on 04-Sep-2021 (Sat)

The exploration of peripheral perfusion through skin analysis has recently been highlighted and represents a promising clinical approach to study semiquantitatively tissue perfusion. The clinical ‘‘pale skin often covered with perspiration’’ dur- ing septic shock was described more than 50 years ago [7]. Altemeier [8] also reported in 1956 a moist and cool skin on septic patients with a bad prognosis.

The capillary refill time (CRT) is another interesting clinical parameter. CRT measures the time required to recolor the tip of a finger, usually the index

Septic shock, within 24 h after ICU admission, was defined by the 2001 international sepsis definitions conference [16]. Patients were included (H0) when vasopressor infusion was started (within 24 h of admission). Patients with dark skin were excluded because accurate clinical evaluation of CRT was not possible.

Capillary refill time was measured by applying a firm pressure to the distal phalanx of the index finger (or to the center of the knee) for 15 s. The pressure applied was just enough to remove the blood at the tip of the physician’s nail illus- trated by appearance of a thin white distal crescent (blanching) under the nail. A chronometer recorded the time for return of the normal color

repeated twice, and CRT was computed as the mean of the two successive readings

Inter-rater concordance was 80 % [73–86] for index CRT and 95 % [93–98] for knee CRT. The coef- ficient of variation of the index CRT was 10 % suggesting that limits of agreement should be defined as ±20 % of a reading. On the basis of this pilot study, the standard deviation of the CRT was approximately 2.5 s.

Macrocircula- tory parameters (MAP, cardiac index) were restored and not different at H6 (Table 2). However at H6, arterial lactate level was significantly higher (7.7 ± 5.8 vs 2.5 ± 1.4 mmol/l, P = 0.007) and urinary output lower (0.17 ± 0.27 vs 0.56 ± 0.64 ml/kg/h, P = 0.009) in non- survivors compared to survivors

CRT was recorded on the index finger and on the knee area at H6 and was significantly longer in non-survivors (respectively 5.6 ± 3.5 vs 2.3 ± 1.8 s, P \ 0.0001 for index CRT and 7.6 ± 4.6 vs 2.9 ± 1.7, P \ 0.0001 for knee CRT) (Table 2)

In univariable analysis, the OR of 14-day mortality significantly increased with CRT: index CRT (\2.5 s OR 1 (ref.), [2.5–5 s] OR 5.4 (1.3–22.3), [5 s OR 18.0 (3.6–89.6), P \ 0.001) and knee CRT (\2.5 s OR 1 (ref.), [2.5–5 s] OR 5.1 (0.5–51.3),[5sOR 61.2 (6.5–578.9), P \0.001).

In addition, adjusting on an overall severity index (SOFA score), a global hemody- namic parameter (MAP) and the dose of vasopressor did not substantially alter the association between CRT and death at day 14 (supplemental table)

At the bedside, we analyzed the prognosis of patients according to CRT changes between H0 and H6. The coefficient of variation of the index CRT was 10 %, leading to limits of agreement of 20 % around a mea- surement. Therefore, a change by 20 % between CRTs was considered of interest.

We did not observe any significant relation between CRT changes and cardiac index changes between H0 and H6 (P = 0.10, r = 0.22; supplemental figure)

A threshold of index CRT at 2.4 s predicted 14-day outcome with a sensitivity of 82 % (95 % CI [60–95]) and a specificity of 73 % (95 % CI [56–86]). A threshold of knee CRT at 4.9 s predicted 14-day outcome with a sensitivity of 82 % (95 % CI [60–95]) and a speci- ficity of 84 % (95 % CI [68–94]) (Figs. 2, 3).

The CRT at H6 was strongly predictive of 14-day mortality as the area under the curve was 84 % [75–94] for the index measurement and the AUC was 90 % [83–98] for the knee area

The CRT did not correlate with cardiac index and ScvO 2 .

The relationship between CRT and SOFA score remained significant when we removed the neurological (index P \ 0.0001, r = 0.53; knee P = 0.001, r = 0.41) or the hemodynamic parameter (index P \ 0.0001, r = 0.54; knee P = 0.001, r = 0.40) from the organ failure score

Regarding the index CRT a cutoff at 2.4 s was predictive of mortality but was different from the threshold at 4.5 s proposed by Schriger et al. [26] 25 years ago. However, in this study, the authors did not include critically ill patients with shock but compared only subjects before and after cold water immersion.

The temperature is a parameter that affects CRT. Unfortu- nately, we did not record skin temperature; but when we analyzed central temperature, we did not observe any difference between survivor and non-survivors. More- over, we did not find any relation between central temperature and CRT

Others factors could modify CRT such as vasopressors. At H6 we observed a significant relation between the vasopressor doses and the CRT measured on the index or the knee area (data not shown); but after adjustment based on vasopressor dose, the association between CRT and 14-day death remained significant.

The association of mortality with CRT was also unaffected by stratification on known arterial disease (defined as a previous vascular event, symptomatic or requiring therapeutic intervention). We recorded four (18 %) patients with a known vascular disease in the non- survivor group, and nine (24 %) in the survivor group.

Moreover, we observed a positive relation between CRT and the SOFA score; Lima et al. [14] also showed that patients with clinical signs of hypoperfusion had a significant lower SOFA improvement compared to patients with normal peripheral perfusion

CRT should not replace the mottling score [10], but could be used as a comple- mentary tool for several reasons. First, the mottling score is a semiquantitative parameter whereas CRT is a quan- titative parameter, leading to more accurate monitoring during resuscitation. Moreover, in the mottling group (0–1) and (2–3), knee CRT improved patient discrimi- nation according to their outcome, with non-survivors presenting a significantly higher knee CRT (data not shown)

Also, our conclusion was based on patients with severe septic shock and can probably not be extrapolated to patients with severe sepsis not requiring vasopressors

We did not include patients with dark skin because knee CRT could not be measured.

