This study was approved by Cochin Hospital Ethics Committee (# CCPPRB 2061, Assistance Publique Hôpitaux de Paris). The mHLA-DR blood tests did not require the patient's informed consent since it was a retrospective study with routine measurements in our institution performed on the remaining routine blood samples, with a guarantee to use the data after their anonymization, according to the Ethical French law.
Study design and population
The cohort of patients was collected on the patients’ database admitted in our center between 2013 to 2015 having routine measurements of HLA-DR, to evaluate the meaning of mHLA-DR monitoring. The only criteria used to select the cohort was to have had at least one measurement of mHLA-DR performed within the first 3 post ICU admission days. Patients hospitalized for less than 3 days, moribund, or treated with a chronic treatment by immunosuppressive drugs were excluded. Based on the motif for admission, 4 main clusters of life-threatening conditions were observed: (1) sepsis, defined by the criteria of the American College of Chest Physicians/Society of Critical Care Medicine [14]; (2) neurologic disorders, related to acute brain injury such as hemorrhagic or ischemic stroke; isolated severe brain trauma; post-neurosurgery; (3) major surgery (abdominal, orthopedic, ENT); (4) miscellaneous etiologies including primary respiratory failure, hemorrhagic shock from gastrointestinal bleeding or obstetric emergencies. IAI was diagnosed using the classic definition [15, 16]: a new-onset infection starting at least 48 h after ICU admission, which motivated a new antimicrobial therapy. The likelihood of infection motivating the clinical decision to administer antibiotics was classified as none, possible, probable, and definite [15]. Details of the classification method are provided in the e-Method section of Additional file 1. At the time of secondary infection diagnosis, the in-charge physician was not aware of the mHLA-DR value. Two senior intensivists (first and last author) blindly reviewed all patients’ medical charts and adjudicated all secondary infections. In case of discordance, a third expert settled the final diagnosis (CD).
Circulating monocyte HLA-DR measurements (mHLA-DR)
The quantification of the expression of HLA-DR on monocytes was assessed using the number of antibodies per cell (AB/C) by flow cytometry (FACS Canto II instrument, FACS Diva software, Becton Dickinson, San Jose, CA USA) as previously described [7] (see detailed protocol in the e-method section of Additional file 1). In our center, the median and IQ range of mHLA-DR expression in healthy people for the measurements at the same period (n = 13) was a median log mHLA-DR value of 40,134 (IQR: 36,315–44,353). The first blood sample and measurement of mHLA-DR were performed within the first 3 days after admission. The 2nd measurement of mHLA-DR was obtained on fixed days (Monday or Thursday) until the patient’s discharge or death. Since we used a survival model, only measurements sampled before the event were considered for analyses, and data were blinded at the time of the event (death or IAI). We used the threshold of AB/C < 8000 to define “low mHLA DR” corresponding to the acquired immune suppression as previously reported (NCT02361528) and because it corresponded to the median value observed in previously published datasets [7, 12]. Only the first nosocomial infection was considered for analysis.
Statistical analysis
The data were described as number and percentage for categorical variables and median (interquartile range (IQR)) for continuous variables. Comparisons relied on the Fisher exact test or χ2 test for categorical data and the Kruskal–Wallis or Wilcoxon test for continuous data. Because of non-linearity, all the mHLA-DR values were log-transformed. Age and SOFA scores were categorized based on the median value. A p-value of less than 0.05 was considered statistically significant.
Standard survival analyses are affected by the time of onset of the event of interest. Patients who have not experienced the event at the end of follow-up were censored. To determine the risk of an event occurring at a certain time-point, a fundamental assumption is that such censoring is not associated with an altered chance of the event occurring at any given moment. In this study, the event of interest is the occurrence of nosocomial infections and followed up until day 28 or until leaving alive from ICU. Indeed, death and leaving alive from ICU are competing events since, by definition, extubation precludes the observation of a ventilator-associated infection [17]. For that purpose, the association between mHLA-DR measurements and outcomes was assessed using adjusted Fine and Gray subdistribution competing risk models [18]. We first took into account the competing ICU discharge for the subdistribution hazard of mHLA-DR measurements on death at day 28. The subdistribution hazard of mHLA-DR measurements on the occurrence of IAI at day 28 was made considering the competing ICU death and ICU discharge. For each model, risk factors for the different outcomes were first researched by univariate analyses. The covariates tested into the models were the following: age, motif of admission, SOFA on day 1, comorbid conditions, and immune suppression. Although parenteral nutrition and the use of a central venous catheter are usual risk factors for NI, they were excluded from the predictors of NI. Almost every patient was managed with a central venous catheter, and enteral nutrition only was given to our patients. Then, the variables yielding p-values < 0.2 in univariate analysis were entered into a multivariate model using a backward selection, with p < 0.05 considered significant. The mHLA-DR measurement was forced into all the models. Results were expressed as subdistribution hazard ratios (sHR) with their 95% confidence intervals (95% CIs).
To analyze other immune factors possibly associated with IAI, we performed the same analyses using lymphocyte count (with lymphopenia defined as lymphocyte count 1000/mm3) and monocyte count (with monocytopenia defined as monocyte count below 500/mm3).
We assessed the robustness of our findings using multiple sensitivity analyses. We performed internal validation using a bootstrapping procedure, which was done by taking a large number of samples of the original one. This technique provides nearly unbiased estimates of the confidence intervals (CI) of the odds ratio (OR) of the independent covariates. Second, we performed logistic regression sensitivity analysis. Third, we used a multivariate cause-specific survival model. Fourth, we analyzed previous immune suppression as a comorbid condition and included age into the model. Fifth, we provide sensitivity analyses focusing on documented secondary infections and more specifically for secondary infections occurring at least 48 h after the second mHLA-DR measurement. Similar testing was applied for the VAP. All analyses were performed using SAS software, version 9.4 (SAS Institute Inc., Cary, North Carolina).