Our large-animal randomized controlled experiment investigating the effects of Cytosorb HA in fecal peritonitis-induced sepsis did not demonstrate protection of animals from deterioration in fulminant sepsis, regardless of whether HA was initiated in the early or advanced phase of septic shock. In contrast, the HA resulted in a more aggressive vasopressor load and worse survival. Finally, HA in healthy animals resulted in hemodynamic deterioration, systemic inflammatory response, and cytopenia.

For over a decade, the critical care community has been discussing the immune response modulation and mediators adsorbing devices used in sepsis. Currently, six devices are marketed with different adsorption capabilities and research backgrounds [10]. The CytoSorb device stands out due to its demonstrably strong in vitro efficacy [19, 20], the number of clinical runs, and the evidence gathered over this time. CytoSorb decreased the levels of inflammatory markers, improved the hemodynamic parameters, and prolonged survival in small animals after cecal ligation puncture [21, 22] and fatal endotoxemia [23]. Based on such preliminary experimental data an over-optimistic clinical application has started. However, experiments on rodents cannot be directly clinically replicable on humans and the current evidence from human-based studies remains extremely inconsistent. Koehler et al. recently reviewed the evidence from 170 studies, analyzing the indications, possible effects, and research findings [24]. Besides the potential for eliminating exotoxins (anticoagulants) or endogenous substances (bilirubin, myoglobin, or free hemoglobin), the vasoplegic shock owing to systemic inflammation holds the greatest clinically relevance. Such effect has been observed in several observational studies and a recent meta-analysis of 33 studies by Hawchar et al. [3] reported a positive impact on norepinephrine dose and hemodynamic stability. An overall Hedge’s g index of 1.64 should indicate a large clinical effect, but the population comprised 140 patients from four controlled trials, of which only one was randomized. Unlike this one randomized trial with 20 septic shock patients [6], in two much larger trials, no significant improvement by the HA was observed [5, 7]. Schadler et al. [5] randomized 97 patients with septic shock, but was unable to demonstrate any effect on systemic levels of interleukin-6; unmatched mortality was higher in the treatment group (effect, which disappeared after adjustment). Another randomized multicenter pilot compared 23 CS-treated patients with 26 controls [7]. The primary outcome of this study was negative (vasoplegic shock resolution). Besides, it was negative in multiple secondary endpoints including systemic levels of inflammatory markers, catecholamine requirements, and adverse events rate. During the COVID-19 pandemic, several small randomized controlled trials were used to test the application of CS to decrease the impact of inflammation without any meaningful effect [7, 8, 25]. This conflicting evidence may be a result of our poor understanding of the actual mechanisms and the role of inflammatory mediators during septic shock, and a lack of rational treatment dose and timing.

Based on the in vitro data, CS allows the adsorption of significant amounts of substances ranging from 5 to 55 kDa. The removal rate for most interleukins was above 90% after 120 min exposition [19]. However, in vivo, such an effect was not always observed. In a pig model of burn injury, CS led to a decrease in the inflammatory markers in the circuit without affecting the systemic concentrations [26]. Because the concentrations of the pre-post filter mediators differ, the dose delivered may be inadequate to cause any significant effect. Schultz et al. demonstrated that a higher delivery dose was favorable, while a low dose was not [27]. Based on the Schultz equation [27] the delivery dose in the Linden study [26] was 2.25 L/kg (one 300 mL CS cartridge per 6-h treatment with a 250 mL/min blood flow in a 40 kg pig). In our study, the delivery dose was 2.4 L/kg (one 300 mL CS cartridge per 10-h treatment with a blood flow rate of 200 mL/min in a 50 kg pig). However, in the study by Schultz [27], the higher delivery dose was based on prolonged and multiple device runs rather than high flow rates. Therefore, the initial effect of CS should consider the blood flow and body weight only, and it was 375 mL/kg during the first hour in a study by Linden et al. [26] and 240 mL/kg in our experiment. This means that the doses are two-to-three times more intensive as compared to normal human treatments with 150 mL/h blood flows and a weight of 80 kg (112.5 mL/kg). One may hypothesize that the mediator production in unopposed peritoneal sepsis may exceed the device capacity. This seems unlikely because the systemic concentrations between the treated groups and septic controls did not differ significantly both in our study and the study by Linden et al. [26]. Besides Linden [26] demonstrated the pre-post filter decrease even after 6 h of the CS run.

