In this study of critically ill COVID-19 patients, plasma endostatin levels of 100–200 ng/mL at ICU admission were associated with AKI development on ICU day 1, RRT need, and 90-day mortality independent of age, sex, CRP, and creatinine. Adding endostatin to creatinine improved AKI prediction on ICU day 1 but not RRT. Incorporating endostatin into a model containing age, sex, CRP, and creatinine enhanced the prediction of both AKI on ICU day 1 and 90-day mortality, but not RRT.

The relationship between endostatin and AKI, RRT, and 90-day mortality was non-linear, with the highest risk observed near 200 ng/mL. SAPS-3, SOFA score, lactate, AKI, RRT, and mortality were highest with endostatin levels of 100–200 ng/mL, suggesting a correlation with organ dysfunction. Furthermore, this group had the highest CCI and prevalence of diabetes and CKD. The strong association between endostatin and these markers of disease severity supports the role of endothelial dysfunction and dysregulated angiogenesis in the pathophysiology of critical COVID-19. Approximately 4% of the population had very high endostatin levels (> 200 ng/mL), a finding not previously reported in critically ill patients, with or without COVID-19 [12, 13, 22]. The reason for the decreasing AKI, RRT, and mortality rate with endostatin > 200 ng/mL is not entirely clear. We hypothesised that the high rate of aspirin treatment (47%) in the endostatin > 200 ng/mL group may have provided protective effects against AKI. Endostatin has been associated with platelet activation in COVID-19, and platelet activation has been implicated in the pathophysiology of COVID-19-associated AKI [23, 24]. Furthermore, aspirin use has been linked to reduced mortality in COVID-19 in a large meta-analysis [25]. Another possible explanation is that the effects of endostatin follow a non-linear, concentration-dependent pattern. This has been observed in oncological research, both in vitro and in vivo, where endostatin inhibits endothelial cell proliferation and migration in a dose-dependent manner up to a certain threshold. Beyond this point, however, further increases in endostatin levels paradoxically decrease its inhibitory effects, indicating a biphasic or U-shaped response. Such dose–response relationships have also been shown for other angiogenesis inhibitors [26, 27].

In a general ICU setting, a 2016 study showed that adding endostatin to a predictive model enhanced AKI prediction [13]. Similarly, a recent study concluded that admission endostatin, age, and creatinine effectively predicted AKI and RRT [15]. In contrast, a larger multicenter study reported limited utility for endostatin as a predictive marker for AKI and RRT, despite levels increasing with KDIGO stages [22]. Unlike this study, this investigation was not focused on COVID-19 patients, had a lower AKI rate, and did not stratify patients by endostatin levels. The low area under the curve of endostatin observed in that study could reflect a non-linear relationship between endostatin, AKI, and RRT, as highlighted in the present study.

Endostatin has also been linked to mortality in critically ill patients with AKI and in a general ICU cohort [14, 15]. In critical COVID-19, elevated endostatin levels have been associated with hypoxia and mortality [12]. These findings further underscore the prognostic value of endostatin in critical disease.

Endostatin is expressed in various tissues, including the renal tubular epithelium, Bowman’s capsule, and the basement membrane of the kidneys and lungs [28]. Therefore, it is plausible that endostatin may be released during pulmonary and renal injury. With a molecular weight of 20 kDa, endostatin is expected to be filtered through the glomeruli [29, 30]. Declining GFR during AKI likely contributes to elevated plasma levels [31]. In addition, experimental data suggest that endostatin may play an active role in AKI development [32,33,34]. Thus, elevated endostatin levels may reflect kidney injury and reduced GFR and actively contribute to AKI.

Renal impairment is often detected late, as creatinine changes appear only after significant kidney injury [35, 36]. Incorporating novel biomarkers like endostatin could enhance early risk stratification and AKI diagnosis, allowing timely intervention for reversible factors like hypoperfusion and nephrotoxic exposures, ultimately improving ICU outcomes.

The observed association between endostatin, RRT, and mortality may partly be explained by its strong relationship with AKI. Given that AKI is a well-established risk factor for RRT and mortality, it is possible that endostatin primarily reflects AKI severity rather than being independently associated with RRT and mortality per se. Even though factors beyond AKI classification influence RRT need, it is often a consequence of severe AKI, making it difficult to distinguish between the two fully. However, it may be more clinically relevant to interpret RRT as a severity marker of AKI rather than as an isolated outcome. While endostatin was associated with RRT, its lack of predictive improvement likely reflects the complexity of RRT need, the strong predictive role of creatinine, and the potential need for dynamic biomarker assessment rather than a single admission value. RRT initiation is not solely determined by AKI severity but also by fluid balance, acid–base status, and haemodynamic factors, which may limit endostatin’s added value in RRT prediction. These findings suggest that while endostatin enhances AKI risk stratification, its role in guiding RRT prediction may be more limited.

Future research should include larger studies to validate these findings and explore the intriguing observation of decreasing risk of poor outcomes with very high endostatin levels (> 200 ng/mL). In addition, larger studies in other ICU cohorts beyond COVID-19 are needed to assess endostatin’s broader applicability. Combining endostatin with other emerging AKI biomarkers could enhance diagnostic accuracy and risk stratification, potentially guiding therapeutic interventions in future clinical trials. Further research is also needed to clarify the precise role of endostatin in AKI development and its potential as a therapeutic target.

The multicenter design of this study enhances its generalisability to critically ill COVID-19 patients. Its prospective approach enabled systematic follow-up and detailed data collection on comorbidities, laboratory parameters, and daily AKI status. Robust statistical models, adjusted for age, sex, CRP, and creatinine, minimised confounding and reinforced the associations between endostatin levels, AKI, RRT, and 90-day mortality.

The observational design introduces the possibility of residual confounding despite adjustments for age, sex, CRP, and creatinine. Factors such as ICU burden and SARS-CoV-2 variants may have influenced patient outcomes during the study period. While we did not have variant-specific data for individual patients, surveillance data indicate that the B.1.1.7 (alpha) variant was predominant in Sweden at the time [37]. Although the ICU burden has been associated with increased mortality in critically ill COVID-19 patients, we believe this is unlikely to have significantly impacted our findings, as our models adjusted for key confounders, and ICU admission policies remained relatively consistent across centres [17]. Although the study involved multiple ICUs in Southern Sweden, the findings may not be fully generalisable to other regions or healthcare settings. The study’s focus on COVID-19 also limits its applicability to non-COVID ICU populations, as mechanisms of AKI may differ. The non-linear relationship between endostatin and outcomes may add complexity to its interpretation. Endostatin levels were categorised based on a combination of prior evidence, the observed distribution in our cohort, and pragmatic considerations. However, we acknowledge that endostatin is a continuous variable, and future studies should explore optimal thresholds and alternative modelling approaches. Furthermore, the subgroup with very high endostatin levels (> 200 ng/mL) was relatively small, potentially limiting statistical power to draw firm conclusions. This prevented a meaningful sensitivity analysis of aspirin’s potential protective effects in this group. Larger studies are needed to characterise patients with very high endostatin levels better.