This study not only confirms previous work demonstrating that admission E2 levels are associated with mortality but also supports our hypothesis that E2 levels during the course of critical illness correlate more strongly with mortality than admission values. First, for the repeat E2 value, there was a larger separation in the median E2 levels for survivors versus non-survivors and a greater separation in mortality between the lowest and highest quartiles relative to the admission value. In addition, mortality for patients in whom E2 was in the highest quartile at repeat analysis was significantly greater than those in the lowest quartile at the second analysis, regardless of the admission quartile. Finally, the unadjusted AUROC for the repeat E2 was greater than the AUROC for the admission E2 and adding the trend in E2 to multivariate regression models including admission E2, APACHE II, age, trauma status, gender, and diabetes improved the model’s performance. Although the association of cytokines with outcome of critical illness has been studied to a greater extent than sex hormones, E2 correlated at least as strongly with mortality whether obtained upon study enrollment or during the hospital course. This may be explained in part by the relatively short half-life of cytokines relative to serum E2, making serum E2 a clinically preferable indicator of mortal threat. Whether E2 is actually a mediator of the inflammatory process and contributing to outcome or simply a marker of disease severity is not known[36
] and cannot be determined from these data. Answering such questions demands further investigation regarding the contribution of sex hormones to the outcome of critically ill and injured patients and, until such work is performed, calls for estrogen supplementation in critically ill patients are premature.
While this study did not investigate the source of E2 in this patient population, previous studies demonstrate that increased levels are related to the increased peripheral conversion of androgens to estrogens by the peripheral aromatase enzyme under the stimulatory effect of class I cytokines and TNF[24
]. Furthermore, E2’s potential to contribute to inflammatory processes in critical illness has been established. It is a well-documented immunomodulator[37
] altering monocyte and macrophage function and life span[39
] and cytokine and chemokine production. These changes have translated into early survival in animal models, but could lead to an exaggerated inflammatory response later in the clinical course. In addition to its inflammatory properties, estradiol has also been shown to induce and activate nitric oxide synthase in endothelial cells which may contribute to low peripheral resistance in response to shock[40
]. Estrogens are also known to modulate insulin resistance, an important contributor to outcomes after critical injury.
The strengths of this study include its large sample size, prospective data collection by trained experts, and multi-institutional nature. There are several important limitations however. The initial admission E2 level was obtained within 48 hours of study enrollment, introducing the possibility of sampling bias, and possibly missing early elevations in E2. Additionally, patients who died or were discharged from the ICU within 48 hours were excluded, limiting the ability to generalize data to these patients. Finally, nearly 30% of patients in our cohort lacked complete cytokine data. Therefore, comparisons made between models containing E2 and changes in E2 vs. models containing cytokines and changes in cytokines based only upon available data in the subgroup analysis may not be generalizable to the larger cohort. Potential bias resulting from missing cytokine data is minimized however by the prospective study design, in which blood samples were drawn from all patients twice weekly, regardless of changes in clinical status or risk of death. Hence, the lack of cytokine data was not in any way affected by decisions made by the patient’s clinicians. Additionally, the adjusted AUROC for E2 in the entire 1408 patient cohort is nearly identical to the AUROC for E2 in the 1010 patient subgroup with complete cytokine data. This reduces the concern that our cytokine results were biased by missing data.