The impact of the HA timing was one of the primary aims of our trial. In several studies, early-initiated HA was associated with better results [28,29,30]. On the contrary, in current praxis, HA is considered the last resort in irreversible vasoplegia. A dose of norepinephrine above 0.5 μg/kg/min in patients with adequate fluid loading and cardiac output was considered as a trigger by Hawchar et al. [6]. A recent meta-analysis reported a similar pre-treatment norepinephrine dose (median 0.55 (0.39–0.9) μg/kg/min) [3]. In our EARLY animals, the HA led to immediate norepinephrine rate escalation, and the survival time after HA initiation was significantly shorter in the LATE animals, despite similar sepsis initiation time. Moreover, the survival time of the septic animals without HA was the longest. From this point of view, our data do not support “the earlier, the better” notion and even raise certain safety concerns. In our experiment, HA seems to propagate vasoplegia, increase endothelial permeability, and negatively affect the platelets and leukocyte counts and circulating protein levels. No convincing explanation exists for our observation. The ability of blood-biomaterial interaction to activate either the humoral and cellular host response (i.e., bio-incompatibility) or remove beneficial substances should be taken into consideration. The increase in inflammatory mediators and drop in albumin levels observed in our animals supports the plausibility of these mechanisms. However, we did not specifically address this issue by pre- and post-cartridge samples. On the contrary, the safety signal observed in our study is in line with the results of two prospective studies demonstrating potential HA-induced harm [31].

Our experiment has several limitations, which may have affected our results and influenced extrapolation to humans. First, the fact that human and pig native immunology reactivity to insults may differ has to be taken into account. Besides, animal aging does not directly correspond to human development, and hence it may significantly differ from matured or senescent human pathophysiology. Further, it was an open-label study, but the strict experimental protocol should have reduced any sources of experimental bias, and blinding was imposed on data analysts (all biochemical data analyses).

No source control and antimicrobials were provided. Our model was designed to create hyperdynamic sepsis, with a progressive increase in severity over time. Antibiotic therapy was expected to blunt the host response, thereby attenuating the development and full manifestation of a true clinical septic response over 24 h, which is what we instead sought to elicit in our experiment. Such ongoing inflammatory stimulus may overcome the HA's ability to counteract. However, such a situation may not be that rare in clinical routine. Source control is not always possible as demonstrated previously [32] and administration of inadequate antimicrobials may still occur. Prior research suggests that HA can affect antimicrobial levels [33], introducing further unwanted uncertainties into our experiment.

To reduce the number of experimental animals we did not include a sham-operated HA untreated CONTROL group. To analyze the unwanted (and previously not expected) HA effects in the SHAM animals, we used a historical control from a previous experiment [18]. Since our methodology has remained consistent over a large period, we believe it does not introduce errors in our results.

Finally, a potential limitation of our study was the small number of experimental animal subjects. Such a complex experimental approach is expensive, highly time-demanding, and requires adequately trained critical care staff; hence, precludes using high numbers of animals. This limits the use of robust statistical methods. However, it enables us to study in detail the underlying pathophysiology on individual subject level and create extreme conditions (i.e., no-source-control fulminant sepsis); a process never possible in human subjects, but potentially useful to prove or refute certain hypothetical treatment effects. For instance, the time to initiate HA therapy in the LATE group was shorter than that in the EARLY group, which may indicate divergent sepsis dynamics on the individual level; however, the HA effect was not diverse. We believe that our experimental approach still provides a valuable alternative to fill the actual knowledge gap in settings where adequately large randomized clinical trials are difficult to conduct. Encouragingly, all pigs that underwent Cytosorb treatment demonstrated consistent non-beneficial responses. Thus, it is unlikely that increasing the sample size would have altered the results.