Skip to main content

Foundational Trials

PAC-MAN

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Assessment of the clinical effectiveness of pulmonary artery catheters in management of patients in intensive care (PAC-Man): a randomised controlled trial
  • Acronym: PAC-Man
  • Year: 2005
  • Journal published in: The Lancet
  • Citation: Harvey S, Harrison DA, Singer M, Ashcroft J, Jones CM, Elbourne D, et al; PAC-Man study collaboration. Assessment of the clinical effectiveness of pulmonary artery catheters in management of patients in intensive care (PAC-Man): a randomised controlled trial. Lancet. 2005;366(9484):472-477.

Context & Rationale

  • Background
    Pulmonary artery catheters (PACs) were widely adopted in critical care as invasive haemodynamic monitoring devices, but the link between PAC-derived variables and improved patient-centred outcomes remained unproven in routine ICU practice.
    Concerns about harm, cost, and “monitoring without a treatment algorithm” intensified after a large risk-adjusted observational cohort study reported worse outcomes and greater resource use in patients receiving right heart catheterisation early in ICU care.1
    By the early 2000s, clinical equipoise persisted: PACs were still used for complex shock and multiorgan failure, yet practice variation was substantial and prior randomised evidence was limited and heterogeneous.
  • Research Question/Hypothesis
    In adult ICU patients for whom the treating clinician believed PAC-guided management was indicated (but not mandatory), does a strategy of management with a PAC reduce hospital mortality compared with management without a PAC?
  • Why This Matters
    The PAC is an invasive device with recognised insertion risks; if it confers no net benefit in contemporary ICU decision-making, routine use would represent avoidable harm and cost.
    Conversely, if benefit exists only in specific phenotypes or when coupled to protocolised responses, PAC-Man’s pragmatic design could help define what question was answered (effectiveness) versus what remained unanswered (efficacy/algorithmic care).

Design & Methods

  • Research Question: In adult ICU patients judged by clinicians to warrant PAC use, does PAC-guided management (vs no PAC) reduce ultimate hospital mortality?
  • Study Type: Pragmatic, multicentre, open-label, randomised controlled trial in 65 adult general ICUs in the UK; allocation minimised by ICU and key prognostic variables; ICUs pre-specified a stratum regarding use of alternative cardiac output monitoring in controls (stratum A: no alternative permitted; stratum B: alternative permitted).
  • Population:
    • Setting/timeframe: UK adult ICUs; enrolment October 2001 to March 2004.
    • Inclusion: ICU admission; age ≥16 years; treating clinician intended to manage the patient with a PAC (clinical judgement of indication), with sufficient equipoise to allow randomisation.
    • Key exclusions: Age <16 years; elective pre-operative haemodynamic optimisation; PAC already in situ on ICU admission; previous enrolment; haemodynamic optimisation for organ donation.
    • Screening yield: 1263 eligible; 1041 randomised; 222 excluded (110 lack of equipoise; 47 refusal by next of kin; 65 other reasons).
  • Intervention:
    • Management with a PAC, inserted as soon as possible after randomisation according to local practice.
    • Subsequent clinical management (fluids, vasoactive drugs, ventilation, renal support) at clinician discretion rather than protocolised targets.
    • PAC remained in place for as long as the treating clinician deemed necessary (median for first catheter 2 days; total days indwelling 3 days).
  • Comparison:
    • Management without a PAC (usual care).
    • Rescue PAC insertion permitted if clinicians lost equipoise (occurred in 24/522 allocated to control).
    • In stratum B ICUs, alternative cardiac output monitoring devices were permitted in the control arm (use frequency not reported in the main trial paper).
  • Blinding: Unblinded (device-based intervention; clinicians and bedside teams were aware of allocation). Primary outcome (mortality) was objective, but co-interventions and resource-use outcomes were susceptible to performance bias.
  • Statistics: Initially, 5673 patients were required to detect a 5% absolute change in hospital mortality (10% relative change) with 90% power at the 5% significance level, allowing for non-compliance (4% PAC group; 8% control) and 5% loss to follow-up; during recruitment, the target sample size was revised to 1281 to detect a 10% absolute change in hospital mortality (14.5% relative change) with 90% power at the 5% significance level, with similar non-compliance allowances and planned 100% follow-up; primary analysis was intention-to-treat (Fisher’s exact test; Cox proportional hazards with and without adjustment for prognostic factors and randomisation minimisation variables; non-parametric testing for secondary outcomes).
  • Follow-Up Period: Until final discharge from an acute hospital ward; in-hospital survival analysed to 90 days; one control-group patient still in hospital 3 months after recruitment ended was censored.

