Context-specific clinical applicability of the end-expiratory occlusion test to predict fluid responsiveness in mechanically ventilated patients

Abstract

BACKGROUND

The emergence of context-specific clinical evidence from the end-expiratory occlusion test (EEOT) may change the perception of its operative performance to predict fluid responsiveness.

OBJECTIVE(S)

Assessment of predictive performance of the EEOT in the intensive care unit (ICU) and operating room.

DESIGN

Systematic review of observational diagnostic test accuracy studies with meta-analysis.

DATA SOURCES

MEDLINE, Embase and Scopus were used as data sources for relevant publications until February 2024.

ELIGIBILITY CRITERIA

Prospective clinical studies in which the EEOT was used to predict fluid responsiveness in mechanically ventilated adults, regardless of the clinical care context. The operative performance characteristics must also have been reported.

RESULTS

Twenty-four studies involving 1073 adult patients (588 receiving intensive care and 485 in the operating room) were systematically reviewed, and 22 studies comprising 1049 volume expansions were meta-analysed. The pooled sensitivity [95% confidence interval (CI)] of the EEOT was 0.87 (0.81 to 0.92), and the pooled specificity was 0.90 (0.85 to 0.94); the median [interquartile range] cardiac index (CI) threshold for a positive test was a 5.0 [3.3 to 5.3] increase. The clinical context, the method used for haemodynamic monitoring, the ratio of the averaging time of the monitoring method to the occlusion time, the levels of positive end-expiratory pressure and the choice of cardiac output marker were identified as significant sources of heterogeneity. However, the occlusion duration and tidal volume did not significantly affect its performance. A novel insight is that performance was notably lower in the operating room setting. The likelihood ratios were 14 (positive) and 0.12 (negative) for the ICU, both better than 3.1 and 0.21 for the operating room. The overall quality of the evidence was assessed to be very low, mainly due to high heterogeneity and risk of bias; however, no publication bias was detected.

CONCLUSION

The EEOT for predicting fluid responsiveness in critical care performs acceptably well overall and is a confirmative test. In the operating room and/or with specific technical settings, its performance and clinical utility are reduced, driving the need for more context-specific and patient-specific fluid responsiveness assessments.

KEY POINTS

• This systematic review assessed the effectiveness of the end-expiratory occlusion test in predicting fluid responsiveness in the critically ill in the intensive care unit and during surgery in the operating room.

• The analysis revealed that the end-expiratory occlusion test generally demonstrates clinically acceptable sensitivity and specificity, and is a confirmative test in the intensive care unit.

• The performance of the test is statistically significantly better in the intensive care context compared with the operating room and with the averaging time of the monitoring method shorter than the occlusion time.

• As the clinical utility of end-expiratory occlusion test varies among patients and settings, the choice of fluid responsiveness assessment method should be tailored to the individual patient.

Introduction

Intravenous fluid administration is a cornerstone of the clinical management of haemodynamic instability during peri-operative care and in the critically ill. However, well informed tailoring of volume therapy for individual patients in different clinical contexts remains a challenge. Both hypovolaemia and hypervolaemia are associated with poor clinical outcomes.^1,2 Only about half of haemodynamically unstable patients increase cardiac output (CO) with fluid administration.^3–5

The commonly used passive leg raising (PLR) and pulse pressure variation (PPV) to assess fluid responsiveness have several limitations. For example, PLR is labour-intensive and it is not always possible to perform, especially peri-operatively.⁶ PPV is not reliable with low tidal volume ventilation, irregular heart rhythm or spontaneous breathing.⁷ Therefore, PPV is suitable only for a specific and limited subset of critically ill patients.⁸ These practical challenges and shortcomings have created an impetus for further development of predictive tests to improve patient-specific fluid management.

In mechanically ventilated patients, the end-expiratory occlusion test (EEOT) has been proposed as the most valid dynamic test, based on heart–lung interaction, which enables the prediction and monitoring of fluid responsiveness at the bedside in the intensive care unit (ICU) and operating room (OR).⁹ It is presumed to work regardless of breathing activity, cardiac rhythm, tidal volume or lung compliance.¹⁰ Between 2019 and 2023, five systematic reviews and meta-analyses compiled the available clinical evidence on the use of EEOT, including other dynamic tests to predict fluid responsiveness.^11–15 Overall, the discriminatory and operative performances in terms of sensitivity, specificity and positive-predictive and negative-predictive values appear clinically acceptable.