Accumulating evidence suggests that discordance between systemic hemody- namic parameters and microcirculatory alterations is more preeminent during shock [3].

Moreover, microcirculatory dysfunctions were identified as risk factors of morbidity and mortality, whereas systemic hemodynamic parame- ters were not [4].

In a mixed surgical population, cool skin temperature was associated with lower cardiac output, lower central venous saturation and higher lactate levels as opposed to warm skin temperature [9]

Mottling, an easy to assess clinical sign, is defined as patchy skin discoloration that usually starts around the knees. It is due to heterogenic small vessel vasocon- striction and is thought to reflect abnormal skin microperfusion

Mottling is widely described and taught in medical school as a sign of shock. More than 40 years ago, Vic-Dupont et al. [10] described clinical patterns of patients with septic shock and noted frequent mottling on the knees (65%).

Septic shock, within 24 h after ICU admission, was defined by the 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions conference [11]. Four patients with black skin were excluded because accurate clinical eval- uation of mottling was not possible.

Macrocirculation was assessed using MAP, heart rate (HR), CVP and the Cardiac Index. Microcir- culatory dysfunction and organ perfusion were assessed by arterial lactate levels, urinary output and mottling [14]

We quantified the extent of mottling on the legs on a 6-degree scale ranging from 0 to 5. The mottling score is based on mottling area extension on legs: score 0 indi- cates no mottling; score 1, a small mottling area (coin size) localized to the center of the knee; score 2, a mot- tling area that does not exceed the superior edge of the knee cap; score 3, a mottling area that does not exceed the middle thigh; score 4, a mottling area that does not go beyond the fold of the groin; score 5, an extremely severe mottling area that goes beyond the fold of the groin (Fig. 1)

The repro- ducibility of this score was excellent between observers [kappa 0.87, 95% CI (0.72–0.97)]

The MAP, CVP, HR and cardiac index at H6 were not significantly associated with 14-day mortality

Mottling score was the strongest predictor of mortality [score 0–1 OR 1; score 2–3 OR 16 (4–81); score 4–5 OR 74 (11–1,568), p \ 0.0001] (see Table 2). Fourteen-day mortality according to the H6 mottling score increased from 13% for a score of 0–1 to 70% for a score of 2–3 and 92% for a score of 4–5 (v 2 test for trend p \ 0.001).

We found that death occurred earlier in patients with a higher score (p \ 0.0001). Patients with a mottling score of 4–5 died mostly on day 1, whereas patients with a score of 2–3 died on day 2 and day 3 after admission (see Fig. 2)

At the bedside, we observed rapid changes of skin coloration during shock resuscitation, suggesting that the mottling score changes could be useful. We analyzed the prognosis of patients according to mottling score change between H0 and H6; patients with an initial score of 0–1 were excluded, and we focused on patients with initial moderate or extensive mottling (n = 38). Among the 13 patients whose score decreased, 10 survived (77%). Conversely, among the 25 whose score did not change or increased, only 3 survived (12%).

An increase in the mottling score was associated with increasing lactate levels (p \ 0.0001) and decreasing urinary output (p \ 0.0001). There was no such trend for the cardiac index according to mottling. Finally, the SOFA score also displayed a positive correlation with the level of mottling (p = 0.0002).

In a blind test performed on three independent intensivists, we demonstrated that agreement on the mottling score was excellent [Fleiss’kappa = 0.87 (0.72, 0.97)]

For the first time, we have shown that persistent extensive mottling, evaluated using a newly developed score, is a strong predictor of 14-day mortality. Moreover, the higher the mottling score was, the earlier the mortality occurred, suggesting a direct link among the initial event, severe infection and ICU mortality

Recently, Lima et al. [26] also reported a significant positive relation between clinical signs of hypoperfusion and arterial lactate level in an ICU pop- ulation. Furthermore, they showed that patients with clinical signs of hypoperfusion had significantly less SOFA improvement compared to patients with normal peripheral perfusion.

In an emergency department, others authors also reported that sublingual microcirculation improvement during the first hours of shock resuscitation was significantly associated with an improvement of SOFA score [27]

Several factors could modify clinical evaluation of mottling, such as vasopressors. However, after stratifica- tion of drug dosage [norepinephrine, high dose ([0.5 lg/ kg/min) and low dose], the association of a high mottling score with mortality remained strong (p \ 0.0001). More precisely, in those receiving high doses ([0.5 lg/kg/min) at H6, survival was 72% when the mottling score was 0–1 and 0% in the others. In those receiving smaller doses, survival was 95% with the mottling score 0–1 and 55% otherwise.

Finally, the exact physiopathology of mottling and the link between mottling and microvascular dysfunction remained unproved, and further investiga- tions are necessary

Mottling is a red-violaceous discoloration of the skin due to blood flow reduction in small skin vessels [1]. This dynamic process involves vessels with high noradrenergic receptor density, such as the knee [1]. In mottled territories the use of laser-Doppler demonstrated a frank decrease in skin per- fusion [2] and near infrared spectroscopy (NIRS) showed a reduction in tissue oxygenation [3].

Recently, two studies reported the incidence of skin mottling over the knee in a subset of patients admitted for septic shock [3, 5]. In these studies, mottling was frequent, occurring in 46 and 70 % of the patients, respectively, and was associated with increased mortality at day-14 [3, 5]

Indeed, although non-survivors were more likely to have skin mottling, they also had higher severity scores. Therefore it is not clear if the association between skin mottling and mortality was dependent or not on disease severity as indicated by severity scores [3, 5]

Additionally, mottling may occur in patients with non-septic shock

We hypothesized (1) that skin mottling over the knee could be frequent in non-selected ICU patients, and (2) that skin mottling over the knee and its duration could be associated with in-ICU mortality

Skin mottling was defined as a red-violaceous discolor- ation of skin visually assessed over the knee by the nurses regardless of skin temperature or capillary refill time. In our ward, presence of skin mottling over the knee is usually and prospectively qualitatively assessed by nurses since more than 10 years every 1–4 h as all other vital parameters from admission until death or discharge. The extent of skin mottling was not assessed according to the recent mottling score [5].

