Although there is an ever-increasing demand for standardisation and improved quality in psychiatric treatment, it is extremely difficult for busy psychiatrists to keep up-to-date with clinically important advances. About 5500 potentially clinically relevant scientific papers are published annually: this is equivalent to reading 15 papers per day, which is clearly unrealistic for most clinicians (Reference Geddes, Wilczynski and ReynoldsGeddes et al, 1999). This leads to a gap between research and practice, which is likely to be reflected by inappropriate variations in clinical practice.

EVIDENCE-BASED MEDICINE

Evidence-based medicine was developed as a set of strategies derived from developments in information technology and clinical epidemiology, to assist the clinician in keeping up-to-date with the best available evidence (Reference Geddes and HarrisonGeddes & Harrison, 1997). The aim of evidence-based medicine is to provide a set of tools to assist clinicians in decision-making. One of the key developments under-pinning the feasibility of evidence-based medicine has been the development of the methodology of reviewing. Systematic review of primary research studies can critically examine the evidence for both established and new interventions and should provide reliable information to inform treatment choices. The advantage of the reviews contained in the Cochrane Database of Systematic Reviews is that, on average, they are of a higher methodological standard than other reviews (Reference Jadad, Cook and JonesJadad et al, 1998) and that, being primarily electronic, they can be continually updated as new primary studies are completed or identified.

Systematic reviews can also identify areas of substantial clinical uncertainty in which there is a dearth of good-quality research (Reference GeddesGeddes, 1999). In the UK and elsewhere, mechanisms are being developed to feed back this clinical uncertainty to inform the clinical research agenda, to ensure that scarce research funds are targeted at the areas of highest priority (Reference Stein and MilneStein & Milne, 1999).

To make sure that the results of systematic reviews are introduced effectively into everyday clinical practice, it seems likely that it will be important to consider defined interventions concentrating on the problems raised in management by relatively coherent patient groups. Clinical epidemiology, which is one of the main methods of evidence-based medicine, has been called ‘a basic science for clinical medicine’ and can be seen to strengthen and enhance traditional medical approaches to the care of patients (Reference Sackett, Haynes, Guyatt and TugwellSackett et al, 1991). The introduction of evidence-based medicine is likely to lead to a renewed interest in accurate diagnosis and the efficient delivery of effective medical care for individual patients with defined problems. This approach is in contrast to the application of policies of service delivery based on ill-defined interventions for vaguely specified patient groups that have been fashionable recently, and have been described as a disaster for psychiatry in the 1990s (Reference FreemanFreeman, 1999).

LARGE-SCALE RANDOMISED STUDIES IN BIPOLAR DISORDER

We believe there is sufficient clinical uncertainty about maintenance treatment in bipolar disorder to warrant a large, simple, randomised, controlled trial. The key clinical question is: in patients for whom prophylaxis has been recommended, which treatment most decreases the subsequent admission rate? The main uncertainty concerns the relative efficacy of lithium and sodium valproate (and the combination of the two drugs) in the maintenance treatment of bipolar disorder.

A large-scale, randomised study, the Bipolar Affective Disorder: Lithium Anticonvulsant Evaluation (BALANCE) study, is being planned to compare the combination of lithium and valproate with either drug alone. Here we discuss some of the key methodological issues involved in the design of this trial, which is now being piloted in several centres in the UK.

The main requirement is for reliable evidence on the comparative efficacy of the interventions. Although double-blinding might protect against performance and ascertainment biases, it would be difficult to achieve and the pilot study suggests that it would reduce participation. Ascertainment bias can be reduced by the use of a relatively “hard” primary outcome (see below). The primary question is about the relative efficacy of the treatments in patients who are willing to remain on long-term treatment, and so the trial needs to ensure that the number of participants who drop out early is limited. There will therefore be a non-randomised run-in phase of 8 weeks. Participants who satisfactorily complete this run-in phase, and for whom there remains clinical uncertainty about the optimal maintenance treatment, will then be eligible for randomisation.