The most recent analysis from 2023 included 17 studies on EEOT, reporting an unadjusted sensitivity of 0.76 (0.70 to 0.81) and specificity of 0.77 (0.71 to 0.83), whereas a preceding analysis from 2020 reported values greater than 0.80 for both performance metrics.^12,13 A possible explanation may relate to the growing evidence for EEOT in the medical, surgical and postsurgical contexts and the use of different, less invasive haemodynamic monitoring technologies. An increasing body of clinical evidence usually modifies the boundaries of the operative performance of any diagnostic test over time, as was recently observed for EEOT. However, a detailed analysis of the literature is hampered by heterogeneous patient groups and different clinical contexts. Additionally, technological differences in CO monitoring may strongly limit proper clinical evaluations of EEOT.

Therefore, we conducted a systematic literature review and performed a meta-analysis of currently available clinical studies to assess the overall operative performance of EEOT for predicting fluid responsiveness, and the impact of clinical context-specific and detailed technical factors in both the ICU and OR.

Methods

This study represents a systematic literature review and meta-analysis of observational diagnostic accuracy studies. The manuscript was prepared in line with the Meta-analyses of Observational Studies in Epidemiology (MOOSE) reporting guidelines.¹⁶ The study protocol was prospectively registered in PROSPERO (ID: CRD42024514338).

Identification of relevant studies

The MEDLINE, Embase and Scopus databases were systematically searched for potentially relevant original full-text research articles in English with no restrictions on the publication period. The complete search strategy is shown in Supplemental Digital Content 1 (additional Tables and Figures) Table S1. Articles were eligible for inclusion if they consisted of prospective clinical studies involving mechanically ventilated human adult ICU or peri-operative patients. In these studies, EEOT was performed by holding the ventilatory cycle at the end of expiration and measuring the induced haemodynamic effect. No restrictions were placed on the duration of the hold or type of haemodynamic measurement. Actual fluid responsiveness had to be determined with an increase in measured CO or related haemodynamic marker following an (autologous) volume expansion. The discriminatory ability of EEOT to identify fluid responsiveness was required as one of the study outcomes.

Screening was conducted using the web-based tool Rayyan (2022; Rayyan Systems, Inc., Cambridge, Massachusetts, USA), where duplicates were automatically detected and removed. Screening of the title and abstract first, followed by full text review, was conducted by two independent experienced researchers. Conflicts related to the inclusion or exclusion of studies were resolved through consensus with a third researcher. All relevant studies were included in the systematic review but only those with sufficient data reporting operative performance, such as sensitivity and specificity, were meta-analysed. There was no contact with the authors of the eligible articles, and no hand searching was performed.

Data extraction

Relevant data were extracted by two researchers using a standardised form. Personal and clinical characteristics of the study cohort, data related to publication, mechanical ventilation settings, specific EEOT variables and relevant details on haemodynamic monitoring technology were selected. The ‘detection time ratio’ was calculated by dividing the averaging time of the haemodynamic monitoring method by the occlusion time. In addition, the method and characteristics of volume expansion, as well as the number of fluid responders and nonresponders, were extracted. Operative characteristics included sensitivity, specificity, area under the receiver-operating characteristic curve (AUC) and the threshold of the EEOT-induced increase in a cardiac output marker to predict fluid responsiveness.

Statistical analysis

As several studies report results of EEOT performance with different settings, only one result per study was selected for meta-analysis. Only settings with complete performance data, that is, sensitivity and specificity, were considered. The result from the setting, which matched the settings of the ventilator, haemodynamic monitoring method and CO marker as used during the volume expansion for actual fluid responsiveness in that study was chosen. Whenever necessary, the occlusion duration with the highest AUC was chosen.

For each study, true positive, false-positive, false-negative and true-negative test results were calculated based on the reported sensitivity, specificity and number of true fluid responders and nonresponders. Furthermore, positive and negative likelihood and diagnostic odds ratios were calculated. Generally, a diagnostic test is considered confirmative when the positive likelihood ratio (LR+) is at least 10 and exclusive when (LR−) is 0.1 or less. A +0.5-continuity correction was added, as some of the instances from the confusion matrix were zero, and therefore a diagnostic odds ratio could not be determined.

