Propensity-Score Analysis Reveals that Sex is Not a Prognosis Factor for Mortality in Intensive Care Unit-Admitted Patients with Sepsis

Purpose: Classically, men have been considered to have a higher incidence of infectious diseases, with controversy over the possibility that sex could condition the prognosis of the infection. The aim of the present work was to explore this assumption in patients admitted to the ICU with sepsis using a robust statistical analysis. Methods: Retrospective analysis (2006-2017) in patients with microbiologically conrmed bacteremia (n=440) by majoritarian bacterial pathogens. Risk of ICU and in-hospital mortality in males respect to females was compared by an univariant analysis and a propensity score correspondence analysis integrating their clinical characteristics. Results: Relevant differences were related to the infection source: urinary origin for females (28.7% vs 19.8%) and abdominopelvic surgery for males (8.8% vs 4.8%). Sepsis occurred more frequently in males (80.2% vs 76.1%) as well as in-hospital (48.0% vs 41.3%) and ICU (39.9% vs 36.5%) mortality. Escherichia coli was 2 times more frequent in survivors whereas Staphylococcus aureus was 3 times more frequent in deceased patients. Univariate analyses showed that males had a higher Charlson comorbidity index, a poorer McCabe prognostic score; however the propensity score in 296 patients demonstrated that females had higher risk of both ICU (OR 0.72; 95% CI 0.46 to 1.13), and in-hospital mortality (OR 0.84; 95% CI 0.55 to 1.30) but without statistical signicance. Conclusion: Men with sepsis have worse clinical characteristics when admitted to the ICU, but sex has no inuence on the prognosis of mortality. Our data contributes to help reduce the sex-dependent gap present in health care provision.


Introduction
Sepsis is currently de ned as life-threatening organ dysfunction caused by a dysregulated host response to infection [1]. Microorganisms trigger the systemic immune response with direct cytopathic injury after release of proin ammatory mediators, resulting in endothelial injury, microvascular thrombosis and ultimately tissue ischemia. The cellular damage leads to organ dysfunction that can eventually result in septic shock and death. Sepsis-related mortality has been estimated at 18%-35% [2], being higher in patients admitted to intensive care units (ICUs) [3]. The assessment of the global sepsis impact remains undetermined for developed countries; however the World Health Organization estimates an incidence of 30 million cases and 6 million deaths per year, which represents a major public health concern with a high economic burden [4,5].
One of the major challenges for ICU clinicians is to identify the prognostic factors that can predict the clinical course and outcome of sepsis. In this sense, the patient's sex has recently been hypothesized as a possible mortality prognostic factor in sepsis of ICU-admitted patients [6]. Both sex (biological and physiological factors speci c to male or female) and gender (the social role, the activities attributed to men and women) can in uence the acquisition, progression and prognosis of infectious diseases [7].
Examples of how sex differences could constrain the susceptibility to infective agents and the severity of the clinical presentation are those related to the immune system [8], and to pregnancy-related hormonal changes [9,10]. In the same way, gender-related differences, such as the smoking use to be more common among men, making them more susceptible to respiratory tract infections [11]. It has been classically established by empirical evidence that males are at higher risk of sepsis than females [12], but it is still poorly understood whether sex differences could in uence sepsis outcome.
Some studies have speci cally assessed the role of sex or gender as a prognostic factor in patients with sepsis, yielding contradictory results [13]. Such differences could be attributed to methodological heterogeneity, lack of consideration of potential confounding factors and to setting (local or country) related factors in uencing sepsis outcome. This was a retrospective observational study designed to analyze the impact of sex on the prognosis of ICU-admitted patients with sepsis, only considering the microbiologically con rmed cases by the most common pathogens, and adjusting for a wide set of potential confounding factors through a propensity score matching analysis.