Key Results

This trial was not stopped early. An independent data monitoring and ethics committee performed two interim safety analyses and did not recommend halting recruitment.

Outcome PAC-guided management No PAC Effect p value / 95% CI Notes
Ultimate hospital mortality (primary) 346/506 (68%) 333/507 (66%) Adjusted HR 1.09 95% CI 0.94 to 1.27; P=0.25 Fisher’s exact P=0.39; unadjusted HR 1.07 (0.92 to 1.24); P=0.40; one control patient censored (still in hospital at 3 months after recruitment ended).
ICU mortality 304/506 (60%) 291/508 (57%) Not reported P=0.37 Fisher’s exact test.
28-day mortality 314/506 (62%) 305/508 (60%) Not reported P=0.52 Fisher’s exact test.
ICU length of stay (survivors) 12.1 days (IQR 6.2–22.3) 11.0 days (IQR 5.7–21.0) Not reported P=0.26 Wilcoxon rank-sum.
ICU length of stay (non-survivors) 2.6 days (IQR 0.7–8.4) 2.5 days (IQR 0.8–7.2) Not reported P=0.71 Wilcoxon rank-sum.
Hospital length of stay (survivors) 34 days (IQR 23–61) 40 days (IQR 21–70) Not reported P=0.43 Wilcoxon rank-sum.
Hospital length of stay (non-survivors) 3 days (IQR 1–11) 3 days (IQR 1–11) Not reported P=0.90 Wilcoxon rank-sum.
Organ-days of ICU support (survivors) 19 (IQR 12–33) 19 (IQR 10–32) Not reported P=0.32 Composite “organ-days” based on augmented care period dataset definitions.
Organ-days of ICU support (non-survivors) 9 (IQR 4–20) 8 (IQR 4–21) Not reported P=0.74 Composite “organ-days” based on augmented care period dataset definitions.
Direct PAC insertion complications 46/486 (10%) — Not reported Not reported Most frequent: haematoma 17 (4%), arterial puncture 16 (3%), arrhythmias requiring treatment within 1 h 16 (3%; including 1 cardiac arrest); also pneumothorax 2, haemothorax 1, lost guidewire retrieval 2; none fatal.
  • Mortality: No evidence of benefit (or harm) for PAC-guided management; ultimate hospital mortality 68% vs 66% with adjusted HR 1.09 (95% CI 0.94 to 1.27; P=0.25).
  • Process/uptake: Median time from randomisation to PAC insertion was 1.7 h (IQR 1.1–2.7); one or more management changes within 2 h of insertion were reported in 389/486 (80%) of PAC insertions (fluid bolus ≥200 mL in 205 [42%]; vasoactive dose change >25% in 211 [43%]; vasoactive initiation in 156 [32%]).
  • Subgroups: No convincing heterogeneity for the primary outcome; for example, stratum A HR 1.21 (0.87–1.68) vs stratum B HR 1.06 (0.90–1.26), interaction P=0.48; “acute respiratory failure” subgroup HR 1.39 (0.91–2.13), interaction P=0.69.