The bivariate-random effects (logistic-normal) meta-analysis model (BRMA) was used to meta-analyse the data, this hierarchical model jointly models sensitivity and specificity, taking into account their relationship at different thresholds.¹⁷ The study-specific sensitivity and specificity with exact 95% CI were graphically displayed using a forest plot. Furthermore, Fagan plots were constructed to illustrate clinical utility and also summary receiver-operating characteristic curves (sROC), from which the pooled AUC was computed.

A bivariate I²-statistic proposed by Zhou and Dendukuri¹⁸ was used to quantify between-study heterogeneity. The I² value is the percentage of variation across studies due to heterogeneity and not chance; greater than 50% was considered substantial. Potential sources of heterogeneity were investigated using a meta-regression analysis. Separate regression models were estimated for the following covariates for assessing their effect on EEOT's performance:

• (1)Clinical context: ICU versus OR.
• (2)Tidal volume: ≤7 versus >7 ml kg⁻¹. As tidal volume was not fixed, only studies or a subset within a study, with ∼75% of the patients well above or at/below the benchmark of 7 ml kg⁻¹ were used.
• (3)Positive end-expiratory pressure (PEEP) level: ≤7 versus >7 cmH₂O.
• (4)Occlusion duration: ≤15 versus >15 s.
• (5)Haemodynamic monitoring technology: calibrated pulse contour analysis (CPCA) versus other modalities: nonCPCA (NCPCA), echocardiography and Doppler.
• (6)Detection time ratio: <0.5 versus 0.5 < 1 versus ≥1.
• (7)Cardiac output marker: direct – CO and cardiac index (CI) versus indirect – velocity time integral (VTI), stroke volume (SV) and stroke volume index (SVI).

The likelihood ratio test was used to statistically compare the joint models for sensitivity and specificity. Tau (τ²) was displayed as a measure of dispersion of true effect sizes between studies, again 0.5 was considered high.

Publication bias was assessed using Deek's funnel plot asymmetry test. Statistical significance was set at P < 0.05. All analyses were performed with Stata 18.5 SE for Windows (StataCorp., College Station, Texas, USA) and its user-written modules metadta and midas.^19,20

Assessment of study quality and risk of bias

Two independent researchers evaluated the included the body of evidence for quality according to GRADE and the risk of bias using the QUADAS-2 scale, which includes four domains: patient selection, index test, reference standard, and flow and timing.^21–23 The QUADAS tool is specifically designed to assess risk of bias and applicability of diagnostic test accuracy studies. Disagreements were resolved through consensus with a third researcher.

Results

Included studies and patients

A total of 24 studies, published between 2009 and 2023, were identified.^24–47 The study selection process is shown in Fig. 1, and the final decisions and reasons for exclusion in Supplemental Digital Content 2. A total of 1073 adults were included, and 1204 EEOT assessments were reported. Fifteen studies evaluated EEOT in an ICU context (n = 588), and nine studies were conducted in the OR (n = 485), (Supplemental Table S2–S3). The ICU cohort were patients with acute circulatory failure and were critically ill with a large presence of sepsis and ARDS. The timing of EEOT in OR patients varied, with some studies assessing it immediately after the induction of anaesthesia, others after incision and some postsurgery. Six OR studies used haemodynamic stability as a prerequisite. Vasopressor use was reported in most studies performed in the ICU, but they were either excluded or not reported in studies performed in an OR setting. The use of inotropes (dobutamine) was reported in two studies.^29,36 Arrhythmias were present in 33 (3%) patients, all part of the ICU cohort.

Fig. 1.

Selection methodology of included studies.