Patients who exhibited at least one episode of skin mottling over the knee were included in the mottling group. In patients who had had several mottling episodes, we analyzed only the first episode of skin mottling to assess the impact of mottling duration on outcome

For all patients with at least one mottling episode, we additionally collected clinical characteristics at the onset and at the end of the mottling episode including heart rate (HR), mean arterial pressure (MAP), duration of the first mottling episode, urinary output at the mottling onset, and fluid challenge and urinary output throughout the episode

Neurological SOFA score at the onset of mottling was not included as numerous patients required sedation. Lactate level within the 12 h prior to mottling episode was recorded

As this time threshold was a recommended objective for resuscitation [8] and in accordance with the literature [5], a mottling episode was considered as long-lasting if it persisted over 6 h

The reliability of qualitative assessment of mottling in patients with non-extensive mottling (mottling score B2) was excellent as indicated by a Cohen’s Kappa of 0.87 (95 % CI, 0.63–1.0)

During their ICU stay, the 230 patients (29 %) who exhibited at least one mottling episode were older, displayed higher SAPS II and SOFA scores at admission, required more fre- quently mechanical ventilation, vasopressors, or renal replacement therapy, and had a higher in-ICU mortality than patients without mottling (Table 1; Fig. 1a). They were more likely to be admitted for cardiac arrest or shock.

Incidence of mottling was 43 % (62/141) in patients admitted for shock, 49 % (32/65) in the subset of patients with septic shock, and 25 % (72/283) in those admitted for acute respiratory failure

Characteristics of patients at mottling onset are dis- played in Supplemental Table 1. Mottling occurred at admission in 65 % of the patients (150/230). At mottling onset, 60 % (138/230) of patients had a mean arterial pressure C65 mmHg without vasopressors. Median dura- tion of the mottling episode was 7.0 h (4.0–12.0) following exclusion of 3 patients in whom mottling episode duration could not be assessed due to mottling persistence at time of discharge.

Using an a priori definition [3, 5], 134 patients (59 %) had a persistent mottling episode for more than 6 h. They had higher SOFA and SAPS II scores, and received more frequently vasopressors at mottling onset than patients with shorter skin mottling (Table 2)

Between the onset and the end of skin mottling, heart rate and mean arterial pressure significantly dropped from 100 ± 24 to 95 ± 20 bpm, and from 81 [69–95] to 77 mmHg [69–86], respectively (p \ 0.005 for both).

Using multivari- ate analysis, the occurrence of at least one skin mottling episode over the knee was associated with in-ICU mor- tality independently from SAPS II (Table 3)

In the mottling group, the in-ICU mortality increased according to the mottling duration (Figs. 1b, 2). However, the in-ICU mortality of patients with skin mottling at admission was not significantly different from that for those in whom skin mottling occurred later in their ICU stay: 27 % (41 of 150 patients) vs. 36 % (29 of 80 patients), respectively, p = 0.18 (Supplemental Figure 1)

Using multivariate analysis, the persistence of skin mot- tling was associated with in-ICU mortality independently from the use of vasopressors or mechanical ventilation at mottling onset, and from hyperlactatemia before mottling onset (Table 3)

This study is the largest to investigate skin mottling in a non-selected cohort of critically ill patients. Skin mottling occurred in 29 % of the overall population

The incidence of skin mottling had previously been reported only in patients with septic shock in two small observational studies from the same group and was 46 [3] and 70 % [5] which was close to the 49 % in the subset of patients admitted for septic shock in our study.

In a large con- trolled clinical trial including patients with acute respiratory distress syndrome [10], skin mottling was reported in only 2.5 % which was in marked contrast the 25 % incidence in our patients admitted for acute respiratory failure. This major discrepancy can probably be explained by the difference between randomized controlled trials including highly-selected patients and cohort studies such as ours without exclusion criteria

Importantly, in 60 % of cases reported, skin mottling occurred while MAP was greater than 65 mmHg without the need of vasopressors. Such a discrepancy between normal mean blood pressure values and altered peripheral tissue perfusion has already been pointed out using mon- itoring devices such as sublingual laser-Doppler [11–13], peripheral perfusion index [14–18] or NIRS [15, 17, 19, 20], or using other clinical signs such as capillary refill time [14–18, 20, 21] or skin temperature gradient [14–18]

However, we describe for the first time that the occurrence and the persistence of skin mottling over the knee are independently associated with in-ICU mortality. A possible explanation for this association is that mottling may be an early severity sign as in the course of menin- gococcal disease in children [22]. Therefore, the occurrence of mottling may preclude the clinical or bio- logical signs of organ failure used to calculate severity scores such as SAPS II score. Moreover, skin mottling is associated with skin hypoperfusion [2], the largest organ of the body [23] and may lead to a ‘microcirculatory and mitochondrial distress syndrome’ through various mech- anisms such as the production of nitric oxide, endothelial cell damage or leukocyte activation [24–29]

However, adjustment to disease severity have led to con- flicting results [11, 12, 42] and all of these peripheral perfusion parameters have major limitations

First, thresholds defining normal capillary refill time [18, 21], peripheral perfusion index [15, 16] or skin temperature gradients [14, 17] are conflicting in the literature.

