The TBIcare project: Evidence based Diagnostic and Treatment Planning Solution for Traumatic Brain Injuries

By: 

Olli Tenovuo MD, PhD, Turku University Central Hospital, Finland; David Menon2 MD, PhD, Addenbrooke’s Hospital, Cambridge, UK

Despite the vast impact on public health and economy, traumatic brain injury (TBI) has remained without many core elements that form the cornerstones of modern medicine. To name a few of the most important of these, the clinicians taking care of subjects with TBI lack:

  • reliable means to exclude the presence of an acute TBI
  • reliable measures to assess the severity of the injury
  • specific medications to treat critical pathophysiological processes within the brain
  • firm scientific evidence for practically all therapies that are commonly used

Indeed, almost all of our methods, ranging from diagnostic procedures to the various treatments used, are based on so-called observational medicine. In other words, through trial and error, the clinicians treating these patients have concluded what seem to be the best methods to assess and treat these patients. Of course, scientific research has produced evidence to favour some measures and avoid others, but hard evidence is almost non-existing. Especially frustrating in the history of TBI medicine has been the failure of all of the thirty or so randomized controlled trials (RCTs) that have sought new medications for acute TBI [1]. As a consequence, pharmaceutical companies abandoned the field for ten years, and are still wary about returning to evaluate new therapies.

Although there are also other reasons operating, the most important one leading to the current state is the vast complexity of TBI as “our most complex disease in our most complex organ”. As a concrete example, most RCTs have recruited severe TBIs, which in terms of macroscopic pathology, may include contusions, subdural bleeds, epidural bleeds, traumatic subarachnoid haemorrhage, diffuse oedema, diffuse axonal injury, or any combination of these. Thus, we may have 64 different types of injuries at a macroscopic level, not counting the mechanisms of cellular pathology, which are undoubtedly at least as variable.

The purpose of the TBIcare project is to find new solutions to tackle this problematic complexity, which causes our common methods, such as RCTs, to be ineffective or impossible to carry out in a sufficiently large scale.

Idea of the TBIcare project

In principle, the TBIcare project tries to combine modern statistical methodology, data mining, and system simulation modelling in analysing large clinical TBI databases to produce methodology that enables the clinician to make individually best choices in treating patients with TBI.

The project has two scientific objectives; development of

  1. a methodology for finding efficient combinations of multi-modal biomarkers in statistical models to objectively diagnose and assess an individual TBI patient, and
  2. a simulation model based for objectively predicting outcome of the planned treatment of an individual TBI patient.

Methodology of the project

The first objective is addressed by using an approach in which a high number of biomarkers, relevant to TBI, are explored from sets of heterogeneous data. In this connection the word ‘biomarker’ is used in its widest sense and includes, for example, demographic and injury details, structural and functional changes visible in imaging data (computerised tomography, CT; magnetic resonance imaging, MRI; positron emission imaging, PET), changes in neurophysiology (electroencephalography, EEG); changes in bedside multimodality monitoring parameters including systemic circulatory and respiratory physiology, intracranial pressure (ICP), and brain chemistry (monitored by oxygen sensors and microdialysis); and changes in metabolomics or proteomics detectable in the blood. Figure 1 depicts the structure of the project.

Figure 1. Structure of the TBIcare project.

We define sets of biomarkers from several thousands of brain injury cases retrospectively, and from several hundreds of TBI cases and healthy controls prospectively. The retrospective datasets include large existing clinical databases from the participating hospitals in Turku, Finland and Cambridge, UK. In addition, the contents of the IMPACT database will be utilized – this database contains information of tens of thousands of subjects that have participated in international RCTs [2]. As a prospective dataset, the aforementioned participating hospitals will collect extremely detailed information from 400 subjects with TBI and 100 controls during the project. The goal is to build statistical models allowing standardised and objective interpretation of data from a single patient. The diagnostic rules will be derived by comparing individual patient data to corresponding cases in the database, by using statistical inference.

Work towards reaching the second objective uses these statistical models as basis for the construction of a simulation model. Due to the unique responses to treatments, the simulation model must be individualised. The model is personalized for each patient separately using data only from similar cases. Various approaches can be used for the simulations, such as, concepts from system dynamics or Bayesian networks. In the TBIcare concept individual physiological measures and various treatments form the building blocks of the system dynamics model, which is used to predict the outcome.

A tool for clinical work

In practical terms, this project tries to develop a software tool for clinical practice whereby the clinician may make as accurate assessment of the injury as possible. Currently, we know about one hundred variables that are known or suspected to affect the outcome of TBI. This huge amount of variability means that even an experienced clinician is unable to make accurate assessments of an individual injury. The modelling process does not only define the most important variables that are expected to contribute to outcome in a particular case, but also indicates whether further diagnostic measures that could improve prognostic reliability or therapy choice for a particular patient. Such interactions between the model and clinical data can also be used to validate and refine the model.

