Towards a new neurological model of language that explains outcome after stroke.

Every five minutes one person in the UK has a stroke. More than a third of these patients will develop aphasia, which can impair the ability to speak, understand speech, read and/or write. Aphasia is the most feared outcome after stroke because it severely impairs quality of life and makes stroke patients prone to social isolation, dependence and depression. One of the most demoralizing outcomes for aphasic patients is the lack of any information on their likely prognosis for recovery. This is a consequence of there being no established way to predict when, how or whether language skills will recover, which in turn makes it difficult to decide which course of therapy is best suited to an individual patient. Our research aims to Predict Language Outcome and Recovery After Stroke (PLORAS, http://www.ucl.ac.uk/ploras). In pursuit of this goal, we generating a database that records as many details as possible about hundreds of different stroke patients.

Our initial investigations show that we can make confident predictions for language outcome for many patients. For example, we have identified 2 brain areas where the presence or absence of damage (caused by the stroke) very reliably predicts whether or not a patient will have "persistent" difficulty producing speech (i.e. years after stroke). These findings are important to share with new patients and their carers as they help to provide realistic expectations of likely recovery. Prognoses of good outcome are valuable because they substantially reduce anxiety about the future. Prognoses of poor outcome are more challenging for patients and their carers to deal with in the short term, but they can ease future frustration and depression associated with a failure to meet unrealistic targets.

The aim of the proposed research is to provide a scientific explanation of how the predictions are working. This comes in the form of what we refer to as a "neurological model" of language that predicts speech and language outcome after stroke by identifying which brain regions are critical to language, the functions they support (e.g. speech comprehension or reading) and how the regions connect to one another.

The current neurological model of language, illustrated in medical textbooks today, is based on crude post mortem data from less than a handful of aphasic patients tested at the end of the 19th Century by Paul Broca (1861) and Karl Wernicke (1874). Despite its enduring popularity, this model has not stood the test of time because it does not reliably predict outcome in new patients. Thus, although the classic neurological model is conceptually appealing, it does not make predictions that are accurate enough to be clinically useful. Conversely, our data-led analyses are starting to make very accurate predictions but we do not understand their scientific basis.

Our proposal aims to provide a new neurological model that can be used to explain and ultimately improve the predictions we make. The new neurological model will illustrate the set of brain regions that support different language abilities (e.g. the ability to speak or understand speech). These brain regions will correspond to the parts of the brain that are typically damaged in patients who have selective difficulty with one particular language skill. In order to find these areas, and demonstrate how consistently they affect other patients, we have developed a completely new method for analysing the exact location and extent of brain damage. Our studies also place great importance on how much time, since the stroke, the patient has had to recover. We also aim to investigate the influence of many other factors on a patient's recovery, including the amount of speech therapy they have been given.

We will investigate the combinations of brain areas that are critical to core language functions. In Step 1, for each function of interest, we select patients who each have difficulty performing all the tasks that are assumed to tap that function. For example, when the function of interest is speech production, we select patients who have difficulties with all speech production tasks (object naming, reading and fluency) but not on tasks that don't (picture-word-matching etc). The lesion site in each of the selected patients is defined in high-definition 3-dimensional space and becomes a region of interest (ROI) with the assumption that the ROI must include brain areas that contribute to all the neural pathways that have the potential to support the tasks and functions that the patient has difficulty with. Further analyses are then required to determine which parts of the ROI are critical or not, and whether the different parts contribute to the same neural pathway or a separate pathway (if there are different ways that the same task can be performed).

In Step 2, we select other patients who either have damage that includes all or a part of an ROI. The language abilities of each of these patients determine which parts of the ROI are critical and how the lesion-deficit relationship varies (i) across patients and (ii) with time post stroke. If different parts are from the same neural pathway, we expect that damage to each part will impair task performance. If different parts contribute to different neural pathways (for the same task) then we expect task performance to be worst when damage includes all relevant pathways than when it includes just one or a subset of pathways. Each result will be validated in new cohorts of patients who are closely matched for lesion site and followed up longitudinally. All results will be integrated in a new neurological model of language that can predict both initial symptom severity and the likelihood of recovery in new patients.

Our proposal fits with the NHS and the UK government's goals to improve quality of life and reduce unnecessary costs for people with long term impairments caused by diseases such as stroke. Our target is to provide (A) an extensive database of brain scans and language measures from stroke survivors tested at many different time points during the course of their recovery; (B) predictions of how a lesion will affect language and (C) a neurological model that explains the predictions. This will allow clinicians, new patients, and their carers, to find out whether, how and when they are likely to be affected and recover. Below, we emphasize the social and economic benefits that souch a resource could provide to stroke survivors, healthcare professions, NHS resources and the general public.

Stroke survivors: Patients want to know whether, when and how they can recover their language skills after stroke. The neurological model we are creating aims to address these questions confidently and precisely for as many patients as possible. In the short term, our predictions will only indicate how likely it is that a patient's language skills will return to normal or not in the first year or two after their stroke. This "likelihood" could range from confident (more than 95% likely) to unknown (50% likely). If we are able to provide confident prognoses of recovery, patients will be relieved to know that they should be able to return to normal life. If our predictions are confident that recovery will be limited, patients may initially become demoralised. However, realistic long term expectations can greatly reduce the emotional and economic burden on the patient and their family. For example, realistic expectations can reduce the frustration of wasting time, money and hope on treatments and therapies that are unlikely to work, while guiding patients to other therapeutic assistance that maximises their wellbeing through the most appropriate support. Most patients and their families therefore tell us that they want to know their likely outcome whatever the scenario. Having some information is better than not knowing. Therefore, patients for whom we are unable to make confident predictions (good or poor), will motivate further research.

Healthcare professionals: Physicians, speech and language therapists, neuro-radiologists and neurosurgeons will ultimately benefit from our research. Clinicians need to know what to advise patients and their families who are asking for details on their prognosis. Speech and language therapists need to select therapies that are best suited to the brain structures that are available to support recovery. Currently, therapies are selected according to the patients' symptoms, not the integrity of recovery pathways we are identifying. Nevertheless, it is clearly recognised that the success of a therapy varies greatly from patient to patient. The predictions we generate by considering brain structure should provide (a) testable hypotheses to indicate which therapies are most likely to work for an individual patient; (b) a means of stratifying patients when testing the efficacy of a new treatment and (c) baseline predictions from which to quantify how much a therapy has accelerated recovery relative to that expected regardless of any intervention.

NHS economic efficiency: Healthcare provisions need to maximise treatment and support for aphasic patients while minimizing costs. Selecting appropriately tailored therapies will facilitate recovery and reduce the inevitable cost of selecting the wrong treatment.

Public: Provision of quality medical information on the web empowers the general public to find answers to frequently asked questions. Our research has a growing online presence and, through the project website and social media, is raising awareness of the potential to predict language recovery after stroke.