Third, blood lactate level is not continuously monitored and hyperlac- tatemia involves mechanisms other than tissue hypoxia [45]

We suggest that even in normotensive patients, clinicians should be aware that patients with persistent skin mottling over the knee may have an increased risk of in-ICU mortality.

Although mottling assessment was not possible in patients with dark skin, these patients were not excluded from the analysis given the retrospective design

Although higher scores and their persistence have been associated with poorer prognosis [3, 5], in these two studies mottling score was exclusively assessed by physi- cians at admission and 6 h later. In our study, occurrence of skin mottling was delayed after ICU admission in 35 % of patients. Consequently, the daily skin mottling assessment by nurses during the ICU stay seems more appropriate to screen skin mottling incidence in a large population of critically ill patients. Despite this limitation, our qualitative mottling assessment was significantly associated with in- ICU mortality.

It could be argued that only extensive mot- tling in the more severe patients was detected by our nurses. And so, we had to ascertain the reliability of their detection of patients with the lowest mottling scores, and we indeed found their assessments to be highly reliable, even in patients with low-extent mottling. This type of qualitative measurement consequently seems appropriate in prediction of prognosis

Third, given the retrospective nature of our analysis, we cannot exclude that occurrence or persistence of skin mottling would have led to therapeutic interventions by the physicians. However the effects of vasopressors [46, 47], dobutamine [48, 49], fluid therapy [50–52], or nitro- glycerin [13, 17, 53] on microcirculation are debated in the literature. Furthermore, to our knowledge, there are no data suggesting a relationship between peripheral perfusion improvement and mortality

First, it demonstrated the heterogeneity and the complexity of microcirculatory disorders in response to bacterial invasion, such as modi - fications in vasomotor tone, activation of the coagulation cascade and increased platelet–leucocyte interactions [1].

Second, it demonstrated the discrepancy between global hemodynamic parameters (such as arterial blood pressure) and microcirculatory blood flow [2, 3]. Studies also have reported a decrease in microcirculatory blood flow in patients with septic shock [4], even when global hemodynamic parameters seemed adequate [5]

Moreo - ver, microcirculatory alterations have been considered to be a stronger determinant of outcome than global hemo - dynamic parameters [6]. These accumulating pathophysi- ological and epidemiological evidences explained why the last conference of European experts excluded arterial blood pressure from the definition of shock but stressed the identification of microvascular alterations through the detection of tissue hypoperfusion [7]

In a cohort of patients who required ICU admission for trauma, sepsis and surgical complications, Kaplan et al. [12] reported that patients with cold extremities, evalu - ated in a subjective manner, had a higher arterial lac- tate level and a lower SVO 2 than patients with warm extremities.

Fifty years ago, Joly et al. [13] studied toe- to-room temperature gradient in a mixed ICU popula - tion (patients with cardiogenic, hypovolemic or septic shock and shock due to poisoning). They reported a significantly lower toe-to-room temperature gradient in ICU non-survivors than in ICU survivors

Dur - ing a 10-month period, we included all adult patients (≥18 years of age) who required ICU admission for a severe sepsis or septic shock (according to the 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference) from any causes [14].

Patients with hypothermia (defined as central temperature <H0

In patients with sep - tic shock, intravenous volume expansion was provided to achieve predefined endpoints: pulse pressure varia - tion </h

Toe and knee skin temperature was measured using a skin temperature sensor (STS-400 Level1 ® , Smiths Medi- cal, Rockland, MA, USA) applied at patient inclusion and remained for the next 24 h. Central body tempera - ture was measured with an electronic rectal thermometer (SureTemp ® Plus 692 Electronic Thermometer, Welch Allyn ® ), and room temperature was also recorded. The four temperature gradients (°C) were calculated as fol - lows: central-to-toe (central T–toe T), central-to-knee (central T–knee T), toe-to-room (toe T–room T) and knee-to-room (knee T–room T)

All patients with septic shock received vasopressor therapy, but four of them were weaned in the first 6 h after inclusion. The vasopressor mainly used was norepinephrine (at H6, 58/59 patients, median dose 0.5 [0.2; 0.9] µg/kg/min) and one patient received epineph - rine (at H6, 0.9 µg/kg/min). Thirteen patients had dark skin leaving 90 patients for the evaluation of mottling score and knee CRT

Of the 63 patients with septic shock, 20 patients (20/63, 32 %) died within the first days from multiple organ fail - ure and shock (deaths from MOF group). Forty-three patients (43/63, 68 %) survived the episode of initial sep - tic shock and were weaned from vasopressors, but 11 of them died later from various causes (11/63, 17 %, late deaths group, Additional file 1: Figure S1). Those patients died from acute myocardial infarction (n = 2), stroke (n = 1), bleeding (n = 2), pulmonary aspiration (n = 1), secondary infections (n = 2) and following a withdrawal of life-sustaining therapies (n = 3). Thirty-three patients were discharged alive from ICU (33/63, 52 %).

In the severe sepsis group, two patients developed a shock after 24 h of inclusion and died from multiple organ fail - ure. Six patients died later (28 [18; 44] days) from others causes (Additional file 1: Figure S1)

After initial resuscita - tion, mean arterial blood pressure and cardiac index were not different between the severe sepsis and the sep - tic shock groups. However, tissue perfusion parameters were significantly different between groups (Table 2).

Compared to patients with severe sepsis, patients with septic shock had a significantly lower urine output, higher arterial lactate level, higher knee CRT and a larger mottling score. Central-to-knee and knee-to-room tem - perature gradients were not different between groups. However, central-to-toe temperature gradient was sig - nificantly higher (12.5 [9.2; 13.8] vs 6.9 [3.4; 12.0] °C, P < 0.001) and toe-to-room significantly lower (1.2 [−0.3; 5.2] vs 6.0 [0.6; 9.5] °C, P < 0.001) in patients with septic shock (Fig. 1)

However, central-to-toe temperature gradient was sig - nificantly larger and knee-to-room and toe-to-room tem- perature gradients were significantly lower in the deaths from MOF group in comparison with survivors. The dif - ference was more pronounced for toe-to-room gradient (−0.2 [−1.1; +1.3] for death from MOF group and +3.9 [+0.5; +7.2] °C for survivors, P < 0.001).

