Can you predict who will benefit from rehabilitation? When I was the editor of Clinical Rehabilitation, I saw a steady stream of studies that attempted to predict who would benefit from a rehabilitation intervention. One term commonly used was responder analysis, which assumes that a person who responds (to rehabilitation) can be identified and then their characteristics used to select patients for treatment. Separately, managers and commissioners want selection criteria, to choose patients who will benefit or to exclude people who will not benefit. Rehabilitation teams always look for people ready for or suitable for rehabilitation. These ideas all assume that a patient’s chances of benefiting can be estimated, and indeed categorised (good prospect or poor prospect). Is this true? Is it even likely that it can ever be true? I will explore these questions.
A disabled patient who is involved with a rehabilitation team will have a better outcome. This statement is supported by much evidence, and applies widely (see here). The question this post considers is, “Are there any criteria that can be used to predict who will benefit from rehabilitation?” An alternative question might be, “Are there any criteria that predict who will not benefit from rehabilitation?”
In addition, within the overall package of rehabilitation, there are specific interventions that lead to benefits, usually in some more restricted area such as the degree of spasticity (e.g. botulinum toxin) or the extent of arm function after stroke (e.g. constraint induced movement therapy). Therefore a second similar but more restricted question arises, “Are there any criteria that predict who will, or who will not benefit from the specific intervention?”
These questions are not unique to rehabilitation. Identical questions apply to many services – intensive care units, drug and alcohol services, hospitals in general, and many specialist services – and also to many specific medical or surgical interventions, such as chemotherapeutic anti-cancer drugs, anti-hypertensive drugs, carotid endarterectomy etc.
For a few specific interventions, it is possible, on the basis of evidence, to identify patients who are unlikely to benefit. For example, patients who cannot actively extend their fingers are unlikely to benefit from constraint-induced motor therapy.
However, for the great majority of situations where evidence shows that something is beneficial to a group of patients, it is not possible to say with any certainty who will benefit. Sometimes, it is more likely that everyone may benefit to a greater or less extent, and the response is on a spectrum. In other circumstances, such as taking drugs to reduce blood pressure and thus prevent a stroke, the person will either have a stroke or will not, and many people will not have that benefit.
The evidence available
To understand the difficulty, one has to understand the evidence. Almost all evidence comes from randomised trials where two interventions are compared (see here), and one intervention is found to be better that the alternative. If the intervention that is associated with the less good outcome is assumed not to be doing harm, then one can assume that the other is associated with benefit.
When the data show a difference between the two groups, such as a better mean score, or a greater number achieving some category, or a faster achievement of an event, they you that the patients in one group have benefited. The data do not tell you who has benefited.
In most studies, one can assume that the change seen in each group has a statistically normal distribution. In other words, the difference seen is a shift in the distribution of outcomes towards a better outcome. It is unlikely that the patients in the experimental groups will fall into two populations, with an obvious and discernible population showing a much better outcome and the remaining patients showing the same spread of outcomes as patients in the control group.
Therefore, all that one can say is that, if a person has the experimental treatment, there is a probability that their outcome will be a bit better. For some people that ‘bit’ might be quite large, and in others, it might be quite small. One can say, from the design, that the difference is associated with and probably caused by the experimental treatment.
The second important fact is that, in many studies, there is a natural underlying change within the studied patients. Sometimes the study is in a recovering condition, such as early after stroke; sometimes it is in a deteriorating condition, such as motor neurone disease; and sometimes change is less predictable so that people may fluctuate with some getting better, some getting worse, and others changing little.
In that context, it is worth looking at the techniques used to try and predict ‘response’; these methods are supposed to identify people who benefit, so that their characteristics can be studied and compared with non-responders, allowing one to identify criteria to distinguish the two groups – responders and non-responders.