Internal Validity

  • Randomisation and allocation concealment: Central 24-hour telephone randomisation with minimisation by ICU, age band, presumptive clinical syndrome, and surgical status (with pre-specified ICU stratum A vs B); allocation concealment was maintained until assignment.
  • Post-randomisation exclusions: 27/1041 excluded from analysis (control: 13; PAC: 14) due to withdrawal or lack of retrospective consent (control: 9) and relatives refusing/withdrawing agreement (control: 4; PAC: 5).
  • Crossover/non-adherence: Control allocation: 24/522 (5%) were managed with a PAC (predominantly due to loss of equipoise; one staff error); PAC allocation: 34/519 (7%) were not managed with a PAC (insertion unsuccessful 14 [3%]; clinical condition changed such that PAC deemed inappropriate 14 [3%]; coagulopathy/safety concerns 6 [1%]).
  • Timing and “dose” of intervention: Median time from randomisation to PAC insertion 1.7 h (IQR 1.1–2.7); median duration of first PAC 2 days (IQR 1–3) and total PAC days 3 (IQR 2–4).
  • Baseline comparability (selected examples): Mean age 64.7 vs 65.3 years; APACHE II total score 22.1 vs 22.7; median APACHE II risk of death 0.37 (IQR 0.23–0.57) vs 0.39 (IQR 0.23–0.55); mean SOFA at randomisation 8.6 vs 8.6; median time from ICU admission to randomisation 16.2 h (IQR 5.8–42.0) vs 15.3 h (IQR 4.3–34.8).
  • Separation of the variable of interest: Despite crossovers, the intervention arm received rapid PAC placement and demonstrable “actionability” (management changes within 2 h in 389/486 [80%] insertions), supporting meaningful separation at the level of monitoring strategy.
  • Performance/detection bias: Unblinded delivery could influence discretionary co-interventions and resource use; however, the primary endpoint (ultimate hospital mortality) was objective and less susceptible.
  • Heterogeneity and subgroup signals: Broad case-mix (e.g., multiorgan dysfunction 65–66% at enrolment) increases clinical heterogeneity; pre-specified subgroup interaction tests did not show compelling effect modification for mortality.
  • Statistical considerations: Revised target sample size was 1281, but 1041 were randomised (1014 analysed); the trial reported power sufficient to detect a large (10% absolute) mortality difference, but limited ability to exclude smaller effects.

Conclusion on Internal Validity: Overall, internal validity is moderate: randomisation and objective mortality endpoints support causal inference for the pragmatic “PAC strategy” question, but unblinded care, crossovers, and limited power for modest effects constrain interpretability for efficacy in specific phenotypes or algorithm-driven use.

External Validity

  • Population representativeness: Adults in UK ICUs with severe acute illness and high mortality (66–68% ultimate hospital mortality), enrolled when clinicians considered PAC use relevant; major syndrome at randomisation was multiorgan dysfunction in 65–66%.
  • Practice context: Pragmatic delivery reflects “real-world” clinician decision-making and local PAC expertise; however, most patients were enrolled in units where alternative cardiac output monitoring in controls was permitted (stratum B: 401/506 [79%] PAC vs 401/508 [79%] control).
  • Applicability limits: Findings do not directly address elective perioperative haemodynamic optimisation (explicitly excluded) or tightly protocolised resuscitation strategies where PAC-derived variables trigger standardised interventions.
  • Temporal generalisability: Subsequent expansion of echocardiography and less-invasive monitoring may further reduce the marginal utility of PACs in many ICUs, but PAC-Man remains relevant as an effectiveness trial of routine PAC deployment.

Conclusion on External Validity: External validity is moderate: results generalise well to adult ICUs considering PACs for severe shock/multiorgan failure in settings similar to early-2000s UK practice, but are less informative for specific phenotypes, algorithmic haemodynamic optimisation, or modern echocardiography-dominant workflows.

Strengths & Limitations

  • Strengths:
    • Large, multicentre pragmatic RCT across 65 ICUs with broad case-mix, enhancing relevance to routine practice.
    • Central randomisation with minimisation and pre-specified subgroup framework (including ICU policy strata).
    • Objective primary endpoint (ultimate hospital mortality) with near-complete follow-up (one censored control patient).
    • Detailed reporting of adherence, early management changes attributable to PAC-derived data, and insertion complications.
  • Limitations:
    • Unblinded “monitoring strategy” trial; co-interventions were at clinician discretion and could dilute any benefit of PAC information.
    • Permitted rescue crossover (5% of controls received PAC) and some non-delivery in the intervention arm (7% without PAC).
    • Revised sample size target (1281) was not achieved (1041 randomised), limiting exclusion of smaller but clinically meaningful effects.
    • Trial did not evaluate a standardised treatment algorithm tied to PAC variables, nor quantify competency/interpretation quality.
    • Control-arm use of alternative cardiac output monitoring was permitted in most participants (stratum B), narrowing inference to the incremental value of PAC over other monitoring strategies.

