Two weeks ago I posted a blog about an impeccable NEJM study on the effects of procalcitonin (PCT) on antibiotic use in patients with lower respiratory tract infection. I stated that this RCT was one of the first diagnostic studies in this disease area targeting the correct patients and ended by an invitation to identify the fatal flaw. Last week one of the PhD students (Valentijn Schweitzer, absent when the paper was discussed in our journal club) told me that searching a fatal flaw was not needed; as the RCT was unnecessary in the first place. Here is why.
In short, they randomised 1,656 patients in the emergency department with an initial diagnosis of acute lower respiratory tract infection (LRTI) to either treat according to PCT results (no antibiotics if the PCT is low) or as usual care (as PCT was not known real-time). The PCT information, coupled to a guideline how to act upon PCT results, failed to change overall antibiotic use and patient outcome, compared to usual care.
In the intention-to-treat analysis, which best mimics clinical practice, the adherence to the PCT guideline was high, 72.9%. The question then is what would the maximum possible effect on antibiotic prescription be, if every patient was treated according to the PCT guideline (100% adherence). The investigators estimated this by doing a “per guideline analysis” using an instrumental variable approach. This approach quantifies/estimates the effect on antibiotic prescription when all (100%) patients would have been treated according to the PCT guideline, and the difference in antibiotic-days between both study groups would have been −0.1 day (95% CI, −1.0 to 0.8). In other words, the PCT algorithm to guide antibiotic therapy never could have changed antibiotic use, even not with 100% compliance to protocol. Instead of performing a costly non-inferiority RCT, the authors could have come to this conclusion with a much more simple and cheaper observational cohort study to quantify the amount of patients that would be ‘eligible’ to an altered treatment regimen based on the current PCT algorithm and compare that to the antibiotics actually prescribed.
So, why did full adherence to protocol not lead to a change in antibiotic prescription?
- Many patients in the usual care group with low PCT did not receive antibiotics;
- Only few patients had low PCT;
- Less antibiotics in patients with low PCT was off-set by more antibiotics in patients with high PCT levels;
- Any combination of the three.
As 77.4% of all patients had low PCT, we can strike out number 2. The study data suggest that the first and third reason occurred frequently; 65.9% of usual care patients with low PCT were not treated with antibiotics and among those with high PCT those in the study group were 23.5% more likely to receive antibiotics (Figure S3; 21.9% more in PCT>0.25-0.5 and 1.6% in PCT>0.5 group). A higher likelihood of getting antibiotics in case of a high PCT could be beneficial for patients if they indeed have a bacterial infection. Yet, the number of patients receiving antibiotics due to high PCT is very small (approx. 6 in the PCT>0.25-0.5 group and approx. 1 in the PCT>0.5 group of a total of 826 patients). Therefore, the first explanation is the most important reason for the PCT failure.
The study domain was formulated as follows: “We enrolled patients in the emergency department for whom the treating clinician had given an initial diagnosis of acute lower respiratory tract infection (LRTI), but had not yet decided to give or withhold antibiotics and about whom there was uncertainty regarding the need for antibiotics, such that PCT data could influence the prescribing decision.” Apparently, PCT was not needed to make the clinicians decide not to give antibiotics in (too) many episodes.
What would then be a better patient population to demonstrate the usefulness of PCT? Simple, patients usually treated with antibiotics, but in whom clinicians dare to withhold based on low PCT (example Christ-Crain et al. where antibiotic was started in 99% of patients with CAP during usual care and in 85% with PCT) or use the PCT algorithm solely to discontinue antibiotics in patients with low PCT, rather than use it to start antibiotics based on high PCT (example De Jong et al.). And then the question remains whether only PCT can motivate clinicians to do so.
In the spirit of the ongoing soccer World Cup; In the NEJM study PCT played a game that could not be won (as the final result had already been set before the game started). And in the spirit of world politics: Antibiotic stewardship in the US hospitals is too good for PCT (they are great, again). And in the spirit of evidence-based medicine: an RCT – even if unnecessary in hindsight – is more convincing than any observational study.
Valentijn Schweitzer does a PhD on effectiveness of antibiotic stewardship in patients with CAP
2 thoughts on “Procalcitonin-guided antibiotics for respiratory tract infections (part 2)”
Thanks Valentijn for your well-explained point of criticism. The ProACT trial shows that changing antibiotic prescription behavior of physicians is an extremely difficult task (due to low algorithm adherence, do you agree?), or as you describe: ‘patients usually treated with antibiotics, but in whom clinicians dare to withhold based on low PCT’. Regarding the latter, I believe the article you cite as an example (Christ-Crain et al.), antibiotic was started in 99% of patients with CAP during usual care and in 85% with PCT. Though, the difference is a bit smaller than you suggested, or did you mean something else?
This is correct. We mixed up percentages between studies. Will be adjusted immediately.