Interpreting research findings to guide treatment in practiceBMJ 2008; 337 doi: http://dx.doi.org/10.1136/bmj.a1499 (Published 24 October 2008) Cite this as: BMJ 2008;337:a1499
- Tony Kendrick, professor of primary medical care1,
- Kelsey Hegarty, associate professor2,
- Paul Glasziou, director3
- 1University of Southampton, Primary Medical Care, Aldermoor Health Centre, Southampton SO16 5ST
- 2University of Melbourne, Department of General Practice, Carlton 3053, Australia
- 3Centre for Evidence-Based Medicine, University of Oxford, Department of Primary Health Care, Oxford OX3 7LF
- Correspondence to: T Kendrick:
- Accepted 30 March 2008
Trials are important but not sufficient for good clinical decision making. Recommendations derived from trials in groups of patients must be interpreted and adapted by clinicians to the context of each individual patient seen in practice.1 The spectrum of patients in primary care is often very different from that in secondary care and clinical trials. In general, practitioners have two options: to consider how the treatment’s benefits and harms will differ given the severity, risk, and context of the individual patient, or to use a “try it and see” approach (or, more formally, do an “n of 1” trial). There is a range between these options. Take the example of the case of depression described in box 1.
Box 1 A patient with depression after marriage break up
A 54 year old man presents with low mood after separating from his wife of 30 years. He is usually seen only twice a year for review of his epilepsy drugs, tends to play down his symptoms, and has no history of depression. He has been unable to work for three weeks owing to reduced sleep, energy, appetite, and lack of motivation. He is now living alone in bed and breakfast accommodation and drinking more alcohol, is having thoughts about suicide but has no specific plans, and is not keen on antidepressants as he believes they can be addictive. He scores 14 out of 27 on the patient health questionnaire (PHQ-9), indicating mild major depression.2 The GP discusses the pros and cons of antidepressants, explores with the patient what strategies he will use to address his situation, and certifies him unable to work.
A week later, with his wife now taking steps to divorce him, he says he feels more down, and his PHQ-9 score is 18, indicating moderate to severe major depression. However, a friend has given him a room to stay in for a few months; he has stopped drinking and now wants to try antidepressants so he can get back to work.
Do depression guidelines help clinicians for this individual case?
Current guidelines in Australia and New Zealand, Canada, the United Kingdom, and the United States all recommend similar treatments for major depression or dysthymia (chronic mild depression) diagnosed according to psychiatric classification systems (DSM-IV or ICD-10).3 4 5 6 The two management strategies they most strongly support—antidepressants or cognitive behavioural therapy (CBT)—have been evaluated mainly among patients with major depression, mostly in secondary or tertiary care settings.
Guidelines from NICE (the UK’s National Institute for Health and Clinical Excellence) on antidepressant treatment were based on a review that included 31 outpatient studies, three inpatient studies, and only one study from primary care.5 A subsequent systematic review of primary care based trials identified 12 studies, but most were small, commercially funded studies of short duration and low methodological quality.7 8 Little of the evidence about antidepressants is derived from the type of depression seen more commonly in primary care, which is less severe than major depression and less chronic than dysthymia. The most recent systematic review of 35 published and unpublished trials from the US Food and Drug Administration database included only one trial in mild depression.9 Most trials have excluded patients with considerable alcohol use or comorbid physical conditions such as epilepsy that could be adversely affected by antidepressants.10 The treatment recommendations also fail to acknowledge patients’ reluctance to take drugs and do not address whether treatment works in the face of the social adversity usually associated with depression in primary care.11 12 More primary care based evidence exists to support the use of counselling and cognitive behavioural therapy, but only for patients with depression of a severity comparable to that of patients seen in outpatient settings.13 14 For milder depression, the guidelines mention the need to deal with social issues and suggest patient education and self help strategies based on principles of cognitive behavioural therapy, but the evidence for these interventions in primary care settings is limited.15
How might benefits and harms in clinical settings differ from the research setting?
The figure ⇓ provides a model to help think through the issues of transferability and application of research findings.1 In general, benefits will be higher in patients with higher risk or more severe illnesses.1 9 The top line shows the likely increase in benefit if the trial’s results apply across the spectrum of risk or severity of the condition. However, the relative effectiveness in primary care may be less (lower line) for several reasons—for example, because compared with everyday practice, trials may use tighter diagnostic rules, ensure higher levels of adherence to treatment, and fail to follow up patients with a poorer outcome. As trials often include a higher proportion of patients with severe disease or at high risk, the reported net benefit will be higher than in actual practice. This difference will be even greater for primary care, where most patients are likely to be at lower severity or risk. However, a primary care patient who is similar to the patients in the trial (the patient represented by the open circle near the right in figure 1⇓), should receive similar benefit. Every clinician must answer the following question: “For this patient does the severity or risk reach the threshold where benefits outweigh harms such as adverse effects and inconvenience?”
How should we estimate an individual patient’s likely response to treatment?
Let us apply this approach to the depressed man in box 1. Box 2 shows some of the factors to be considered.16
Box 2 Factors to take into account in the individual case: the PICO approach16
Population—is the individual in front of you sufficiently similar to trial participants to be likely to gain a similar relative benefit from treatment?
Is the patient diagnostically similar to patients in the trials? (Take into account the limitations of secondary care diagnostic tests in the primary care setting)
Are the likely benefits and harms of treatment similar to those in the trials?
Intervention—how similar will the treatment be to that given in the trial?
Is sufficiently similar treatment available and accessible?
Will the patient adhere to it? (Have you explored the patient’s preferences?)
