The patients for this prospective observational study were recruited from a 16-bed mixed intensive care unit between March 2017 and December 2018. Consecutive adult (more than 18 years of age) patients with sepsis admitted to the unit were screened within 24 h of admission. Patients with septic shock were eligible for inclusion. Septic shock was defined as an organ dysfunction related to an infection [8] and the presence of hypotension uncorrected by fluid resuscitation. This group was chosen because of the high incidence of visible mottling reported in previous studies [9]. Patients with extensive wounds or skin discoloration in the knee and thigh areas unrelated to mottling were excluded.
The study protocol was approved by the Institutional Research Ethics Committee (26/23.02.2017), and all participants or their closest relatives provided written informed consent.
Clinical management of patients
All clinical management was determined by local protocols. Clinical staff not involved in the study tailored therapy with fluids, vasopressors, and inotropes individually to maintain a mean arterial pressure of > 65 mmHg.
Study design
Visual evaluation and hyperspectral imaging of the skin around the knee area were performed when a mean arterial pressure of > 65 mmHg was achieved, and there had been no change in vasopressor requirements for at least 1 h.
The following clinical information was collected from the patients’ hospital records: demographic data (age and sex), primary site of infection, and clinical and laboratory data necessary to calculate Acute Physiology and Chronic Health Evaluation (APACHE) II [10] and Sequential Organ Failure Assessment (SOFA) [11] scores. Survival status was recorded at 28 days.
Mottling of the anterior aspect of the knee was assessed visually on both legs. Patients were placed supine with the legs straight and at the level of the heart. Mottling score (MS) which describes the extent of the mottled area on the knee and thigh was determined on a 6-point scale ranging from 0 to 5 as described previously [2]. If mottling was present, then the leg with more prominent mottling was chosen for scoring and imaging. Mean arterial pressure, doses of vasopressor agents, partial pressure of oxygen and lactate in the arterial blood, and core temperature at the time of imaging were also recorded.
Hyperspectral imaging equipment and measurement procedure
Hyperspectral images of the skin overlying the patella were acquired by a custom-built system consisting of multispectral camera Nuance EX (PerkinElmer, Alameda, USA) combined with a light source, both fixed on a tripod. An HSI data cube acquired from each patient contained 75 12-bit 1392 × 1024 pixel monochromatic images captured in the range of 450–820 nm with a step of 5 nm. All HSI cubes were saved as a set of lossless monochrome .tiff files. The processing of the HSI data was performed offline in semiautomatic mode using custom MATLAB (MathWorks, Natick, USA) code. The HSI image cube was divided into three clusters depending on the total hemoglobin concentration, obtaining precise skin regions with the highest pooling of blood. The oxygen saturation value at each pixel was calculated utilizing a three-layer optical model of the skin [12, 13] which infers the content of oxyhemoglobin, deoxyhemoglobin, and melanin from measured reflectance spectra. A full description of the image acquisition and analysis is provided in Additional file 1.
The mean values of microcirculatory blood oxygen saturation percentage (μHbSO2) and relative total hemoglobin concentration (μHbtot) in arbitrary units (a.u.) were calculated from the region of the imaged skin with the highest pooling of blood. The obtained values for each patient were used for further statistical analysis. Examples of the images obtained at different stages of HSI processing are shown in Fig. 1.
Biomarker analysis
Venous blood samples were collected from all patients in ethylene diamine tetraacetic acid tubes within 24 h of intensive care unit admission at the time of HSI. Within 30 min, the samples were centrifuged for 15 min at 1000 rpm, immediately aliquoted, frozen, and stored at − 80 °C until use. Plasma intracellular adhesion molecule-1 (ICAM-1), vascular cell adhesion molecule-1 (VCAM-1), and tissue plasminogen activator inhibitor-1 (PAI-1) were assayed by xMAP technology on a Luminex 200 analyzer (Luminex Corporation, Austin, USA) using a Milliplex MAP Human Sepsis Magnetic Bead Panel 1 kit (HSP1MAG-63K; Merck KGaA, Darmstadt, Germany). Plasma thrombomodulin was detected by the ELISA method using a human thrombomodulin/BDCA-3 Quantikine ELISA Kit (DTHBD0; R&D Systems, Inc., Minneapolis, USA). Plasma angiopoietin-2 was detected by ELISA using a human angiopoietin-2 Quantikine ELISA Kit (DANG20; R&D Systems, Inc., Minneapolis, USA). Plasma syndecan-1 was detected by ELISA using a human syndecan-1 ELISA kit (AB46506; Abcam plc., Cambridge, UK) and an Infinite®M200 analyzer (Tecan Trading AG, Switzerland). The intra-assay coefficient of variation was calculated after the analysis of two samples of low immunological marker concentration and two high concentration five times in a single assay run. The intra-assay coefficient of variation was 2.6%, which was in accordance with the manufacturer’s instructions.
Statistical methods
All data in this study were analyzed using R version 3.3.2 (The R Foundation for Statistical Computing, GNU General Public License, Boston, USA) with Hmisc, rms, ROCR, and ggpubr packages. Demographic, hemodynamic, microcirculatory, and biomarker data were expressed as median (interquartile range) or counts (percentages). Statistical comparisons between survivors and non-survivors were performed by the Wilcoxon-Mann-Whitney two-sample rank-sum test. Data regarding skin μHbSO2 and μHbtot across patients with different MS were compared using the Kruskal-Wallis test with post hoc Mann-Whitney analysis. The association between endothelial biomarkers and MS or μHbSO2 was assessed using Spearman’s rank correlation analysis. Binary logistic regression was used to estimate the odds ratios (OR) and 95% confidence intervals (CI) for the association between 28-day survival and MS or μHbSO2. Discrimination in univariate models was assessed by the C statistic. We used multivariate logistic regression to model 28-day survival as a function of MS, μHbSO2, μHbtot, and concentrations of thrombomodulin, angiopoietin-2, ICAM-1, VCAM-1, PAI-1, and syndecan-1. A p value of less than 0.05 was considered statistically significant.