The toe-to-room temperature gradient was predictive of death due to MOF at H6 with an area under the curve (AUC) of 0.76 [0.65; 0.86]. Pre - dictive value increased over time, at H12 the AUC was 0.83 [0.71; 0.95] and at H24 it reached 0.84 [0.74; 0.94]. At H24, a threshold of toe-to-room temperature gradient of 1.75 °C was predictive of death from MOF with a sen - sitivity of 75 % (CI 95 %, 53; 98) and a specificity of 75 % (CI 95 %, 62; 85)

The toe-to-room temperature gradient did not correlate with cardiac index but correlated weakly with vasopressor doses (r = −0.20, P = 0.05). In contrast, we observed a significant relationship between toe-to- room temperature gradient and tissue perfusion variables such as arterial lactate level (r = −0.54, P < 0.0001), urine output (r = 0.37, P = 0.0002), knee CRT (r = −0.42, P < 0.0001) and mottling score (P = 0.001) (Fig. 2).

In this prospective observational study of 103 critically ill patients with severe infections, central-to-toe and toe- to-room temperature gradients were statistically differ - ent between patients with severe sepsis and patients with septic shock

The toe-to-room temperature gradient showed the largest difference between survivors and patients dead from MOF.

Finally, the toe-to-room temperature gradient sig- nificantly correlated with tissue perfusion markers such as urine output, arterial lactate level, knee CRT and mot - tling score.

Kaplan et al. [12] subjectively described in a mixed surgical ICU population that patients with cold extremities had higher arterial lactate levels

Lima et al. [21] proposed a subjective combined approach of tissue perfusion and defined an abnormal peripheral perfusion if the examined extremity had an increase in index CRT or if it was cool to the examiner hands.

Central-to-toe tem - perature differences that we observed were in the same range than data reported in the Lima’s study with a median central-to-toe temperature gradient of 12.5 [9.2; 13.8] °C in the septic shock group and 6.9 [3.4; 12.0] °C in patients with severe sepsis.

In a small population of patients with severe sepsis and septic shock, Hernandez et al. recently described that an early recovery of both CRT and central-to-toe temperature gradient at H6 was predictive of lactate nor - malization at H24 [22].

The tem - perature gradient decreased in patients dead from MOF during the first 24 h of management, whereas it increased in survivors and patients in the late death group. In non- selected ICU patients with acute circulatory failure, Hen - ning et al. [23] also observed an increase in toe-to-room temperature gradient to more than 4 °C in survivors, whereas this gradient did not reach 3 °C over an interval of 12 h in non-survivors.

There was no relationship between toe-to-room temperature gradient and cardiac index.

Joly et al. [13] found a significant correlation (r = 0.73, P < 0.01) between the temperature gradient and cardiac output in a mixed ICU population, but another group did not. Vincent et al. [27] reported a good correlation in patients with cardiogenic shock but not in patients with septic shock.

We assessed whether confounding factors could affect temperature gradients. We observed a significant but weak correlation between toe-to-room temperature gradient and vasopressor doses in a pooled analysis of patient with severe sepsis and septic shock. However, when analysing only patients with septic shock, this relationship was not significant.

The association between MOF-related mortality and toe-to-room gradient tem - perature measured at H6 was unaffected by stratification on known arterial disease (defined as a previous vascular event, symptomatic or requiring therapeutic interven - tion) or by stratification on room temperature [28]. It is noteworthy that we applied no exclusion criteria in order to be as close as possible to the “real life”, and to iden - tify a parameter that could be widely used in critically ill patients.

Some of the patients included suffered cardiovascular disease or cir - rhosis, both conditions known to impair vascular reac- tivity and peripheral perfusion [29].

Lima et al. [30] described that nitroglycerin infusion improved forearm-to-fingertip temperature gradient in few patients with acute circulatory failure from 3.3 ± 0.7 to 0.7 ± 0.6 °C (P < 0.05). Van Genderen et al. [31], in a proof-of-concept study including 30 septic shock patients, reported that a therapeutic strategy based on peripheral perfusion evaluation, including temperature gradient, led to a trend towards less fluid infusion com - pared with conventional regimen

In this study, we investigated the predictive value of temperature gradients, but we did not analyse the under - lying mechanisms leading to gradient changes according to outcome. In the context of severe infection, sympa - thetic activation and endothelial dysfunction could both participate to impairment of distal blood flow and in fine to extremities temperature changes.

In a meta-analysis of 252 patients, De Backer et al. [2] showed that microcirculatory perfusion alterations predict mortality during serious infections, whereas mean arterial pressure or cardiac output did not.

In another study, modulating mean arterial pressure by increasing norepinephrine dose had variable unpredict - able effects on microcirculatory flow, which occasionally worsened [5, 6]

This dissociation between macro- and microcirculatory compartments, defined by Ince as «a loss of hemodynamic coherence» [7], brings a need to assess end organs tissue perfusion in patients with septic shock and to develop tools to analyze microcirculatory blood flow [8]

The direct identification of severe micro - circulatory alterations remains difficult at bedside. Tra- ditional markers of tissue perfusion may not be readily available (lactate) or may take time to assess (urine out - put).