End-expiratory occlusion test application

The mechanical ventilation settings while applying EEOT are shown in Supplemental Table S4; volume-controlled ventilation was the most used (n = 10), followed by (volume) assist controlled ventilation (n = 9). The tidal volume ranged between 5.8 and 8.8 ml kg⁻¹, there was no major difference in the distribution of tidal volumes between ICU and OR. The PEEP ranged between 4 and 15 cmH₂O, and in the OR studies, PEEP was lower (≤6 cmH₂O) than in the ICU studies. Several studies have evaluated the EEOT performance in different scenarios of PEEP, tidal volume or patients’ respiratory system compliance.^{35,42,43,46,47}

The EEOT application settings per study are displayed in Supplemental Table S5. The duration of EEOT ranged between 12 and 40 s, but most ICU studies reported 15 s. In the OR, the EEOT duration was more variable and often longer than 15 s. The most common haemodynamic monitoring technique used to assess EEOT-induced haemodynamic changes was CPCA (n = 12), followed by NCPCA (n = 7) and echocardiography (n = 5). CPCA was used only in ICU studies, whereas NCPCA was used solely on the OR. The most common marker used to assess the effect of EEOT was cardiac index or output (n = 17), followed by stroke volume (index, n = 8) and echocardiography measures (VTI or V_max, n = 7). Several articles have studied the differences in the predictive value of multiple haemodynamic monitoring techniques and/or markers.^{24,29,31,34–39,41,45}

Overall performance and heterogeneity

Of all included fluid responsiveness assessments, 617 (51%) were actually positive as assessed by fluid administration, PLR, or the Trendelenburg manoeuvre. As part of this reference method CI, CO, SV, SVI or VTI were measured by NCPCA, echocardiography or Doppler, with thresholds ranging from 8 to 15% (Supplemental Table S6).

Two studies were excluded from the meta-analysis due to incomplete performance data (Fig. 1).^32,33 As shown in Table 1, the pooled sensitivity and specificity of EEOT from the 22 meta-analysed studies were 0.87 (0.81 to 0.92) and 0.90 (0.85 to 0.94), respectively. Further details on EEOT performance in individual studies are displayed in Supplemental Tables S5 and S7–S8. A summary ROC curve is shown in Figure S1. The median [IQR] reported cutoff value for predicting fluid responsiveness with CI was a 5 [3.3 to 5.3]% increase during EEOT, calculated solely from the meta-analysed studies that used this most prevalent marker. The correlation between sensitivity and specificity on the logit scale was −0.05, indicating that a threshold effect is not of major concern. The bivariate estimate for generalised variability was I² = 57.79 (τ² = 1.12), suggesting substantial heterogeneity. The P value of the likelihood ratio test comparing the fitted random-effects model to a fixed-effects model was less than 0.0001, indicating that the random effects model was a better fit to the data.

Table 1.

Meta-regression results

			Pooled estimates
		n	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	P-value (global test)	Heterogeneity τ2
Overall		22	0.95 (0.92 to 0.96)	0.87 (0.81 to 0.92)	0.90 (0.85 to 0.94)	–	1.12
Clinical context	ICU	15	0.96 (0.94 to 0.97)	0.88 (0.80 to 0.93)	0.94 (0.90 to 0.97)	<0.01	0.34
	OR	7	0.87 (0.84 to 0.90)	0.85 (0.71 to 0.93)	0.72 (0.58 to 0.82)
Duration occlusion	Low (≤15 s)	16	0.96 (0.94 to 0.97)	0.88 (0.79 to 0.93)	0.93 (0.87 to 0.96)	0.08	0.83
	High (>15 s)	6	0.90 (0.87 to 0.93)	0.88 (0.73 to 0.95)	0.79 (0.62 to 0.90)
PEEP level	Low (≤7 cmH2O)	12	0.92 (0.89 to 0.94)	0.86 (0.76 to 0.92)	0.83 (0.73 to 0.90)	0.03	0.72
	High (>7 cmH2O)	10	0.93 (0.91 to 0.95)a	0.89 (0.79 to 0.95)	0.95 (0.89 to 0.98)
Tidal volumeb	Low (≤7 ml kg−1)	7	0.94 (0.92 to 0.96)	0.85 (0.71 to 0.93)	0.92 (0.74 to 0.98)	0.87	0.76
	High (>7 ml kg−1)	5	0.93 (0.90 to 0.95)	0.83 (0.65 to 0.93)	0.92 (0.71 to 0.98)
Monitoring method	CPCA	12	0.96 (0.94 to 0.97)	0.90 (0.82 to 0.95)	0.95 (0.90 to 0.97)	<0.01	0.54
	Other	10	0.90 (0.87 to 0.92)	0.84 (0.72 to 0.91)	0.80 (0.69 to 0.88)
Detection time ratio	Low (<0.5)	8	0.92 (0.89 to 0.94)	0.87 (0.76 to 0.93)	0.85 (0.71 to 0.92)	0.03 c	0.49
	Medium (0.5<1)	10	0.94 (0.92 to 0.96)a	0.91 (0.82 to 0.96)	0.95 (0.90 to 0.98)	0.01 c
	High (≥1)	4	N/E	0.76 (0.52 to 0.90)	0.80 (0.58 to 0.92)	0.27c
CO marker	Direct	14	0.96 (0.94 to 0.97)	0.87 (0.77 to 0.93)	0.94 (0.88 to 0.97)	0.05	0.80
	Indirect	8	0.90 (0.87 to 0.93)	0.89 (0.77 to 0.95)	0.80 (0.66 to 0.89)