Case selection strategy
Our institution is a tertiary hospital comprising 7 independent ICUs attending a metropolitan area with more than 1 million of inhabitants. We included all adult (≥ 18 years old) ICU-admitted patients with laboratory-con rmed bacteremia over an 11-year period (2006-2017), stratifying by etiological agent and excluding those caused by sporadic microorganisms (< 1% frequency). This sampling scheme allowed the inclusion of each microorganism as an individualized factors in the subsequent statistical analysis.
Only cases that ful lled the current de nitions of sepsis or septic shock were selected (n = 731). Trained personnel performed a complete and exhaustive review of the clinical charts of the selected patients from ICU admission until death or hospital discharge. Only the rst ICU admission of each patient was included, and the recorded variables included demographic data (age, sex), comorbidities measured by the updated Charlson comorbidity index [14], reason for admission, diagnosis, McCabe prognosis index, outcome and length of hospital stay. Other information regarding the bacteremia episode was also documented, including clinical presentation (no sepsis, sepsis, or septic shock), origin (community or nosocomial), primary focus, source control measures, causal agent and antimicrobial susceptibility of the causing isolates (multiresistant or not). The distinction between gender and sex is often di cult, and for this work we have used the sex (male/female), as previously recommended [15]. The primary considered outcomes were mortality in the ICU and in-hospital mortality.

De nitions
To consider a bacteremia episode as laboratory-con rmed, the culture and identi cation of a recognized pathogen (an organism not included on the National Healthcare Safety Network common commensal list), from at least one blood culture was required. Cultures were performed in both aerobic and anaerobic blood culture bottles (BD BACTEC™ Plus Aerobic/F and BD BACTEC™ Lytic/10 Anaerobic/F, respectively) that were incubated in a BACTEC™ FX instrument (Becton & Dickinson, Belgium) following the manufacturer's instructions. Sepsis and septic shock status were assessed following the criteria agreed upon by Singer et al. [1]. Sepsis is de ned as an acute change in the total Sequential Organ Failure Assessment (SOFA) score ≥ 2 points consequent to the infection, whereas septic shock corresponds to a subset of sepsis with persisting hypotension requiring vasopressors to maintain mean arterial pressure ≥ 65 mm Hg and a serum lactate level > 2 mmol/L despite adequate volume for resuscitation. Nosocomial acquisition is designated when a patient has a positive blood culture from 48 hours after their hospital admission, whereas community-acquired sepsis refers to a positive blood culture that occurs within the rst 48 hours of hospital admission. Control measures correspond to key interventions in primary focus, including surgery, debridement, drainage and removal of a potentially infected device. Antibiotic multiresistant bacteria are those without in vitro phenotypic susceptibility to rst-line antimicrobial therapy and with resistance to at least 3 antibiotic families. ICU mortality was de ned as death before ICU discharge, and in-hospital mortality as that occurring within the period of hospitalization.

Statistical analysis
Firstly and to characterize our population, a descriptive analysis strati ed by sex (male/female) were realized. We expressed categorical variables as absolute and relative frequencies and continuous variables as means and standard deviation or medians and interquartile ranges. Thereafter, we performed a univariate analysis evaluating all potential factors likely associated with the 2 primary outcomes (ICU and in-hospital mortality). Categorical data were compared by chi-squared tests, whereas Student's t-tests were used for sex comparison of continuous variables when normal distribution could be assumed or by Mann-Whitney U test otherwise. We assessed normality of the continuous variables using the Shapiro-Wilk test. Subsequently, we performed a propensity score analysis to assess the effect of sex on primary outcomes while adjusting for the imbalance in the patient's characteristics observed between males and females [16]. To obtain the propensity score, we tted a logistic regression model, with sex as a binary dependent variable and the potential confounders as independent variables. A propensity score was used to match female to male, without replacement in a ratio 1:1. Matched pairs were chosen using a caliper of 0.2 of the standard deviation of the propensity score (logit scale). We computed standardized differences for all variables included in the propensity score before and after matching to assess the effect of matching on the imbalance. We deemed a 10% standardized difference as the limit for a correct balance. After matching, we compared mortality (ICU and in-hospital) between females and males using a generalized estimating equations model to account for matched data. Stata (V.15; StataCorp. 2017) software was used for the statistical analysis.