As a visible and easily accessible organ, the skin allows simple observation of local microcirculatory perfusion through skin temperature alterations (skin temperature gradient), perfusion (capillary refill time) and color (mottling). The pathophysiology of these clini - cal disorders has not been investigated in depth, but sev- eral authors assume that the main driven mechanism of reduced blood flow is local vasoconstriction mediated by sympathetic neuroactivation [8]. Additional mechanisms could participate to impair microvascular blood flow (Fig. 1) [9, 10] such as local endothelial dysfunction [11, 12] (Fig. 2), leukocyte adhesion, platelet activation and fibrin deposition [13].

Subjective assessment of peripheral skin temperature may be a valuable tool in the evaluation of patients with septic shock. Eighty years ago, Ebert et al. [17] described the skin of septic shock patients as being «pale, often sweaty». Altemeier et al. [18] then noticed that a moist and cold skin was a factor of worse prognosis in patients with septic shock

Temperature gradients do not correlate with cardiac output [22, 25, 26] but are predictive of both organ failure severity and worse out - come.

Normalization of central-peripheral temperature gradients (< 7 °C) within the 6 first hours of resuscitation predicted correction of hyperlactatemia in septic shock patients [27]

The CRT gives important information on skin per - fusion and microcirculatory status but does not reflect cardiac output [25, 29]. Visual measurement of CRT associated with other clinical signs (tachycardia, mucosal dryness, etc.) helps to diagnose dehydration in children [30]

Inter-rater variability of CRT was weak in non-trained physicians [33], but is better in centers expert in tissue perfusion evaluation [34], espe - cially in the knee area [35]. Standardization of finger-tip pressure (i.e., How long? How strong the applied pres - sure?) might improve CRT reproducibility. Ait-Oufella et al. [35] obtained good inter-rater concordance by “applying a firm pressure for 15 s. The pressure applied was just enough to remove the blood at the finger tip of the physician’s nail illustrated by appearance of a thin white distal crescent (blanching) under the nail.”

Capillary refill time measurement correlates with the pulsatility index, a surrogate ultrasound-derived param - eter that reflects vascular tone of visceral organs in septic shock patients [36].

In the intensive care unit, Lima et al. [21] reported an association between a prolonged CRT (> 4.5 s on the index finger) and hyper - lactatemia and a higher SOFA score. In septic shock patients, a prolonged CRT 6 h after resuscitation has been shown to be predictive of 14-day mortality, with an Area Under Curve (AUC) of 84% for a measure on the index finger, and 90% for a measure on the knee

A 2.4-second threshold value on the index finger predicted mortality with an 82% sensitivity (95% CI [60–95]) and a 73% specificity (95% CI [56–86]). On the knee, a thresh - old value of 4.9 s predicted 14-day mortality with an 82% sensitivity (95% CI [60–95]) and an 84% specificity (95% CI [68–94]) [35]

Overall, when used as a qualitative variable (prolonged or not), CRT is a reliable triage tool to identify critically ill patients at risk of negative outcome. Quantitative measurement of CRT should be mainly used by trained physicians in patients with non-dark skin (Table 1, Fig. 3)

Mottling, a characteristic discoloration of the skin fol- lowing reduced skin blood flow [9], is taught as a marker of shock, but its clinical relevance has been poorly inves - tigated until recent years. A significant relationship between mottling extension and visceral organ vascular tone has been reported suggesting that mottling could reflect gut, liver spleen and kidney hypoperfusion [36]

In a study including septic shock patients, the mottling score at 6 h after resuscitation was predictive of death at day 14 (odds radio [OR] 16, CI 95% 4–81, for stages 2–3; vs 74, CI 95% 11–1568, for stages 4–5). Mortality occurred within 12–24 h for stages 4–5, within 24–72 h for stages 2–3 and later than 72 h for the rare deaths for stages 0–1 (Kaplan–Meier charts, p < 0.0001). In the same study, cardiac output and blood pressure were not associated with mortality at day 14, confirming the disparity between microcirculatory and macrocircula - tory parameters [37]

Another South American study confirmed these results in septic shock patients. Mortality rate at day 28 was 100% when the mottling score was higher or equal to stage 4, 77% for stages 2 and 3, and 45% for stages 1 or lower [39]

Prognostic value of mottling was also reported in unselected ICU patients: Persis - tent (> 6 h) mottling extending over the knee (> stage 2) was an independent risk factor for mortality (OR 3.29, 95% CI 2.08–5.19; p < 0.0001) [40].

Finally, Preda et al. [41] found the good predictive value of the mottling score for mortality at day 28 in patients with sepsis not receiving vasopressors.

Peripheral perfusion index is defined as the difference between the pulsatile and non-pulsatile portion of pulse wave, measured by plethysmography. Peripheral perfu - sion index (PPI) gives information on peripheral vascular tonus by the pulsatility, decreasing in vasoconstriction and raising in vasodilation [42]. Peripheral perfusion index is an early predictor of central hypovolemia [43]

Despite some differences between micro and macrovas- cular compartments, it would be over-simplifying and possibly wrong to completely separate these two vascu - lar compartments. In the study by Ait-Oufella et al. [37] focusing on mottling, global hemodynamic improvement within the first hours following resuscitation, based on blood volume optimization and catecholamine use, was associated with mottling improvement. Patients whose mottling score improved through the first 6-hour resus - citation had a good prognosis, whereas those whose score was stable or even worsened had a poor progno - sis (14-day mortality: 23% vs 88%, p < 0.001).

Finger-tip CRT is also quickly normalized in septic shock patients within 2–6 h after resuscitation, whereas hyperlactatemia requires longer time to recover [27, 46].