People who respond are usually identified by looking at the difference between baseline measurement and final follow-up measurement and selecting people who changed by more than a certain amount, or people who were in the top 20% (or some other arbitrary centile) of change. In other words, a ‘responder’ is someone with a greater change over time. (In deteriorating conditions, the reverse logic would apply – it would be patients with the least change.)
However these are not ‘those who responded most’, they are simply ‘those who changed most’. And, if there are any factors associated with change, all one is achieving is identifying prognostic factors for change. In all studies I have seen, the prognostic factors are either studied in only one group all of whom had the active intervention or, in trials, the factors are extracted from both study groups combined.
Logically, if a group study on patients selected in a certain way has shown benefit, then all patients fitting the same selection criteria should be treated with the intervention. After giving the treatment, one may decide that the intervention should be stopped or changed on the basis of no detectable change after a suitable duration of treatment. This approach is used for disease-modifying treatments in multiple sclerosis and many other diseases, such as leukaemia.
It is simply not possible, on the basis of evidence, to select patients who are more or less likely to benefit from rehabilitation in general, nor is it possible on the basis of evidence to select people for specific interventions. As with other medical treatments, often one has to start a treatment that has been proven effective, and then modify and adjust management in the light of success or failure.
(false positives and negatives)
There is a further logical and statistical difficulty that must never be forgotten, one that has gained prominence in this pandemic (lateral flow tests). In the unlikely event that some factor were identified to predict that someone would (or would not) respond to rehabilitation or to a particular rehabilitation intervention, one must realise the following.
First, even to be considered, a factor would need to be established in one large population (several hundred) and then verified on a second completely separate large population.
Second, one needs to know the diagnostic characteristics: how may testing positive will truly benefit; how many testing negative will truly not benefit; how many testing positive will actually not benefit; and how many testing negative would have benefited.
Third, these figures will only be valid if applied to a similar population; if the population has a different overall proportion of people who will or will not benefit, then the sensitivity and specificity will be different, and the predictive value will be different.
In the unlikely event that there were valid selection criteria, nevertheless, as I pointed out in 2003 (here), one is likely to select for rehabilitation many patients who would not benefit and, at the same time, to refuse rehabilitation to many patients who would benefit from rehabilitation. The article shows the calculations supporting this assertion, in Table one.
The solution has several parts. The first part is that there must be a sufficiently resourced and flexible expert rehabilitation service to ensure that everyone needing or wanting rehabilitation can be assessed for their needs, and given advice.
The second part is to allow the rehabilitation service flexibility to make professional judgement about access to parts of the service that are under-resourced. The third part is to ensure that there is a minimal service available to everyone needing it, if only to monitor change and change a patient’s priority for treatment if needed.
At present these solutions are not available in the UK, but they could be (see here).
In this solution, professional judgement is the selection mechanism. Professional judgement is obviously not perfect, and it is subject to many biases and other problems. But, until a better method is developed, it is probably the least bad option and least likely to make terrible errors. Moreover, the decisions should be taken by, or overseen by, a multi-disciplinary team which will reduce the risk of any one individual making biased or irrational decisions.
The last part of this solution is to accept that a ‘trial of rehabilitation’ (or a ‘trail of an intervention’) is a sensible, clinically appropriate policy. At the same time, the team must accept that a patient who is judged not to be benefiting sufficiently from their involvement must be transferred out of the service so that other patients may benefit. Unfortunately, this is often difficult, given the unsatisfactory nature of residential care placements and the support services for people with long-term care and support needs.
In summary, it is not possible on any evidential or rational basis to determine who will or will not benefit from rehabilitation, or from a specific intervention. Every patient should have an expert assessment which should identify their rehabilitation needs. These needs should be prioritised in the context of the available services, and as many of the needs as can be should be met. Whether or not any needs can be met, every person should be assessed and advised. If the rehabilitation service is organised on a District-wide basis (see here), then this approach may well ensure that most patients have many of their needs met. Using selection criteria will lead to lower quality service and a waste of resources by selecting patients inappropriately.