Procalcitonin‐guided or C‐reactive protein‐guided antibiotic discontinuation protocols for adult patients who are critically ill with sepsis

The systematic review by Rafiq et al. [1] addresses an important question, but several included studies appear to deviate from the authors' prespecified research question. The authors specified that eligible studies should include patients with sepsis who were receiving antibiotics and organ support and were treated in ICU settings, while studies without ICU patients and those using broader antibiotic management strategies were not included. However, the study by Mahmutaj et al. was not conducted in an ICU setting [2]. Although the PROGRESS study by Kyriazopoulou et al. met Sepsis-3 criteria, the original article states clearly that it was conducted in seven medical wards; the patients were not treated in an ICU, but received advanced supportive care on the wards in the context of ICU bed shortages [3]. Thus, the question addressed by Rafiq et al. appears to have expanded from patients in the ICU with sepsis to patients hospitalised with sepsis or infection. The authors' own meta-regression showed that baseline Sequential Organ Failure Assessment (SOFA) explained 70% of the between-study variance in the effect on antibiotic duration, and that for each one-point increase in SOFA, the effect of procalcitonin in reducing antibiotic duration was attenuated by 0.55 days. Therefore, the inclusion of non-ICU or lower severity studies may amplify the antibiotic-reducing effect of procalcitonin. Inconsistency between the target population or intervention and the research question constitutes indirectness and should affect the certainty of the evidence [4]. The SISPCT study by Bloos et al. also appears not to meet the criterion of a discontinuation-only protocol [5]. That study evaluated procalcitonin-guided anti-infectious therapy. When procalcitonin failed to decrease, the algorithm recommended optimising or changing antibiotics and even implementing new source-control interventions [5]. This conflicts with the authors' statement that their review avoided mixing initiation, intensification and discontinuation protocols. Moreover, the authors reported long-term mortality event counts for Bloos et al. as 529/552 and 516/537, approximately 96%, which is inconsistent with the 28-day mortality rates of 26% and 28% reported in the original study [5]. Given that this study contributed 50% of the weight in the long-term mortality analysis, if these figures were not mortality events, both the pooled long-term mortality result and the GRADE rating require correction. We would welcome the views of the authors on the points we have highlighted as to whether the conclusions of their analysis require moderation.