Pulmonary artery catheterization was adopted about 30 years ago (Swan 1970) and, in common with most technologies used within intensive care units (ICUs), was widely disseminated without any rigorous evaluation as to whether it reduced deaths or the cost‐of‐care in these patients. Despite this, many doctors consider the use of a pulmonary artery catheter (PAC) to be beneficial. PAC is used to measure cardiac output and pressures in the pulmonary circulation. These parameters, together with derived figures calculated from these measurements, are then used to adjust treatment. It is the ability of PAC to provide these data that clinicians feel is beneficial, even in the absence of consensus about the specific interpretation of the data.

In 1996 a review of most comparative and randomized trials involving pulmonary artery catheterization was published (Cooper 1996). Of the 34 published studies reviewed, only one was considered "level 1" evidence (Gattinoni 1995). This showed no benefit for goal‐directed therapy (using PAC) in a mixed ICU population. The remaining studies were equally split between those showing no or worsened outcome with PAC and those showing a benefit. Since the review was published two further randomized controlled trials of PAC in patients undergoing major vascular surgery (Bender 1997, Valentine 1998) have been published in high‐impact journals, and one trial on high risk elective surgery (Wilson 1999). These showed no benefit, increased complications and an improved mortality respectively. A meta‐analysis of major morbidities in 12 randomized controlled trials involving PAC was also published recently (Ivanov 2000) showing a very small reduction in morbidity with PAC.

Connors et al. (Connors 1996) rekindled the debate around PAC with the publication of an observational study using the data from a high quality, clinical database to provide a risk adjusted, but non‐randomized, comparison of outcomes of critically ill patients receiving or not receiving PAC within the first 24 hours following admission to ICU. The results indicated that PAC was associated with increased risk of hospital mortality (39%, 95% confidence interval 15%‐67%) and increased utilisation of resources. This led the Intensive Care National Audit & Research Centre (ICNARC) to respond to the "Consultation to identify NHS R&D Priorities" in October 1996 by identifying PAC as a technology urgently requiring evaluation.

There was considerable media coverage in the United States (Winslow 1996; Friend 1996; Ass. Press 1996) in relation to the Connors et al. study, which led to a formal press release from the Society of Critical Care Medicine (SCCM) (Hoyt 1996) in the USA challenging the conclusions of the study, predominantly because it was a non‐randomized comparison. The SCCM promised, "to convene an expert panel to discuss PAC".

In December 1996, the SCCM convened a multidisciplinary Consensus Conference on PAC (funded in an unrestricted manner by industry). In this context, multidisciplinary was defined as ensuring representation of members from all the relevant clinical specialties rather than involvement of all the relevant scientific disciplines (such as epidemiologists and health economists). The Consensus Statement was published in June 1997 (PAC Consensus 1997). In general, the statement identified the level of published evidence to support the use of PAC was paltry and, scientifically, very poor. However, there was no recommendation for a moratorium on PAC use, although the need for clinical trials was highlighted.

In December 1997, a critique of the Consensus Statement was published referring to it as an unsystematic, narrative review (Sibbald 1997). The Consensus Statement supplied insufficient methodological detail and was flawed due to the:

• potential for biased selection of the participants (no detail on selection was provided)

• positive bias in the questions posed (for example, "does PAC improve outcomes for…", "does PAC reduce complications from…")

• selection and review of the evidence not based on defined criteria, for example, no information on the sources of literature/databases searched, no explicit search strategies for relevant studies, no criteria for assessing study validity were outlined except study design

• lack of obvious expertise involved in critically appraising the literature

• details of the homogeneity between studies not being provided

The Consensus Statement relied on "expert opinion" for response and the method of consensus for that response was, therefore, of paramount importance. Explicit, scientific methods exist for reaching consensus, for example, nominal group or Delphi techniques (Murphy 1998), but none were used. There was no evidence that the panel was unbiased in its views of the use of PAC (most, if not all, were practitioners) and no conflict of interest statements were published prior to commencing the conference.

The Connors et al. study provoked numerous editorials concerning the clinical and cost‐effectiveness of PAC (Dalen 1996; Reinhart 1997). In the UK this was addressed in an editorial in the British Medical Journal (Soni 1996). The correspondence that followed (Dexter 1996; Lazarus 1996; Konarzewski 1996) suggested there was considerable equipoise amongst UK clinicians. MacKirdy et al (MacKirdy 1997) undertook a risk‐adjusted comparison of outcomes using Scottish data and replicated the results of the Connors study.

The present review aims to systematically search for and combine all evidence from randomized controlled trials relating to the clinical and cost effectiveness of PAC‐Management in adult intensive care patients in order to supply the best evidence currently available on which to base a multi‐centre, randomized controlled trial (RCT) and therefore recommendations for clinical practice.