^aa +1 continuity correction was used.

^bOnly studies with ∼75% of patients well above or at/below threshold are included.

^cSeparate pairwise analysis between the three groups: low versus medium 0.03, medium versus high 0.01, low versus high 0.27; P values in bold font are statistically significant (<0.05). AUC, area under the curve; CI, confidence interval; CO, cardiac output; CPCA, calibrated pulse contour analysis; ICU, intensive care unit; N/E, not estimable; OR, operating room; PEEP, positive end-expiratory pressure.

Clinical context

Fifteen meta-analysed studies were conducted in the ICU and seven in the OR. Their predictive performances are shown in Fig. 2. The pooled estimates for sensitivity and specificity for ICU studies were 0.88 (0.80 to 0.93) and 0.94 (0.90 to 0.97), whereas they were 0.85 (0.71 to 0.93) and 0.72 (0.58 to 0.82) for OR studies. There was a statistically significant difference between the two clinical contexts (P < 0.01; Table 1). The sROCs for both contexts are shown in Fig. 3, where the OR studies had a lower AUC with larger confidence and prediction regions. The relative pooled estimates are presented in Supplemental Table S9.

Fig. 2.

Forest plot per clinical context – meta-analysis of diagnostic accuracy of the end-expiratory occlusion test for predicting fluid responsiveness in intensive care unit and operating room; 95% exact confidence intervals are shown for each study by the black lines and cumulative values in colour.

CI, cardiac index; CI, confidence interval; CO, cardiac output; CPCA, calibrated pulse contour analysis; EDM, oesophageal Doppler measurement; HP, high positive end-expiratory pressure; ICU, intensive care unit; NCPCA, noncalibrated pulse contour analysis; OR, operating room; SV, stroke volume; SVI, stroke volume index; TEE, transoesophageal echocardiography; TTE, transthoracic echocardiography; V_t, tidal volume; VTI, velocity time integral.

Fig. 3.

Summary receiver-operating characteristic curves per clinical context – summary receiver-operating characteristic curve for predicting fluid responsiveness with the end-expiratory occlusion test in intensive care unit and operating room; 95% confidence and prediction regions are shown.

AUC, area under the curve; ICU, intensive care unit; OR, operating room.

Using the computed likelihood ratios for both ICU and OR studies, a Fagan plot was created to show the clinical and patient-relevant utility and informativeness of EEOT (Fig. 4). EEOT is confirmative in the ICU setting (LR+ 14) but not exclusive (LR− 0.12). In contrast, in the OR context, EEOT is neither confirmative (LR+ 3.1) nor exclusive (LR− 0.21). Assuming a 50% pretest probability, a positive EEOT in the ICU context indicates a 93% chance of being truly fluid responsive, whereas a negative EEOT reduces this chance to 11%. In the OR, a positive EEOT indicates a 76% posttest probability, and a negative EEOT indicates a 17% posttest probability.

Fig. 4.

Fagan plot per clinical context – nomogram for clinical probability of fluid responsiveness using the end-expiratory occlusion test in intensive care unit and operating room. Starting with a 50% pretest probability of fluid responsiveness (left axis) in an ICU patient, a positive result of EEOT (solid blue arrow) indicates a 93% chance of this patient being actually fluid responsive (right axis). EEOT, end-expiratory occlusion test; ICU, intensive care unit.

