Full Text View

Novel aspects of the PEPTIC trial design

The PEPTIC trial design incorporated a number of novel aspects that have important implications for the future of ICU research (Table 1). The trial used existing registry data sources predominantly, which greatly reduced the amount of data that needed to be collected from individual patients. This cluster, crossover design ³ tested regimens of stress ulcer prophylaxis implemented at the level of the ICU. ⁴ Each ICU used either a PPI or H₂RB for 6 months and then switched to the alternative class of drug for the subsequent 6 months. The order of treatments used in study ICUs was randomised. The cluster crossover design embedded the trial into usual clinical practice and meant a large number of patients were enrolled in a short time frame; every patient invasively mechanically ventilated within 24 hours was included in the trial by default. Arguably, for the first time in a randomised clinical trial in intensive care research, there was sufficient power to detect what might reasonably be considered a minimum clinically important difference. While the initial, pre-trial sample size calculations suggested the trial would provide 80% power to detect a 2.4% absolute risk difference, ⁴ in reality, the trial provided 80% power to detect a 1.8% absolute risk difference. ¹ The reason for the discrepancy was mainly because of a difference between the estimated within- and between-period cluster correlation coefficients used in the sample size calculations and the observed coefficients in the trial itself. ⁵

Remarkably, the trial afforded similar power to what would have been observed with an individual randomised controlled trial with the same number of participants. This occurred principally because there was little variability in the in-hospital mortality rates of each ICU over time. Such variability is labelled as “between periods within cluster” variability, and is what remains after taking the within-ICU differences between the two interventions — namely, ICU constant factors are cancelled out, and only factors that vary over the two observation periods remain. Minimising this variability is a key component of the power of a cluster crossover design, ⁵ and as indicated in the additional statistical analyses of in-hospital mortality provided in the supplementary appendix of the PEPTIC article, ¹ this variability was essentially zero. This is the absolute statistical best-case scenario for a cluster crossover trial. However, this best case may not be generalisable to future cluster crossover trials in ICU because it depends on design characteristics of the trial (ICU locations, observation period lengths), outcome measures being assessed, and intrinsic variation in outcomes between patients within the same ICU. Nevertheless, based on what was observed in the 50 ICUs in five countries in the PEPTIC trial, it is possible that such variability over 6-month periods among patients who are invasively mechanically ventilated within 24 hours of ICU admission is consistently small.

Relevance to the design of future trials

Even despite these considerations, given the efficient recruitment rates we achieved, the implications of such statistical power for future trials of ubiquitous ICU interventions, such as fluid therapy, oxygen therapy, nutrition and blood pressure targets, are potentially profound. With no observed between-group difference, a 95% confidence interval in a trial similar to PEPTIC would be expected to exclude either an increase or decrease in mortality of one percentage point. Yet, with power to detect absolute mortality differences of 2%, small differences in mortality potentially attributable to idiosyncratic practice variations for ubiquitous therapies are now identifiable. ⁶ While an absolute mortality difference of 2% may appear small, this equates to a number needed to treat of 50, and to 2000 lives saved or lost for every 100 000 patients treated. Even smaller differences will be detectable if multiple crossovers are performed, ⁷ although such designs may pose additional logical difficulties that need careful consideration.

In the PEPTIC trial, while most of the key data were obtained from registries, some data were collected at an individual patient level. Using a combination of registry and individual patient data made it relatively easy to conduct the trial in multiple countries. Using registry data sources greatly reduced costs compared with collecting trial-specific data at an individual patient level. The trial was conducted with less than $500 000 of funding, a cost of under $20 per patient. Such comparatively low cost means that this trial technology might allow future large scale trials to be conducted in low income countries where registries are rapidly developing. ⁸ Even in higher income countries, registry-embedding potentially means sites that do not have resources for research coordinators can contribute, resulting in a greater generalisability of trial results.
 

Trade-offs inherent in the PEPTIC trial design

One limitation of the PEPTIC trial, which has received attention in trial commentary, ⁹ is the amount of non-adherence with assigned treatment. Around 20% of patients admitted when an ICU was assigned to H₂RBs received PPIs. Such non-adherence confounds interpretation of the trial results with regard to the efficacy of individual drugs and may have been reduced if more resources had been devoted to educating staff in study centres about trial protocols and procedures. On the other hand, this may simply be an unavoidable trade-off with a cluster crossover trial that means these types of trials are best considered to be about the effectiveness of implementing particular treatment strategies rather than about the efficacy of the individual medicines being compared. In future trials of this nature, efforts to incorporate process evaluation may give a clearer idea of what drives non-adherence to ICU-assigned therapy. Such process evaluation is likely to be most important for common interventions where there may potentially be strongly held beliefs that drive practice. Collecting data on drivers of non-adherence may also enable compliance-adjusted statistical analyses that address estimates of efficacy of the individual medicine in addition to those of strategies. ¹⁰

Despite the degree of non-adherence to assigned therapy that occurred, there was substantial separation in the exposure to drug classes between treatment groups, and randomisation provides a sound basis for determining that observed differences in outcomes were attributable to these differences in drug exposure. One common criticism of large scale pragmatic trials is that they fail to account for the individual differences in patient characteristics that clinicians might use at the bedside to inform decision making. ¹¹ However, because the PEPTIC trial was so large, it provides a unique opportunity to explore the relative safety and effectiveness of PPI and H₂RB treatment strategies in different patient groups.

The rationale for exploratory analyses using the PEPTIC trial data

We have already seen that a strategy of using PPIs rather than H₂RBs in patients with high illness acuity, ¹² who are at the greatest risk of clinically important upper gastrointestinal bleeding is associated with an increased risk of death. Other groups of potential interest include sepsis, chronic liver disease, trauma and burns, coagulopathy and bleeding, abdominal surgery, brain injuries, and renal failure. The rationale for conducting future exploratory analyses in these patient subgroups are shown in Table 2.
 

Conclusions

Despite some necessary trade-offs, the unique methodology used in the PEPTIC trial has many potential advantages compared with a conventional individual patient randomised controlled trial and may provide the impetus for future research activities. Particularly for ubiquitous ICU therapies, this trial design may prove attractive in the future.  There is also a strong rationale for a number of exploratory analyses using PEPTIC trial data.