Interpretation & Why It Matters

  • Clinical meaning
    In a pragmatic UK ICU context, adding a PAC to usual clinical decision-making did not improve survival and introduced non-trivial insertion complications (10% reported), supporting a selective (not routine) approach to PAC placement.
  • Mechanistic interpretation
    PACs provide physiological measurements; benefit depends on how reliably clinicians translate data into effective interventions.
    PAC-Man showed that PAC data frequently triggered management changes (80% within 2 h), yet these changes did not translate into measurable outcome improvement at the trial level.
  • Economics and health technology assessment
    The broader UK Health Technology Assessment programme incorporated PAC-Man within a combined clinical- and cost-effectiveness evaluation, reinforcing that “no mortality benefit” must be interpreted alongside device cost, training burden, and complication risk in policy decisions.2
  • Research implications
    PAC-Man is best read as an effectiveness trial of a monitoring strategy; it leaves open whether protocolised PAC-guided therapy, or highly selected subgroups, might benefit (an “efficacy” question requiring different trial architecture).

Controversies & Subsequent Evidence

  • What question did PAC-Man truly answer? The accompanying Lancet commentary highlighted that PAC-Man compared PAC use against a control strategy that could include alternative monitoring technologies in many sites, and that the “no-alternative” stratum A represented a smaller (and potentially underpowered) subset of ICUs, shaping inference towards incremental effectiveness rather than PAC “value in isolation”.3
  • Monitoring is not therapy: A subsequent journal commentary emphasised that lack of benefit may reflect (i) variable interpretation of PAC data, and/or (ii) absence of a standardised treatment response to haemodynamic targets—therefore negative results should not be over-interpreted as proof that PAC-derived physiology is never useful.4
  • Consistency with other randomised evidence: Other ICU/acute respiratory failure trials similarly failed to demonstrate improved outcomes from early PAC-guided management compared with controls, aligning with PAC-Man’s conclusion in different health systems and case-mixes.56
  • Synthesis of evidence: A JAMA meta-analysis of randomised trials and a later Cochrane review concluded that pulmonary artery catheterisation did not reduce mortality in adult ICU populations and did not reliably improve key secondary outcomes, supporting de-implementation of routine use.78
  • Practice change and “technology death” narrative: Observational data demonstrate major declines in PAC use in the US across the 1990s–2000s, consistent with accumulating concerns about benefit, harms, and availability of alternatives; this decline accelerated in the wake of key publications and trials.9
  • Guidelines/consensus positioning: Major consensus guidance on shock and haemodynamic monitoring increasingly frames PACs as a selective tool (e.g., complex shock when echocardiography/less invasive monitoring is insufficient and results will change management), rather than as routine ICU monitoring.1011

Summary

  • PAC-Man was a pragmatic UK multicentre RCT testing a management strategy: PAC-guided care versus no PAC in ICU patients for whom clinicians judged PAC use relevant.
  • There was no improvement in ultimate hospital mortality (68% vs 66%; adjusted HR 1.09; 95% CI 0.94 to 1.27; P=0.25) and no signal of benefit in ICU mortality, 28-day mortality, length of stay, or organ-support days.
  • PAC placement changed management frequently (80% had ≥1 change within 2 h), yet this did not translate into outcome benefit, underscoring the distinction between physiological measurement and effective therapeutic strategy.
  • Insertion-related complications were reported in 10% of attempted placements (none fatal), reinforcing the need to justify invasive monitoring by clear, actionable clinical questions.
  • Subgroup analyses (including ICU policy strata) did not identify a convincing subgroup with mortality benefit, though the trial was not powered to exclude smaller effects or evaluate protocolised PAC-driven algorithms.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • PAC-Man is a classic example of “effectiveness of a monitoring strategy” rather than “efficacy of a protocolised haemodynamic intervention”; interpretation should track the level at which randomisation occurred (device strategy) and the level at which treatment decisions were allowed to vary (clinician discretion).

Overall Takeaway

PAC-Man is a landmark effectiveness trial showing that, in routine UK ICU practice, adding a pulmonary artery catheter to clinician-directed management did not improve survival and carried measurable procedural risk. Its enduring influence is methodological as much as clinical: it clarified that invasive monitoring without a standardised, evidence-based treatment response is unlikely to change outcomes, and it helped catalyse a shift towards selective PAC use and alternative haemodynamic assessment strategies.

Overall Summary

  • In adults with severe ICU illness where clinicians considered a PAC relevant, a pragmatic PAC-guided strategy did not reduce mortality versus no PAC, and complications occurred in 10% of attempted insertions.

Bibliography