Technical aspects

The technical aspects of mechanical ventilation, occlusion and monitoring methods were analysed using a meta-regression. The pooled estimates and heterogeneity results can be found in Table 1; PEEP level, detection time ratio, the haemodynamic monitoring method and CO marker showed a statistical difference in performance. This would indicate that EEOT performs better with CPCA measuring CO or CI, with a shorter averaging time than the occlusion time and with a high PEEP. However, restricting the meta-regression for the PEEP level to only the 15 meta-analysed ICU studies, a statistically significant difference was no longer observed (Supplemental Table S14 and Figure S8). Notably, when comparing all meta-regressed covariates, heterogeneity was most reduced by the division between the clinical contexts (τ² 0.34). For tidal volume and occlusion duration, no statistical difference in performance was found. The relative pooled estimates, sROC, and forest plots are shown in Supplemental Table S10–S16 and Figs. S2–S13.

Quality and risk of bias

The risk of bias assessment is shown in Fig. 5 (details in Supplemental Table S17). The risk of bias for patient selection was often unclear, as it did not describe whether a consecutive or random sample of patients were enrolled. The risk of bias for the index test, EEOT, was often scored high, as most of the articles did not prespecify the test's threshold for fluid responsiveness. The other QUADAS-2 domains scored, on average, low on the potential risk of bias or concerns regarding applicability.

Fig. 5.

QUADAS scoring to assess the quality of primary diagnostic test accuracy studies.

The overall quality of evidence from the included studies was assessed as very low according to GRADE because of the observational nature and serious concerns for inconsistency and indirectness (Supplemental Table S18). Deek's funnel plot asymmetry test for publication bias revealed a P value for the slope coefficient of 0.21, indicating no significant asymmetry (Supplemental Table S19 and Figure S14).

Discussion

The EEOT, classically described and studied in the ICU, is gaining broader interest as a generally applicable dynamic test in the detection of preload dependence of cardiocirculatory function.⁴⁴ The overall estimates for pooled sensitivity, 0.87 (0.81 to 0.92), and specificity, 0.90 (0.85 to 0.94) are in line with most previously reported meta-analyses.^11,13–15 However, the overall diagnostic performance we found is considerably better than that of the most recent report in 2023, which included only 17 EEOT studies for the prediction of fluid responsiveness.¹² These lower estimates may be related to different study selection, including data and statistical handling of covariates (categorical versus continuous) in the meta-regression. Compared with other methods for assessing fluid responsiveness recently meta-analysed, EEOT seems to perform slightly better than PPV with a sensitivity of 0.74 and specificity of 0.82, but comparable to PLR with a sensitivity of 0.86 and specificity of 0.92.^48,49 Our findings indicate that, in the ICU, EEOT seems to be clinically acceptable and is at least a confirmative dynamic test. Meta-regression revealed significantly lower performance of EEOT in the OR context and, notably, when the detection time ratio was 1 or higher. These new findings are probably due to the increased number of studies published with a variety of different patient cohorts and clinical care contexts (substantial heterogeneity, I² ≈ 58%) (EEOT_Supplement2_ScreenedRecords).

A large difference in diagnostic performance was observed in the clinical context, together with low heterogeneity (τ² = 0.34) when separating ICU from OR studies. This indicates that EEOT performance is predominantly influenced by the clinical context. This could be caused by inherent patient differences in terms of underlying clinical conditions, sedation levels, and overall clinical stability, with haemodynamic instability and vasopressor use being exclusion criteria in most OR studies. As a result, in 367 out of 485 patients in the OR, where the test was performed in a situation without strict clinical indications for fluid administration, there were concerns about the applicability of EEOT. Another explanation could be the combination of technical settings commonly used in the OR, such as lower PEEP, non use of CPCA, indirect CO markers and lower fluid bolus volumes, resulting in lower performance. The statistically significant decrease in EEOT accuracy for these technical factors found in the 2023 review was confirmed by our meta-regression results. However, the meta-regression analysis for these covariates still showed substantial heterogeneity. In addition, comparing PEEP levels in ICU studies only revealed that the difference in performance was nonsignificant. Nevertheless, the clinical context-specific performance variation has important implications for the clinical utility of EEOT in the OR; around a quarter of patients with a positive EEOT are not fluid responsive and one in five to six patients with a negative EEOT could still be responsive. In this sense, our results indicate that the EEOT may potentially mislead clinicians, and it should be interpreted in a nuanced way, bearing in mind a differentiated critical appraisal of studies from two very different clinical contexts.

