Objectives: It is recognised that the clinical presentation of people in the post-acute phase of recovery after acquired brain injury is very heterogeneous (Arnould et al., 2016; Turner-Stokes, 2008).
Classifications on the basis of medical diagnosis of the brain injury do not fully address the problem, as there is heterogeneity even within the same diagnosis (e. g. Saatman et al., 2008). In addition, medical
diagnoses do not provide a clear understanding of the impairments, rehabilitation needs and “rehabilitation potential” of patients in the post-acute stage of recovery, which has led to the suggestion that various profiles of symptoms should be considered in order to match interventions and service levels to needs (Arnould et al. 2016; Arnould et al., 2015). The aim of this study was to explore whether a data-driven approach (cluster analysis) could be used to derive clinically meaningful subgroups of patients based on profiles of rehabilitation need.
Methods: Global outcome data gathered across all rehabilitation centres of the Brain Injury Rehabilitation Trust (BIRT) between 01 January 2016 and 30 June 2017 were used in the analyses. Data from a total of 300 cases were extracted. Scores on groups of items of the Mayo-Portland
Adaptability Inventory (MPAI-4) were entered in an iterative series of cluster analyses. Logistic regression was then used to identify areas of functioning that were the best predictors of cluster classification, and the outcomes on discharge within each cluster category were also investigated.
Results: The cluster analyses identified three profiles distinguishing individuals on the basis of their rehabilitation needs on admission. These groupings did not overlap with medical diagnoses or other demographic characteristics. Global cognitive and behavioural impairment alone did not reliably distinguish between these three profiles, as problems in these areas were prevalent in all three groups. We also found that all three groups benefited from rehabilitation, but the extent and nature of improvement differed. Results of the logistic regression analyses identified specific areas of function that distinguished between the groups and were thus shown to be related with outcome on discharge.
Conclusions: Determining the rehabilitation needs, key areas for targeting treatment, and likely outcomes of post-acute brain injury rehabilitation is challenging and using medical diagnoses as the basis for such predictions is uncertain, as it may result in underestimation of the level and complexity
of need. Using a data-driven approach, the present study led to the development of a framework for classifying rehabilitation needs, and predicting specific areas of outcome on discharge. We describe
how these findings can be applied to communicating with stakeholders and we suggest areas for further development in service delivery planning

Can cluster analysis help us better plan, communicate, and deliver brain injury rehabilitation?