Case selection
Our retrospective search yielded 5,520 positive blood cultures from 2,063 adult ICU patients during the selected 11-year period. Patients with bacteremia exclusively caused by microorganisms included in the NHSN-CDC common commensal list and by sporadic microorganisms were excluded (n = 1,288), as were duplicated episodes from the same patient (n = 44). The remaining 731 patients/episodes of bacteremia were attributed to major pathogens such as Escherichia coli (24%), Candida sp. (13%), Staphylococcus aureus (12%), Enterococcus faecalis (10%), Pseudomonas aeruginosa (10%), Klebsiella pneumoniae (10%), and Streptococcus pneumoniae (7%). It was also detected 14% polymicrobial episodes. Finally, 440 out of the 731 patients initially selected met the criteria of sepsis or septic shock, and were ultimately used for statistical analysis (Fig. 1).

Descriptive Analysis
Data obtained from the clinical charts reviewed from the 440 patients (62% males and 38% females) are summarized in

Univariable Analysis
Factors signi cantly associated (p < .05) with the two selected outcomes, ICU and in-hospital mortality, were Charlson index, mode of acquisition (higher rates of nosocomial origin between deceased patients), clinical presentation (higher septic shock rates in deceased patients), and the presence of an antibiotic multiresistant bacteria ( Table 2). The anatomical focus of the bacteremia was signi cantly associated with mortality; the urological tract was approximately 3 times more frequent in survivors than in deceased patients and endocarditis was 7 times more frequent in the deceased than in survivors. Finally, the etiological agent was also associated with broad mortality differences, ranging from E. coli, which was 2 times more frequent in survivors, to S. aureus that was 3 times more frequent in deceased patients.

Propensity Score Matching And Risk Estimation
Signi cant variables associated with mortality in the univariable analysis were further selected to perform propensity score matching. After a pairing algorithm, a total of 296 cases (148 from each sex) were matched. Figure 2 shows the standardized differences by sex, before and after perform the matching.
After matching, only endocarditis as the primary focus in men and P. aeruginosa as the etiologic agent for women remained with a minor disbalance (Table 1 and Fig. 2).
The between-sex difference in risks for both ICU and in-hospital mortality were calculated in the matched cohort (Table 3). Our results showed that the risk of ICU mortality was 28% lower in males (OR 0.72; 95% CI 0.46 to 1.13), as well as the risk of in-hospital mortality, which was 16% lower in males (OR 0.84; 95% CI 0.55 to 1.30). Although there was no statistical signi cance in the matched analysis for either ICU or inhospital mortality, these ndings contrast with the univariate results in which men presented higher a ICU (39.9% vs 36.5%) and in-hospital (48.0% vs 41.3%) mortality risk.