Context or comparator treatments— are there specific contextual issues for the individual that are very different from the trial context and likely to influence the outcome for the patient? What are the alternatives to the specific treatment being considered?
Does the patient have complications or co-morbid conditions which would affect the likely benefits or harms?
Are there other prognostic factors which were not measured in the trials?
Is the patient likely to improve without treatment anyway?
Outcome—are the outcomes assessed in the trials, the same outcomes that are important for this individual?
Have you established what is important to the patient?
In the UK the commonest immediate decision is whether to prescribe an antidepressant, as cognitive behavioural therapy is usually not available or has a waiting list of several months. Since 2006, UK general practitioners have been remunerated for using questionnaire measures of the severity of depression at the outset of treatment, and the patient in box 1 scores just above the threshold for major depression, suggesting that antidepressants may be of benefit. However, the doctor is aware that the positive predictive value of the PHQ-9 for major depression is likely to be lower in primary care than in secondary care due to the relatively lower prevalence, and therefore the patient may be a false positive case given his borderline score.
So the doctor does not prescribe antidepressants on the basis of one marginally raised PHQ-9 score alone, especially as the patient is not keen on taking them and may be at risk of taking an overdose while acutely distressed. Also antidepressants may lower the threshold for epileptic convulsions. Instead the doctor discusses with the patient the possible benefits and side effects of antidepressants and arranges a review a week later. The patient clearly needs follow-up, as he tends to play down his symptoms, his functioning is impaired, and—being an older, socially isolated male with increased alcohol use—he is at increased risk of suicide (although the doctor is aware that having all these epidemiological risk factors is still a poor predictor of actual suicide in any individual case).
At review the symptom severity score is higher, indicating moderate to severe major depression. The patient’s preference is now for antidepressants and he is less socially isolated, reducing the suicide risk. He is now seen to be more like the patients included in trials of antidepressants, and the balance of the decision tips towards prescribing antidepressants. Caution is needed however, given the possible adverse effect of precipitating epileptic convulsions, with possible risks to the patient’s driving licence and job, and it would be reasonable to monitor the patient for longer, as his social situation is in flux and changes over the next few weeks may lead to considerable improvement.12 Prescribing can also be viewed negatively as a way of “disposing” of the patient which medicalises understandable reactions to life’s difficulties, precludes further in depth exploration of the person’s circumstances and own resources, and risks subsequent adverse effects on the doctor-patient relationship.12 17 If the doctor does decide to prescribe antidepressants, further discussion of the benefits of treatment, prompted at regular follow-up encounters, is likely to maximise the patient’s adherence to treatment.18 Encouraging questions, in a collaborative patient centred relationship, is likely to lead to a better outcome.19 Case management programmes have been shown repeatedly to ensure the best outcomes through such engagement, but at present are not widely available in primary care.20
The decision to apply research findings in practice depends not just on the setting but on the risk, severity, and context of the individual patient; the relationship between doctor and patient, and the availability of other resources to offer the patient as an alternative to medication. The transferability and applicability of secondary care evidence to primary care will vary between disorders. For depression, the spectrum seems quite different, making it difficult in many cases to base treatment decisions on evidence from trials. In many cases, the default position should be not to treat, at least initially, to avoid doing the patient harm, while watching for the development of more severe depression. The same principles could be applied to other common conditions such as migraine, for example. Yet in other conditions, the trial evidence can be applied directly. For example, for the use of statins after myocardial infarction the trial participants may have been recruited in hospitals but are actually the set of patients discharged to primary care for treatment, and hence the guideline recommendations are much more readily applicable.
N of 1 trials
For some conditions we can use the “n of 1 trial” approach to check whether a particular treatment works for a particular individual. For example, to determine whether a non-steroidal anti-inflammatory drug (NSAID) is better than paracetamol for arthritis, we could ask a patient to alternate paracetamol and the NSAID for a few days at a time over some weeks and record pain daily to measure the relative benefit. This would be most rigorous if similar preparations of each medication were used and the patient was unaware which was which—this would reduce the placebo effect, which would tend otherwise to favour the NSAID.16 Other examples of this “try it and see” approach, which will be familiar to general practitioners, include trying different creams on different parts of an apparently eczematous rash, or different eye drops in each eye for dry eyes.
Unfortunately, the n of 1 approach doesn’t work well in the case of depression, as antidepressants take some weeks to work, and a wash-out period is often needed between different types. Patients who have tried various treatments in the past can obviously tell us which seemed to work best for them. Otherwise we have to consider how well the evidence from groups of patients might apply in the individual case.
How can the evidence base be improved to be useful in individual cases?
To better inform treatment decisions we need more studies of the course of conditions without treatment, to identify predictors of who needs active treatment. We also need trials that include patients with mild as well as more severe conditions, and with comorbidities that might affect the relative benefits and harms of treatments. Patients’ preferences need to be taken into account in research designs, and outcomes important to patients need to be measured.
Studies will often need to be large, so that they have sufficient power for subgroup analyses to measure the effects of important predictors of response, including age, sex, and ethnic minority differences and variable adherence to treatment. Such studies would be better carried out in primary care, and clinicians should be prepared to take part in studies that will directly inform their practice, facilitating the negotiation of a truly informed decision between clinician and patient. However, some extrapolation from trial populations, whether they are from primary or secondary care, will always be necessary, as research studies can never include every possible type of patient presenting in practice.
Cite this as: BMJ 2008;337:a1499
Contributors: TK, KH, and PG researched the background to the article, and contributed to successive revisions of the text. TK wrote the first draft and is the guarantor.
Competing interests: None declared.
Provenance and peer review: Commissioned; externally peer reviewed.