Although we are able to analyse a considerable number of subjects from the existing databases, it is apparent that during this three-year project we are probably able to create basic models only for the most common injuries and combinations of variables. Therefore, one task of this project is to create a basic modelling system, where the accuracy of the tool can be continuously improved by prospectively increasing the number of patients in the modelling database. This is especially important in developing the treatment modelling, where a higher degree of reliability is required.

These scientific objectives are supplemented by realization of technical objectives: a software solution to be used in daily practice to diagnose and plan treatments; new approaches for extracting information from multi-source and multi-scale physiological databases for management of an extremely heterogeneous disease; and innovative data quantification methods for the clinical TBI environment. Thus, TBIcare transfers the scientific concepts of the funding EU-program “Virtual Physiological Human” to clinical practice.

Future prospects

Evidence-based medicine has become a standard requirement in modern clinical practice, partly because of the constantly increasing costs of health care. However, clinicians treating individual patients are often faced with more complicated cases than those included in RCTs that underpin evidence-based recommendations. It is difficult to find a more challenging target for evidence-based medicine than to create recommendations for a condition as complex and variable as TBI. We feel that the one important approach this goal may be to bypass traditional RCTs, and supplement clinical decision making using a system that provides evidence to support clinical care in individual cases, practically irrespective of their individual profile.  However, we need to recognise that the fidelity of the model will be relatively better for patients who display common combinations of commonly encountered disease characteristics, and will become increasingly less dependable as the individual disease variables contributing to a given patient’s clinical picture become less common, or occur in uncommon combinations.

The requirement of cost-efficiency is also increasingly put for decision-makers in health care. Our modelling system does not only help in clinical decision-making, but it also helps to evaluate the true benefit and contribution of various diagnostic measures and treatment options, thus helping to focus efforts on methods that can yield maximal impact in individual cases. Obviously, the more sophisticated the data available, the more accurate the modelling approaches that can be applied. However, such modelling is not restricted to the complex multimodal data that are exclusively collected in specialized units, but can also provide recommendations based on simple clinical variables. This kind of modelling and prospective data collection is also a fruitful environment to do comparative effectiveness research [3], where the different outcomes and factors behind them can easily be analysed.

In our vision the future of TBI medicine may lean on specialized and audited TBI units which provide reliable and validated data for the modelling system, thus continuously improving its accuracy and reliability. As a very common occurrence, TBIs will also in future be treated at several levels of health care, and the modelling system may serve these “lower” units by providing diagnostic and treatment suggestions based on the information provided to the model, through a web-based system. This vision is described in figure 2.

Figure 2. A future vision for organizing continuous development and utilization of the diagnostic and treatment modelling.

Conclusions

A major challenge for the clinician treating individual patients with TBI is to form an accurate picture of the type, severity and prognosis of the injury, and to decide what would be the best way to treat a particular patient. The TBIcare project tries to approach these challenges in creating a tool that helps the treating clinician find the probably best answers and solutions. This goal cannot be achieved during this three-year project, but we aim to provide a core modelling resource that acts as a basis for further development and validation.

Acknowledgements

This project is partially funded by the European Commission under the 7th Framework Programme (FP7-270259-TBIcare). It is co-ordinated by VTT Technical Research Centre of Finland and the consortium includes GE Healthcare Ltd. (UK), Turku University Central Hospital (Finland), University of Cambridge (UK), Imperial College London (UK), Complexio S.a.r.L. (France), Kaunas University of Technology (Lithuania), and GE Healthcare Finland Oy.

The TBIcare webpages can be found at www.tbicare.eu.

 

References

  • Maas AI, Roozenbeek B, Manley GT. Clinical trials in traumatic brain injury: past experience and current developments. Neurotherapeutics. 2010;7:115-26.
  • Marmarou A, Lu J, Butcher I, McHugh GS, Mushkudiani NA, Murray GD, Steyerberg EW, Maas AI. IMPACT database of traumatic brain injury: design and description. J Neurotrauma. 2007;24:239-50.
  • Maas AI, Menon DK, Lingsma HF, Pineda JA, Sandel ME, Manley GT. Re-orientation of Clinical Research in Traumatic Brain Injury: Report of an International Workshop on Comparative Effectiveness Research. J Neurotrauma. 2011 Aug 29. [Epub ahead of print]

 

Corresponding author

Olli Tenovuo
Finnish Brain Injury Research and Development
Hospital District of Southwest Finland
Kiinamyllynkatu 4-8, 20520 Turku
Finland
tel. +358 50 4383802
fax. +358 31 32737