Emergency department non-invasive cardiac output study (EDNICO): an accuracy study

The aim of intravenous fluid therapy is to increase cardiac output and therefore to increase oxygen delivery to hypo-perfused organs. Resuscitation with inadequate intravenous fluid may risk inadequate organ oxygen delivery. Over resuscitation may precipitate iatrogenic tissue oedema leading to compartment syndromes and decreased oxygen delivery, with consequent increases in mortality [1‐3]. Early-stage fluid resuscitation is commonly guided by physiological parameters such as capillary refill time, pulse rate, and blood pressure. Biochemical parameters such as lactate and acid-base assessment may also be used to guide fluid doses [4]. Neither approach offers a sensitive or specific proxy for organ perfusion, so in-precisely guiding fluid dosing [5‐9].

Fluid responsiveness is commonly defined as a stroke volume increase of at least 10% following a fluid bolus of 200-500mls delivered over 10–15 min [10]. Around one to two thirds of emergency department (ED) patients thought to require fluid resuscitation are not fluid responsive and risk harm from aggressive fluid administration [11]. Identifying which patients respond to fluids (and continue to do so) has the potential to individualise fluid delivery and improve outcomes [12]. In intensive care units, invasive monitoring of cardiac output is frequently used to guide fluid administration, with methods including pulmonary artery catheterisation (PAC), arterial pulse pressure analysis or oesophageal Doppler [13]. However, these methods are invasive and unsuitable for routine monitoring in the ED [14, 15].

Non-invasive cardiac output monitoring methods are emerging in the ED and in the pre-hospital environment [16, 17]. These methods include common carotid artery blood flow monitoring (CCABF) [18], suprasternal aortic Doppler (SSAD) [19, 20], plethysmography using the vascular unloading technique (PVUT) [21, 22], and thoracic bioreactance [23‐25]. Inferior vena cava collapsibility index (IVCCI), while not measuring cardiac output, has been studied as an indicator of fluid responsiveness [25‐27]. Stroke volume measured by left ventricular outflow tract velocity time integral (LVOT VTI) [28, 29] is the most widely studied of these techniques, including validation against the PAC, which is widely referred to as the gold standard for cardiac output monitoring. LVOT VTI is widely accepted as a non-invasive method to assess stroke volume and cardiac output, and guide fluid therapy [30‐35]. We thus arbitrarily assigned the LVOT VTI as reference standard in the current study whist being aware of its limitations in terms of operator experience/skill.

The clinical value of a diagnostic test is assessed by a multi-phase process which includes assessing its feasibility, repeatability, accuracy, impact on patient outcomes and cost. We have previously reported on the feasibility and repeatability of the above methods in these same cohort of ED patients [36]. We report here the diagnostic accuracy of these techniques determined by the sensitivity and specificity of each in identifying fluid responders as identified by LVOT VTI.

Of 76 participants recruited in the primary study [36], 33 received 250-500mls of crystalloid over 15 min or less and were included for assessment of accuracy (Fig. 2). Reasons for exclusion included a prescribed change in fluid delivery rate by the clinical care team, interruption of fluid monitoring for patient transfer, and rate-limited fluid infusion equipment. 60.6% (20/33) of participants were identified as fluid responders (Table 1).

The measure of agreement between the reference standard (LVOT VTI) and test method to identify fluid responders was expressed as a kappa value (Table 2). A four-quadrant plot illustrates the concordance of directional change for each patient as measured by each test method as compared to LVOT VTI. An exclusion zone for small changes in stroke volume is applied to exclude changes of stroke volume of less than 10% (Fig. 3). To assess the accuracy of IVCCI in identifying fluid responders, the optimal IVCCI cut-off was 40%. A receiver operating characteristic (ROC) curve was plotted (Fig. 3f).

This study finds that non-invasive cardiac output methods have poor agreement in identifying fluid responders in spontaneously breathing ED patients.

We identified no studies defining the accuracy of bioreactance in assessing fluid responsiveness in spontaneously breathing ED patients. To date studies performed heterogenous results. Squara et al. reported good agreement between cardiac output as measured by bioreactance and PAC in 110 patients [24]. However, accuracy to identify fluid responders was poor when studied on a subset of 23 patients with an assortment of interventions (7 rapid infusion challenges, 6 dobutamine challenges, 6 high PEEP stops, and 4 adrenaline infusions). Marque et al. also compared bioreactance to PAC in a group 19 mechanically ventilated patients who underwent rapid fluid infusion or passive leg raise (PLR) [48]. Bioreactance was found to have a sensitivity of 91% and a specificity of 95% in identifying fluid responders (no confidence intervals reported). Similarly, Galarza et al. reported reasonable accuracy in identifying fluid responders amongst 32 ICU patients with a sensitivity of 92% (95% CI: 62–91%) and a specificity of 80% (95% CI: 56–94%) when using an upgraded bioreactance device against PAC [49]. However, Kupersztych-Hagege et al. reported that the AUROC for bioreactance when compared to PAC to identify fluid responders amongst 48 critically ill patients was not significantly different to 0.5 (p = 0.77; no confidence intervals reported) [50]. The study was criticised for its methodology [51, 52]. Similarly, in a cohort of 22 patients under anaesthesia prior to elective intra-abdominal surgery, Conway et al. reported a sensitivity of 75% and a specificity of 69% in identifying fluid responders against oesophageal Doppler monitoring (no confidence intervals reported) [53]. Lastly, De Pascale et al. reported a sensitivity of 80% (95% CI: 56.3–94.3) and specificity of 82.6% (95% CI: 68.6–92.2) in 21 patients under anaesthesia prior to elective pelvic surgery against oesophageal Doppler monitoring [54].

