In recent years, results from large multicenter RCTs in patients with sepsis therapies including nutrition therapy and high-dose supplementation of micronutrients have essentially been negative or neutral and were, therefore, unable to confirm previous positive findings for (patho-) physiologically plausible therapies [13]. Reasons for this divergence were attributed particularly to the study design itself (e.g., choice of timing and dosing of a drug, synergistic effects seldom investigated, choice of unsuitable outcomes), inappropriate individualization of the intervention and phenotyping of the syndrome sepsis based solely on the clinically established SEPSIS-1 criteria [14,15,16].

Precision medicine approaches, that target patients based on disease subtypes, have transformed treatment approaches for malignancies, asthma and other heterogeneous syndromes and offer equally great potential in critical illness. Innovative approaches even postulate that it is necessary to abandon classical models such as the grouping of different symptoms into syndromes [11]. We obtained analogous results of the missing selenium treatment effect in septic patients, also after applying the subgrouping strategy, as we must note that numbers of individuals in the α- and β-phenotype divided into selenium and placebo group were too small to draw a valid conclusion. These data are, therefore, more of a descriptive nature. However, differences seen in outcome parameters in the α- and β-phenotype should be reevaluated in future studies. Focusing on these two phenotypes in particular to generate a sufficient number of individuals could ultimately answer the question of whether selenium is indicated in any subtype of septic patient. At this point, however, we are still unable to make a recommendation for application of high-dose sodium selenite in septic patients. In agreement with this, a recent systematic review and meta-analysis (SRMA) of 24 trials could not demonstrate a beneficial outcome as a consequence of high-dose selenium application in ICU-patients [17]. On the contrary, doses of more than 1000 µg per day were associated with increased length of stay (LOS) on the ICU [17]. Another SRMA, including 34 studies with 4678 patients, showed a tendency towards reduction of all-cause mortality by antioxidative micronutrient supplementation, albeit the authors did not consider these results to be robust. ICU–LOS and duration of mechanical ventilation therapy were reduced significantly by antioxidant administration [18]. Of note, in this SRMA antioxidative substances were summarized (selenium, zinc, vitamin A, C and E), although selenium accounted for the largest proportion. In 2022, Gudivada and coworkers found in their SRMA further results suggestive of beneficial outcome (e.g., LOS, ventilator days, infections) of ICU patients, again investigating antioxidative cocktails and not selenium as monotherapy, however [19]. In addition, supplementation of high-dose selenium did not improve postoperative organ dysfunction or mortality in cardiac surgery patients, a patient group that is known to have a pronounced inflammatory reaction, as it was shown recently in a large international multicenter trial [20]. As it is hypothesized that reactive oxygen species are to some extent essential for a functioning immune response, antioxidant strategies may, therefore, deteriorate outcomes in late immunosuppressive states of sepsis [21]. Measurements of markers of oxidative stress in blood samples might help to predict the effect of selenium and other antioxidative supplements on outcome parameters. However, we are unable to provide information on oxidative stress levels as measurements were not carried out in the SISPCT trial.

Analyzing further patient characteristics and outcomes, we confirmed, in line with Seymour’s investigation, that patients attributed to the β-phenotype were oldest and had the highest serum levels of creatinine and urea, indicating a higher proportion of renal dysfunction. Furthermore, signs of inflammation such as PCT levels and body temperature were, like in the original work, most elevated in the γ-phenotype, except for plasma CRP levels which were higher in the β-phenotype group. In addition, inconsistent with Seymour’s work, we found that the lowest albumin levels were not found in the γ- but in the β-phenotype. Liver dysfunction, reflected by bilirubin levels, which were highest in Seymour’s δ-phenotype, were on the contrary lowest in the same phenotype of our cohort. Different inclusion criteria in the respective studies are likely to account for these variations, as, for example, liver cirrhosis was an exclusion criterion in the SISPCT-trial [8].

Seymour and colleagues observed an increase in in-hospital deaths, 28- and 365-day mortality from the α- to the δ-phenotype [12]. Interestingly, in the studies analyzed by this group the β- and γ-phenotype showed a comparable mortality rate, whereas patients attributed to the α- and to the δ-phenotype had a considerably lower or higher respective mortality. When looking at the absolute values for early mortality, it is striking that especially the α-phenotype showed a remarkably low mortality-rate varying between 5 and 16% in the original publication compared 20.8% in this study. On the other hand, both early and late mortality in the δ-phenotype group was evidently higher in the original study. In our study a more pronounced, but nonetheless insignificant, gradation of the mortality-rate values was observed at an even earlier timepoint, such as mortality until day 28. In contrast to the phenotype-defining work, we could not detect differences in long-term mortality between the four phenotypes. The mortality in the entire SISPCT trial was comparable to other clinical studies performed by the SepNet Critical Care Trials Group, e.g., the VISEP trial [22] or the MAXSEP trial [23]. However, epidemiological studies like the INSEP trial [24] showed even higher mortality rates which can be explained by different inclusion criteria, sepsis origin, and ultimately by the size of the study population.

In the present study, the four sepsis phenotypes were applied retrospectively to the SISPCT trial and different frequency distributions of the individual phenotype were observed. In our cohort the γ-phenotype was the most common, followed by the δ-phenotype. The α-phenotype was on the other hand the rarest. This indicates a higher proportion of more severely ill patients. Suitably, comparing the SOFA score for the two study cohorts, the SISPCT-cohort had higher values (mean 10 vs. 3.9 points). However, the substantially lower incidence of the β-phenotype in our study cohort (6.3% vs. 27%), representing the phenotype with the most elderly patients likely to suffer from renal dysfunction, was neither reflected by the mean age (68 years in SISPCT vs. 64 years in the Seymour study), nor by mean laboratory markers of kidney function (creatinine 133 µmol/L and blood urea 9.8 mmol/L in SISPCT vs. creatinine 124 µmol/L and blood urea 8.6 mmol/L in the Seymour study) in the respective study cohorts. Recently, the sepsis phenotypes were applied for COVID and non-COVID sepsis patients [25]. Interestingly, the authors also found a different frequency distribution of phenotypes compared to the original publication. This study cohort consisted, like the SISPCT cohort, exclusively of patients who had already been admitted to the ICU, whereas Seymour and coworkers investigated a more inhomogeneous cohort. When considering patients with bacterial pneumonia, the authors found a similar frequency distribution of phenotypes as we did. In line with this observation the largest group of the SISPCT-cohort had a pulmonary focus [8]. In a secondary analysis of the PROWESS trial a pulmonary focus of infection increased the proportion of the γ-phenotype, supporting our data [26]. Although in the original publication by Seymour site of infection could not describe the phenotypes sufficiently, it may have at least an impact on the respective phenotype’s abundance.

In addition to the limitations inherent in the study design of a retrospective analysis, we only examined one trial as a proof-of-concept study. Larger data sets are required for generalizability to gain further insights into potential different treatment effects according to phenotype categorization as opposed to the clinically established SEPSIS-1 standard. Although we used multiple imputation for missing data at baseline, bias cannot be ruled out. Furthermore, we only applied one phenotyping strategy to the SISPCT study cohort. Whether other already established phenotyping methods show differential selenium treatment effects, e.g., hypo- versus hyperinflammatory sepsis cannot be derived from our study population but should be addressed in further clinical trials. Strength of this study is that it analyses a large and well-described cohort from a randomized multicenter RCT, including 33 multidisciplinary ICUs. Data of patients’ characteristics were documented diligently until day 21 with high adherence to the study protocol. Finally, mortality rates were recorded up to both day 28 and day 90.