The primary outcomes in maintenance studies have been problematic in the past (Reference Bowden, Swann and CalabreseBowden et al, 1997). This is because of the complexity of the profile of disability caused by bipolar illness. In practice, the use of maintenance treatment with a mood stabiliser has the multiple aims of reducing the chances of suicide or severe recurrence requiring admission to hospital, and also of improving mood stability, reducing inter-episode mood symptoms and suicidal ideation, and improving overall quality of life. All these outcomes are important, but perhaps the most crucial is the reduction in the chances of severe relapse. We are therefore planning to use severe recurrence, defined by hospital admission, as the main outcome. The principal analysis will be survival analysis of time to readmission. Secondary outcomes will include clinical global impression, self-reported quality of life, adverse events, suicide attempts and use of additional psychotropic medication.

One consequence of using a clearly defined and clinically meaningful outcome such as hospital admission as the primary outcome is that a large sample will be required. An estimate of the probability of admission in bipolar disorder during a 2-year follow-up was provided by the Danish record linkage study (Reference Kessing, Andersen and MortensenKessing et al, 1998). It investigated the probability of readmission at different times in the course of the illness. Using these data, and assuming that most eligible patients would have had two to five recurrences, this produces an estimated expected readmission rate of approximately 50% over 2 years. Assuming an admission rate of 50% over 2 years and considering a 10% difference between treatments as clinically worthwhile, assuming a 30% drop-out rate (based on the mean drop-out rate from previous trials of maintenance treatment in bipolar disorder) for an α=0.05 and β=90%, each group would need to include 713 patients — a total of 2139 in a three-arm study. It is possible that the event rates will be lower because, for a number of reasons, patients in trials may be less likely to relapse (Reference Bowden, Swann and CalabreseBowden et al, 1997). If the event rates were 30% on lithium and 20% on valproate, as observed in the German carbamazepine study (Reference Greil, Ludwig Mayerhofer and ErazoGreil et al, 1997), 483 subjects would be required in each group. The sample sizes for various possible admission rates are shown in Table 1. Taking these estimates into account, we are aiming to recruit 3000 patients into BALANCE.

Table 1 Required sample size per arm for various admission rates (assuming a 30% drop-out rate)

Admission rate (lithium)	Admission rate (valproate)
	0.15	0.20	0.25	0.30	0.35	0.40	0.45	0.50
0.20	1353	—	1746	483	236	145	101	76
0.25	391	1746	—	2092	565	271	162	111
0.30	196	483	2092	—	2381	631	296	175
0.40	88	145	271	631	2608	—	2769	713
0.50	55	76	111	175	315	713	2869	—

Non-randomised studies have suggested a potentially important specific anti-suicide effect for lithium. A review of studies in this area found that the average suicide rate in the studies was 3.2 per 100 patient-years for patients not on lithium, compared with 0.37 per 100 patient-years for patients taking lithium (Reference Tondo, Jamison and BaldessariniTondo et al, 1997). A study of 3000 patients would have over 90% power to detect such a treatment effect. At least some of the apparent anti-suicidal effect of lithium is due to selection bias in the non-randomised studies (patients with a better prognosis are more likely to take lithium); the true effect is likely to be lower and it is doubtful if the study would have sufficient power to reliably detect a more realistic effect. However, it is clearly important to try to obtain some measure of suicidality, and BALANCE will therefore measure the number of suicide attempts in the lithium groups.

Finally, we must mention logistics. We have now spoken about BALANCE to many clinicians with busy practices in the UK and have been impressed by the immediate positive response that has been the rule so far. Continued support from the medical community will be essential for the BALANCE study. The challenge is to devise a treatment choice that expresses the dilemma many clinicians feel — and all, perhaps, ought to feel, given the facts we have reviewed above. The biggest step, however, will be to randomise patients between the alternatives. In principle this is little different from the private randomisations we all currently make when we have no real basis for informed choice of one option over another. If we can make it easy enough in practice to pool that uncertainty by a common act of randomisation, we can both solve the clinician's immediate dilemma and move on definitively to resolve it.

The BALANCE study may represent an important opportunity to define a new direction for clinical trials in psychiatry. Its development will carry lessons relevant to the evaluation of all drugs whose use is established in short-term studies but whose primary impact is intended to be in long-term prevention. By addressing a real clinical choice, it may prevent decisions being unduly influenced by simple marketing factors: it would be unacceptable if lithium use was increasingly discontinued because valproate is marketed actively and lithium is not. However, it is equally possible that undue conservatism in Europe is reducing the access of patients with bipolar disorder to a real alternative to lithium. We need to know which interpretation is correct. If successful, BALANCE could establish the pattern for maintenance trials of new drugs for long-term use in bipolar disorder.

Clinical Implications and Limitations