The ongo- ing ANDROMEDA-SHOCK trial aims to compare two resuscitation strategies during the first hours of sepsis treatment on 28-day mortality, one based on CRT measurement and the other on arterial lactate clearance [48]

A «proof-of-concept» study has been done comparing a volume expansion strategy based on peripheral perfusion, clinical parameter assessment, to a classical strategy based on mean arterial pressure, central venous pressure and cardiac index. Peripheral perfusion was assessed through CRT, index-forearm tem - perature gradient, peripheral perfusion index, and StO2. The resuscitation strategy based on clinical tissue perfu - sion assessment led to a reduction in fluid therapy vol- ume in the first 72 h (7565 ± 982 mL vs. 10,028 ± 941 mL, p = 0.08) and to a reduction in hospital length of stay (16 [5–28] vs. 43 [8–45] days, p < 0.05) [49]. A task force of six international experts with extensive bedside experi - ence recently proposed to integrate peripheral tissue perfusion tools in risk stratification and management of septic patients in resource-limited intensive care units, especially CRT, mottling score and temperature gradients [50]

As bedside evaluation of tissue perfusion using the skin improves risk stratification in patients with sepsis, there is a possibility that it could be used as a tool to guide resuscitation. Lavillegrand et al. [51] reported that a mild arterial hypotension (MAP between 55 and 65 mmHg) could be safely tolerated in patients without any sign of hypoperfusion. Such «personalized» management requires close monitoring (in an ICU) but may decrease the use of invasive devices and vasopressors, both having potential side effects

For example, nitroglyc - erin infusion had no beneficial effect in unselected sep- sis patients [52] but improved peripheral perfusion in selected patients with prolonged CRT and/or increased finger-tip-to-forearm skin gradient temperatures [53]. Ilomedin has been also recently proposed as a rescue therapy in sepsis shock with refractory tissue hypoper - fusion [54] and will be tested soon in a prospective ran- domized multicenter trial (I-MICRO NCT03788837)

Bloodstream infections (BSIs) represent a growing public health concern, with an estimated burden of 1,200,000 episodes of BSI each year in Europe, and 157,000 attributable deaths (Goto and Al Hasan, 2013)

We focused on the impact of BC collection strategies on their performance for the diagnosis of BSI, as this has not been a major focus in most recent reviews (e.g., Kirn and Weinstein, 2013; Garcia et al., 2015)

Published guidelines do not clearly st at e when BCs should be ordered (Baron et al., 2013). Blood cultures are commonly collected when patients have fever, chills, leukocytosis, septic shock, suspected endocarditis or prior to starting antimicrobial treatment in elderly or immunocompromised patients

Physicians significantly overestimate the likelihood of BSI for their pat ients ( Poses and Anthony, 1991). Indeed, in most settings, only 5 to 13% of BCs will turn out to be positive, and of those, 20–56% represent contaminants (B ates et al., 1990; Salluzzo and Reilly, 1991; Little et al., 1997; Dargère et al., 2014).

Despite progress in skin antisepsis since 25 years led to lower risk of contamination (e.g., 0.5–1%, Garcia et al., 2015), rates of contamination as high as 2.1–6% are still commonly reported, and the 20–56% proportion of contaminants holds true today (e.g., Zwang and Albert, 2006; Gander et al., 2009; Roth et al., 2010; Dargère et al., 2014).

Many models for predicting bacteremia have been developed but not all were validated and, when they were, the validation processes were highly heterogeneous ( Bates et al., 1990; Shapiro et al., 2008; Coburn et al., 2012). Eliakim-Raz et al. (2015) identified studies that underwent validation on prediction of bacteremia and that were able to define groups with low (<%) or high (>30%) probabilities of bacteremia in adults. They demonstrated that few studies have been prospectively validated, populations and parameters included were heterogeneous and none of these models were implemented in clinical practice (Eliakim-Raz et al., 2015).

Thus far, the most appropriate timing of BC collection has been poorly evaluated through clinical studies. Most guidelines state that blood spec imens should be collected in the absence of antimicrobials, at or around the time of fever spikes, and a 30–60 min interval between samples has been ar bitrarily recommended ( Weinstein, 1996). However, in a seven-center study that evaluated the timing of BC collection in relation to body temperature in 1436 patients with BSI, Riedel et al. (2008) could not identify any optimal timing for BC collection. Indeed, the likelihood of documenting BSIs was not significantly enhanced by collecting blood during temperature spikes. Yields were similar over a 24-h period before and after temperature spikes (Riedel et al., 2008). Consistent with these results, Li et al. (1994) found no difference in BCs yield whether samples were collected within a 24-h period, either simultaneously or serially (Li et al., 1994)

In a systematic review, Malani et al. (2007) found that: (i) alcoholic iodine tincture is more effective than aqueous povidone-iodine (PVI) to reduce t he risk of BC contamination; (ii) alcoholic chlorhexidine gluconate (CHG) is more effective than aqueous PVI; (iii) alcoholic antiseptics are more effective than aqueous products, and (iv) that alcohol alone is not inferior to any iodine products ( Malani et al., 2007).

A recent randomized crossover trial comparing the effectiveness of 3 sk in antiseptic interventions—10% aqueous PVI solution, 2% iodine tincture (IT), and 2% CHG in 70% isopropyl alcohol—before BCs sampling, found that the choice of antiseptic agent does not impact contamination rates when BCs are sampled by a dedicated phlebotomy team (Washer et al., 2013)

In addition, it has been demonstrated that BCs cont aminat ion rates are significantly lower when an antiseptic agent is applied on B C bottle tops before sampling (Schifman et al., 1998). European and French guidelines recommend the use of an alcoholic solution for antisepsis before BC sampling (Lamy and Seifert, 2012; Accoceberry et al., 2015a)

For an extensive review on skin a ntisepsis before BC collection and prepackaged kits performance, see Garcia et al. (2015).