Discussion
The possible link between sex and infectious diseases has classically been debated, and males are often attributed as having a greater predisposition for infection and a poorer prognosis [12]. Previously published works focused on sepsis have obtained contradictory results [13], making it worthwhile to deeply evaluate the actual sex and gender impact on sepsis prognosis, and validating it in each country or geographical area. This is the rst study from our country regarding this topic, and we performed an exhaustive analysis to mitigate confounding factors by propensity score matching. Our study on ICU patients with sepsis revealed, after propensity score matching, that ICU and hospital mortality rates are slightly lower for males, and this result is opposite to that observed in the univariable analysis, reinforcing the need to adapt the analytical strategy for each situation. However, given this observation had not statistical signi cance, we can only conclude that both ICU and in-hospital mortality rates of ICU patients with sepsis or septic shock are not signi cantly in uenced by sex. This conclusion is consistent with other previously published works [17][18][19][20][21][22][23]. Also, as occurs in other series [23,24], our male patients had more comorbidities, higher septic shock incidence and both ICU and in-hospital mortality. On the contrary, women were associated with urinary primary focus of sepsis and a lower incidence of Gram-positive bacteremia, which has been widely reported [18].
Experimental studies have been performed to unravel the physiological mechanisms that could explain these observations and also to validate sex at a prognostic level. In animal studies, females have more advantageous immunological and cardiovascular responses against severe infections such as sepsis [24,25] by the direct effect of their estradiol [26][27][28]. Moreover, genetic aspects, such as the female X chromosome mosaicism, could confer a diversi cation of leukocyte responses during endotoxemia [29].
In contrast, the 5a-dihydrotestosterone present in males appears to exert a deleterious effect, weakening cardiovascular functions [29] and promoting the cytokine-mediated response [31][32][33]. Several experiments in sepsis have demonstrated the therapeutic utility of the administration of the oestrogen precursor dehydroepiandrosterone (DHEA), and the blockage of androgen-related adverse effects through the administration of androgen receptor antagonists such as utamide [34].
It is generally assumed that males have a higher incidence of severe forms of sepsis [35,36] and more organ failure [18,19], which spurred the interest to evaluate whether females have milder symptoms and better outcomes during a septic episode. However, clinical studies attempting to evaluate the relationship between sex and sepsis prognosis have failed to achieve consistent results. All the recently published studies (in the last 15 years) that include an evaluation of the in uence of sex on the outcome of ICU patients with sepsis among their primary objectives are summarized in Table 4. Several reasons could explain the disparity of results, the most evident being the geographic location of the studies, which implies socio-economic and racial differences of the enrolled population, as well as differences in diagnostic accuracy and therapeutics. Case de nitions are also relevant; given some studies include a mix of surgical and medical ICU patients, or only patients with severe sepsis or septic shock. Importantly, the de nitions of gender and sex were not clear in some studies, so the results could be misleading. There are also important differences in study design, outcome endpoint, statistical approach (i.e. which methodology was used to control for confounding factors) and which studies were selected. The latter point is critical, given the di culty in identifying true confounders. For instance, there is accumulated evidence indicating that males receive both more invasive procedures and earlier antimicrobial therapy than females. A recent systematic review and meta-analysis aimed to evaluate gender-related mortality risk in ICU patients with sepsis [13]. The authors found a slightly higher risk of mortality in women, although this result was considered inconclusive given the heterogeneity of the results obtained in each of the separately selected studies.
The strengths of our study include the fact that is the rst work in which the latest sepsis criteria have been applied [1]; the selection of the patients was made for the rst time on the basis of microbiological data, selecting only episodes caused by the most prevalent and relevant pathogens; and the use of a propensity score matching approach allowed us to obtain 2 groups carefully matched on a large set of confounding factors. Finally, this study is the rst to assess the effect of sex on sepsis mortality for the Spanish population.
Our main limitations include the retrospective nature of the study, the inclusion of a single center, and the fact that we were unable to control for hormonal status and the immunological host-response. We did not perform strati cation by age to separate premenopausal and postmenopausal women, contrary to what was performed in some previous studies [23,36,41]. Finally, we matched in the propensity score by focus and etiologic agent, due to the aforementioned differences obtained by sex. These factors could also be observed as true drivers of the pathophysiology of sepsis; thus matching by them could have biased the true sex mortality differences observed in the present work. Despite our efforts to adjust for a number of possible confounders of the association between sex and survival, we cannot disregard the possibility that residual confounding is still present due to unobserved variables.

Conclusion
Evidence regarding sex-related differences in susceptibility and host-response to sepsis has not been consistently supported by clinical studies due to its heterogeneity and the di culties to control for underlying biases. Our ndings showed that males have a better prognosis of sepsis in the ICU compared with females, although the con dence interval prevents us from excluding a signi cant reduction or increase in mortality risk. Our group is currently working on a systematic review that aims to comprehensively assess the prognostic role of sex in critically ill adults with sepsis; protocol registered in PROSPERO as CRD42019145054 [44]. Accumulated knowledge in this area could lead to the development of sex-targeted therapies in sepsis, and to contributing to a reduction in the sex-dependent gap present in health care provision. The ethics committee for clinical research at our institution approved this study, considering that informed consent of the patients was not necessary due to the retrospective and non-interventional condition of the study.

Availability of data and material
All material used for this work are available.

Consent for publication
Not applicable

Competing interests
The authors declare that they have no competing interests.

Funding
This work was supported by Grant PIE1600050 SEXCOMPLEX from Instituto de Salud Carlos III, Plan Estatal de I+D+i 2013-2016) and co nanced by the European Development Regional Fund "A way to achieve Europe" (ERDF). All authors report no con ict of interest relevant to this article. MP-A was supported by a Rio Hortega contract (CM19/00069) from the Instituto de Salud Carlos III.
Author's contributions MPA and BMFF designed the work and made the acquisition, analysis, and interpretation of the data and draft of the work. RC and AM designed the work and revised it. AH, MRD, and AMSD made the acquisition, analysis and interpretation of the data. JZ and RdC designed the work, interpreted the data and wrote the manuscript. All authors read and approved the nal manuscript.