In the current study, PVUT showed no agreement with LVOT VTI in identifying fluid responders. Movement of the digit on which the cuff device is placed was found to delay calibration and interfere with PVUT measurements. Other studies have investigated PVUT in its accuracy in predicting fluid responsiveness through pulse pressure variation (PPV), rather than measuring stroke volume change. Biais et al. reported that in 35 patients under general anaesthesia, a baseline 15% PPV threshold measured through CNAP identified fluid responders with a sensitivity of 76% (95% CI: 53–92%) and a specificity of 93% (95% CI: 66–99%) [55]. Similarly, Monnet et al. reported a sensitvity of 82% (95% CI: 57–96%) and a specificity of 91% (95% CI: 71–99%) for CNAP-derived PPV against pulse pressure analysis in 39 critically ill mechanically ventilated patients [56].

Previous studies on CCABF were conducted on healthy volunteers [57, 58], ICU patients [18, 59‐61], and peri-operative patients [62, 63]. To our knowledge no studies have investigated the accuracy of SSAD in identifying fluid responders in spontaneously breathing patients in any clinical setting. Marik et al’s retrospective data review of 34 ICU patients found that CCABF identified fluid responders, using PLR as a surrogate for fluid loading, with a sensitivity of 94% and a specificity of 86% (no confidence levels or AUROC reported) [18]. 19 (56%) participants were mechanically ventilated, reducing variations in intra-thoracic pressure compared to spontaneous breathing potentially improving CCABF accuracy. However, Roehrig et al’s study of 33 post-operative mechanically ventilated patients reported an AUROC of 0.54 (95% CI: 0.32–0.76) using PAC as a reference method. The authors suggest that cerebral blood flow autoregulation possibly explains why CCABF did not reflect cardiac output changes after PLR [63]. PLR is a low-cost, easily accessible, and well validated alternative to an intravenous fluid bolus to assess fluid responsiveness. However, the latter is still first line practice in many UK EDs, and was therefore selected as the method of choice in this study.

Although SSAD is reported to reliably detect a 7.5% blood volume loss in healthy volunteers few studies have investigated its accuracy to detect fluid responsiveness [64, 65]. A meta-analysis on SSAD accuracy in mechanically ventilated patients by Chong and Peyton found an overall cardiac output percentage error of 42.7% (95% CI: 38.5–46.9%) in comparison to PAC [66]. The authors hypothesized that inaccuracy arises from operators obtaining a suboptimal angle of incidence on the direction of aortic blood flow.

For IVCCI accuracy, a recent systematic review by Seccombe at al identified six studies conducted in spontaneously breathing patients with sepsis. The authors reported that the high level of heterogeneity amongst study designs prevented pooling of results [67]. The lack of studies conducted in non-ICU settings, where most patients with sepsis are managed, also contributed to poor generalisability of results. The heterogeneity of techniques in measuring IVCCI has been previously discussed [36]. Four studies used LVOT VTI as the reference method, one used bioreactance, and one used systolic blood pressure. The study of Lanspa et al. was the only one based in the ED. The authors reported an IVCCI sensitivity of 100%, a specificity of 67% (no CI reported), and an AUROC of 0.83 (95% CI: 0.58–1.00) to detect fluid responders amongst 14 patients using LVOT VTI as the reference standard [26].

Judicious intravenous fluid delivery plays an important role in increasing oxygen delivery to hypoperfused organs and avoiding fluid overload. As discussed above, few studies have investigated non-invasive methods to identify fluid responsive states in spontaneously breathing patients. This group represents the majority of patients with sepsis who receive intravenous fluid therapy and consequently requires further study. Overall, reviewed studies report low to moderate accuracy across all methods.

This study showed that there was poor agreement between the non-invasive methods used to measure SV/CO (bioreactance, PVUT CCABF, LVOT VTI and SSAD) and LVOT VTI in identifying fluid responders. Multiple studies have shown suboptimal agreement between methods used to assess stroke volume/cardiac output. Our data should not be interpreted as seeing one device as offering more accurate measurements than another but that there is limited agreement between the devices used in this study. Further studies are required to assess the accuracy of these non-invasive haemodynamic methods in spontaneously breathing patients in the ED prior to conducting large outcome trials based on their use.