 0 and standard deviation of 1 for the sample. Therefore, values below 0 by cluster can be interpreted as below average volumes, and above 0 can be interpreted as above average in the sample. We included this figure as an illustration of the estimated marginal means by cluster for each network; the P-values for pairwise comparisons are presented in Supplementary Table 6.