Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
The aim of evidence-based practice is to help clinicians to base their practice on best current evidence. The essential first step in the process is the identification of uncertainty, or the need for information, and the translation of this uncertainty into answerable clinical questions. Developing the ability to formulate precise, structured, and answerable clinical questions has been identified as one of the key tasks for clinicians using the strategies of evidence-based practice in their clinical work.1
The question needs to be:
Directly relevant to the identified problem
Constructed in a way that facilitates searching for a precise answer.
A 24 year old man has been admitted to an acute psychiatric ward complaining of hearing voices (3rd person auditory hallucinations) and persecutory ideas. He had a previous admission about six months ago with a similar clinical picture, but he left the hospital before a definitive diagnosis could be made. He also has a past history of substance abuse. He is unwilling to take haloperidol because of an acute dystonic reaction that occurred with this drug during his previous admission. He wants to leave the hospital, but has no accommodation and the clinical team have concerns about the content of his hallucinations.
Most clinicians generate several clinical questions when they are presented with this example. Some of these questions can only be answered by obtaining more information from the patient, for example does the patient have any depressive symptoms? Has the patient been taking psychoactive drugs recently? This sort of question is obviously important to identify the patient's problems, to understand them, and to help to put them into context.
However, many questions can only be answered by using research that has been done with similar groups of patients with similar problems. Examples of this kind of question include:
What is the likely diagnosis on the basis of the presenting symptoms?
What are the implications of the substance abuse for diagnosis, treatment, and prognosis?
Is there a treatment which would be less likely to cause extrapyramidal side effects than haloperidol?
Is there any way that the patient's concordance with treatment can be improved?
What is likely to happen if the patient leaves hospital?
Here, we will focus mainly on this second category of question.
As in the example, most patients generate at least some clinical uncertainty—perhaps several questions. It seems that clinicians both underestimate their information needs and have difficulty in getting satisfactory answers.2 Clinicians may find it difficult to acknowledge uncertainty, and developing the requisite skills is likely to take practice and some coaching.3 In practice, once we begin to recognise uncertainties and knowledge gaps, the number of questions seems to grow exponentially! When many questions arise, clinicians therefore need to prioritise their information needs. Choosing which question to answer first will depend on which question:
Is most important for the patient's wellbeing
Is likely to be feasible to answer within the time available
Is the most interesting
Is likely to arise frequently in practice.4
Structure and classification of clinical questions
Before looking for the answer, the question needs to be formulated into a precise, structured question. This involves breaking the question down into 3 (or 4 in the case of treatment questions) components:
The patient's problem: this will often be a diagnosis, but may be a non-diagnostic problem such as a lack of housing
The intervention of interest: in treatment questions this will be an intervention, in diagnostic questions it may be any aspect of the history or examination
The comparison intervention: in treatment questions
The outcomes of interest.
In the example, the first question is to do with diagnosis and would become:
If a patient with psychotic symptoms (and therefore a high pre-test probability of schizophrenia) (problem) clearly describes a hallucinatory voice talking in the third person (intervention), how likely is it that the illness will meet reliable criteria for schizophrenia when the assessment is complete (outcome)?
The third question is one of treatment and would become: In a patient with a working diagnosis of schizophrenia (problem) how well does an atypical antipsychotic such as olanzapine or risperidone (intervention), compared with haloperidol (comparison intervention), reduce psychotic symptoms without causing unacceptable extrapyramidal side effects (outcome)?
Most commonly occurring clinical questions are those of treatment (which treatment is most likely to increase the probability of a desired outcome?), diagnosis (what is the most helpful way of classifying a patient's problem?), and prognosis (how likely is it that specific outcomes will occur?). Table 1⇓ shows the general forms of these questions.
Architecture of research
Once the question has been structured and classified, this indicates the research design that is most likely to produce reliable answers (table 1⇑). Furthermore, a hierarchy of research design can be constructed for each kind of clinical question. Such a hierarchy is best established for treatment questions (table 2⇓), but the general principle is that a systematic review of all available high quality studies is likely to be the most reliable evidence for any kind of question. So for a treatment question, a valid systematic review of randomised controlled trials would come at the top of the hierarchy, and for a prognosis question it would be a systematic review of inception cohort studies.
Identifying the most reliable research design will also help efficient searching—a subject we will cover in a future EBMH note.
The structured clinical question is therefore a useful way of linking clinical uncertainty to research evidence. To facilitate this process in the evidence-based journals, we have replaced the objective section of the abstracts with the structured clinical question that the research was primarily designed to answer.3 As well as helping individual clinicians to find rapidly the best available evidence, an active attempt to identify areas of clinical uncertainty is likely to bring another benefit. When it is clear that a substantial number of clinicians have clinical uncertainty about a clinical question, even when the best available evidence has been identified, this question becomes a priority for further research. These research priorities can then be fed into the systems of health technology assessment described by Ken Stein and Ruiaridh Milne in this issue.