Melioidosis, the infection caused by Burkholderia pseudomallei, is endemic in South East Asia¹ and is the commonest cause of death from bacteraemic pneumonia in the Northern Territory of Australia.² Patients with severe infection present with septic shock, a condition associated with a high mortality. The Royal Darwin Hospital has treated 341 patients with culture confirmed melioidosis over 13 years, with an overall 18% mortality. Forty two of these patients presented with severe sepsis.

In 1998, the hospital decided to start treating patients with melioidosis and septic shock with granulocyte-colony stimulating factor (G-CSF). The decision was based on the evidence available at that time (box). G-CSF was introduced at the same time as the hospital appointed a specialist in intensive care medicine. Previously, anaesthetists had supervised the unit, and the appointment resulted in appreciable changes to management protocols, including the more aggressive use of haemodynamic monitoring, empiric antibiotic protocols, the adoption of a closed intensive care model, and early enteric feeding.

Mortality from severe melioidosis fell from 95% (20 of 21 patients) to 10% (2 of 21 patients) (P < 0.001, Fisher's exact test).¹⁰ A more modest reduction also occurred in mortality from septic shock due to other pathogens.¹⁰ The fall in mortality was particularly surprising because newly published data suggested that G-CSF was of little benefit in non-neutropenic infection.^8,11,12

A trial was proposed to determine whether the fall in mortality from severe melioidosis was due to G-CSF or other management changes. We planned to randomise eligible patients to either G-CSF or placebo and assess the end points of death from any cause or deterioration to catecholamine resistant shock. Patients who declined to participate in the study would be treated with the usual protocol, which included routine use of G-CSF. However, it was clear from the outset that the study posed ethical problems.

Discussions of the ethics of randomised controlled trials emphasise the need for equipoise to exist before a trial can proceed. Equipoise has three components: clinical, personal, and patient. Clinical equipoise is usually taken to mean that uncertainty exists about the research question among researchers or the expert community as a whole. In this case the medical community disagreed about the value of G-CSF. This could be seen as evidence that a trial of G-CSF in severe melioidosis was both necessary and ethically justifiable.

Figure 1 B pseudomallei—the bacteria that causes melioidosis

Although clinical equipoise was present, many local experts were uneasy about their personal equipoise and that of their patients. They felt that the risks to patients of forgoing G-CSF were too great. Randomised controlled trials are mostly advocated in situations where only a moderate or small effect is expected rather than the large effect that was possible in this case. Peto et al suggest that in these circumstances randomised trials may be unnecessary.¹³ For example, treatment of syphilis with penicillin has never been tested in a randomised controlled trial, but trials involving withdrawal of penicillin are rightly regarded as unethical.

In 1991, Royall et al attempted to deal with this issue by distinguishing truly experimental trials from unethical demonstration trials that are designed to show the superiority of already successful treatments more convincingly.¹⁴ Our situation shows the difficulties with this concept—does our potentially confounded series constitute sufficient evidence of effectiveness?

Traditionally, equipoise has been viewed from the paternalistic viewpoint of the medical community. However, awareness is growing of the need to consider the perspective of patients and the wider community (see bmj.com). For example, Brody has argued that patient equipoise is present if “a reasonable person of an average degree of altruism and risk adversiveness might consent to being randomised.” This is analogous to clinical equipoise being determined by the normative judgment of a community of physicians.¹⁵

In our case, information about the apparent value of G-CSF was already well known within the community. A local newspaper had described it as a “wonder drug.”¹⁶ It is difficult to see how our survival statistics (95% v 10% mortality) would lead anybody other than the most extreme altruist to agree to participate in the proposed trial, no matter what the experts thought.

The proposal for the trial was eventually abandoned because of these ethical considerations. Yet the question remains whether G-CSF is of any value in the treatment of septic shock associated with melioidosis. Several variations to the proposed trial design were suggested in the hope that they would decrease the risks to participants.

One method would be to test the current dosing regimen against shorter, delayed, or less potent regimens rather than placebo. This is similar to the approaches used in trials of zidovudine to prevent vertical transmission of HIV in developing countries.¹⁷ Because smaller doses are likely to have some effect, a larger trial would be required, but the risks to the individual patient would be decreased. This approach did not prove practical because of the need for a greater sample size than was feasible.

A second method would be to use surrogate end points that might allow deteriorating patients to be withdrawn from the trial early. This approach was used in a trial of steroids in Pneumocystis carinii pneumonia.¹⁸ Unfortunately, appropriate surrogate end points have not been well defined for trials in patients with septic shock.

A third method of making the study safer would be to use an adaptive trial design such as “play the winner,”¹⁹ where allocation of patients to each arm of the trial is based on previous results of the intervention. These designs have been used in trials of extracorporeal membrane oxygenation.²⁰ Although they limit the number of possible adverse outcomes, they are unable to prevent them altogether. Importantly, they do not limit the risks to an individual patient who enters early in the trial.

A fourth method would be to exclude the most seriously ill patients. This would potentially decrease the risk of omitting G-CSF but would also increase the number of patients required for the study and decrease the applicability of the results to seriously ill patients.

Researchers have attempted to use animal and in vitro models to explore the role of G-CSF in acute melioidosis.²¹ However, given that such models have failed to translate into improved outcomes in human trials of G-CSF in non-neutropenic infection,⁶ such evidence is no substitute for a clinical trial.

More than half of the Australian cases of melioidosis are treated at our institution. Thus, if a trial cannot be done at our hospital, it is unlikely to be feasible elsewhere within Australia. We have, however, proposed a trial in conjunction with colleagues in Thailand. Recent guidelines suggest that trials of new treatments may be ethical in developing countries if the results might benefit both the participants of the trial and others in the community. G-CSF is relatively expensive, but its use would be sustainable in Thailand if a substantial benefit could be shown. However, differences in the provision of intensive care between the two countries may mean the results are not directly applicable to patients in Australia.

The use of G-CSF in severe melioidosis remains unproved, and other institutions seem unlikely to start using it for patients with septic shock without additional evidence. Most trials take place in the context of limited but promising evidence of a modest benefit. Our situation shows the problems in defining the state of uncertainty necessary for the conduct of an ethical trial when a substantial benefit has been observed, however flawed this evidence might be. Future advances in the understanding of melioidosis may show other plausible reasons for the fall in mortality. If this were the case, G-CSF might come to be seen as an optional part of treatment and a trial would become possible. However, at present it does not seem possible to limit the risks to individual patients within the current constraints of sample size.