A novel insight and major factor determining EEOT performance is the detection time ratio, that is, the averaging time of the haemodynamic measurement should be shorter than the duration of airway occlusion imposed by EEOT. Studies with the smallest detection time ratio (<0.5) also tend to perform slightly worse; most measurements were echography-based, with a high user dependency, or were not averaged at all, which could lead to more noise. The duration of EEOT itself did not result in statistically significant differences in the pooled estimates, which confirms the results of two previous meta-analyses.^13,14 Theoretically an even shorter occlusion could be sufficient, as pulmonary transit time is only 6.2 ± 2.2 s in healthy controls.⁵⁰ However, the temporal resolution of most monitoring technologies hinders detection of these rapid changes. As the required haemodynamic fluctuation imposed by an occlusion manoeuvre predictive of fluid responsiveness is subtle, that is, a median CI change of 5%, accurate measurement with high temporal resolution is mandatory to detect these changes. When choosing a monitoring technology, the so called ‘least significant change’ should be taken into account, which is around 2% for pulse contour derived CO.^51,52 It should also be noted that different ‘optimal’ thresholds, that is, cut-off values for a positive EEOT, were chosen among the meta-analysed studies. Although a threshold effect was not found in this study, as reflected by the low correlation between sensitivity and specificity, different optimal thresholds could be valid for each haemodynamic marker. As reported previously, the meta-analysis revealed that the method for haemodynamic monitoring influences the predictive performance of EEOT; specifically, it increases with the use of CPCA.^11,12 However, the use of CPCA with CI is strongly linked to the ICU context, and the regularly used PiCCO device has an averaging time of 12 s, which, in combination with an occlusion duration of 15 s leads to a detection time ratio of 0.8. Both the clinical context and detection time ratio analysis showed lower values for heterogeneity. These factors could affect the statistical significance of the enhanced CPCA and direct CO marker performance.

Limitations

A key limitation of all meta-analyses of diagnostic test accuracy is heterogeneity, which is commonly encountered. Although some sources of heterogeneity have been identified, future studies may uncover more and improve the estimates. A per-patient meta-analysis could better reveal the sources of heterogeneity and determine when and where EEOT is most effective. However, with current evidence, this is not possible, as studies have reported only group-level data. For example, factors such as tidal volume, vasopressor use, and prone position were insufficiently analysed in this meta-analysis because of limited (patient-specific) data.

Despite an increase in observational studies on EEOT, the overall quality of evidence remains low, with small sample sizes, single-centre designs, and varied risks of bias. Particularly, the reference method for actual fluid responsiveness assessment varied greatly between studies. Previous systematic reviews have shown inconsistent bias assessments, and the current evaluation may add another subjective layer. A recently published consensus on fluid responsiveness studies may guide more uniform research design and reporting.⁵³

No formal publication bias was identified; however, the borderline P value (0.21) suggested that bias could be significant if ICU and OR studies were evaluated separately. Additionally, two studies and several results from lower performing settings were not included in the meta-analysis due to poor performance, potentially leading to an overestimation of EEOT effectiveness.

Conclusion

This meta-analysis showed that the performance of the EEOT in predicting fluid responsiveness depends on clinical context and technical settings; however, overall pooled estimates indicate acceptable sensitivity and specificity. The diagnostic accuracy was significantly lower when the averaging time of the haemodynamic monitoring methodology was longer than the duration of airway occlusion or when used in the OR. The latter is either due to the inherent patient group differences or because of the absence of calibrated pulse contour analysis with direct CO marker for haemodynamic monitoring, and the use of lower PEEP (≤7 cmH₂O). Nevertheless, as the high heterogeneity and risk of bias of the analysed studies resulted in an overall low quality of the evidence, this conclusion should be interpretated with caution. Overall, the clinical utility of peri-operative EEOT seems to be remarkably lower than that in the ICU context, where the test proved to be confirmative. This finding points to the need to consider EEOT in a more context-specific and patient-specific manner.