Several studies concluded that peripheral venipuncture is the method of choice for BC collection, as compared with sampling through intravenous catheter, with rates of contamination ranging from 1.2 t o 7.3% when samples are obtained from venipuncture compared with 3.4 to 13% when blood is drawn through catheter (Dawson, 2014; Garcia et al., 2015). Indeed, both colonization of the catheter, and breakdowns in sterile procedure increase the risk of BCs contamination when BCs samples are drawn through these devices (Bates et al., 1991)

Although studies have shown that contamination rates are lower for BCs drawn from newly inserted catheter using a sterile technique protocol ( Levin et al., 2013), BCs are usually even less contaminated when samples are drawn by peripheral venipuncture (Snyder et al., 2012)

Impact of Volume Sampled on BCs Yield Data available could be summarized as follows: “The higher the blood volume cultured, the higher the yield.” Indeed, adequate volume sampling is the most important parameter for the detection of bloodstream microorganisms because bacterial or fungal density in blood is very low in most patients with BSI

Basically, the likelihood of detecting a BSI depends on th e bacterial or fungal concentration, and on the volume collected. Using theoretical models, Arpi et al. (1989) estimated the average bacterial concentration in patients with bacterial BSI at 0.25 colony-forming unit (CFU) per milliliter, and Jonsson et al. (1993) demonstrated that the bacterial concentration was less than 0.04 CFU/mL in 29% of Escherichia coli BSI and 18% of Staphylococcus aureus BSI (Arpi et al., 1989; Jonsson et al., 1993)

The sensitivity of BC was estimated to be 95% when 3 CFU are sampled, which implies that at least 30 mL of blood are incubated (Jonsson et al., 1993). The results of this model are in full agreement with Washington’s empirical data obtained 18 years earlier, showing that a total volume of at least 30 ml of blood is required for detecting 99% of BSI ( Washington, 1975).

Overall, data from modeling, as well as clinical studies are remarkably concordant: 50% of BSI episodes are associated with a bacterial concentration in th e range of 0.01–1 CFU/mL (Tables 1, 2)

Comparing the yield of standard-volume BC (mean, 8.7 mL), and low-volume BC (mean, 2.7 mL), a study from the Wisconsin Hospital and Clinics demonstrated that the sensitivity of BCs for the diagnosis of BSI was 92% with standard-volume, and 69% with low-volume (difference of 23%, [95% CI, 9–37%]; P< 0, 001; Mermel and Maki, 1993).

Examining 7783 BCs, including 624 classified as true positive BCs, Li et al. (1994) demonstrated that increasing the volume of blood cultured from 20 to 40 mL increased the yield by 19%, with an additional gain of 10% when the incubated volume was increased from 40 to 60 mL (Li et al., 1994)

The most recent studies showed that sampling blood volumes of 20, 40, and 60 mL was associated with sensitivities of 65.0–75.7%, 80.4–89.2%, and 95.7–97.7%, respectively (Cockerill et al., 2004; Bouza et al., 2007; Lee et al., 2007; Patel et al., 2011)

The French Society of Microbiology recommends that, in patients suspected of BSI, 4–6 bottles (2 to 3 sets) of blood should be cultured, with adequate volume for each bottle. A 6 bottles-procedure is necessary whenever the optimal filling of all the bottles is not ensured (Accoceberry et al., 2015a)

The American College of Critical Care Medicine (ACCM) and Infectious Dise ases Society of America (IDSA) Guidelines recommend that new fever in critically ill adult patients must be investigated by drawing of 3–4 B C sets with appropriate volume within the 24 h of fever onset ( O’Grady et al., 2008)

Generally, in adults with a suspicion of BSI, 2–4 BC sets should be obtained in the evaluation of each septic episode ( Baron et al., 2013; Dellinger et al., 2013 )

For optimal recovery, each BC set should include paired aerobic and anaerobic bottles, the aerobic bottle being filled first. Besides, the culture of only 2 bottles (1 set) during a 24-h period from adult patients (hereafter referred to as “solit a ry BC”), is discouraged in all guidelines, as the sensitivity of only 2 bottles is insufficient

In addition, it must be taken into account that, in the real life, a significant proportion of BCs bottles are not adequately filled (Vitrat-Hincky et al., 2011; Willems et al., 2012; Lin et al., 2013; van Ingen et al., 2013; Coorevits and Van den Abeele, 2015)

Firstly, BC bottles that should contain 8–10 mL of blood (manufacturer instructions) are frequently under-filled, consequently le ading to sets of much less than 20 mL, or— although less commonly—over-filled (>10 mL). Mermel and Maki (1993) first highlighted the issue of under-filled BC bottles, a situation that was not center-specific, since 88% of 71 U.S. laboratories acknowledged that they routinely receive BC specimens from adults containing less than 5 mL of blood (Mermel and Maki, 1993).

On the ot her hand, BC bottles are inoculated with more than 10 mL of blood in 7.6–13% of cases (Table 3). These bottles are at increased risk to be falsely flagged positive by the BC system ( Wilson et al., 1994; Reimer et al., 1997). According to Willems et al. (2012), both Becton Dickinson R (BD) and bioMérieux R admitted that the vacuum in the BC bottles substantially exceeds the optimal blood fill volume (10 mL; Willems et al., 2012). The purpose of the vacuum excess is to ensure a sufficiently long expiration date, and to minimize the colle ction time

Se veral studies reported rates of solitary BC per center between 10 and 33.3% (Table 4). Proportions of solitary BC seem to be lower in t he US, with medians of 26%, and 12.7% (Schifman et al., 1991; Novis et al., 2001), which may be related to the implementation of quality control programs that include this indicator, and to the fact that the personnel primarily responsible for BC collection are dedicated phlebotomists, two major differences with practices in most European countries.

Sampling an adequate volume of blood to ensure optimal sensitivity for BSI detection can be achieved either by incre asing the number of venipunctures (hereafter, “multi-sampling strategy”) or by colle cting the adequate large volume through one single puncture (hereafter, “single-sampling strategy”).