The clinical research question was: What is the sensitivity and specificity of the ΔIVC when using it to predict fluid responsiveness?

The PICO statement is as the following:

P-patient, problem or population: patients with acute circulatory failure in whom the effect of volume expansion (VE) is unknown and needs to be predicted.

I-intervention: Inferior vena cava (IVC) diameter was examined subcostally and measured in M-mode or 2D mode, 2 cm before the IVC joined the right atrium. The IVC respiratory variation (ΔIVC) was calculate by recordingthe largest and smallest IVCdiameter at end-inspiration or end-expiration.

C-comparison, control, and comparator: Fluid responsiveness was defined as a significant increase of stroke volume (SV), cardiac output (CO) or other surrogates during a VE.

O-outcomes: Ability of the △IVC to predict fluid responsiveness.

Our aim was to identify all studies evaluating the ability of the ΔIVC to predict fluid responsiveness compared to the increase in SV, CO or other surrogates induced by subsequent VE.

We searched the MEDLINE, EMBASE, Cochrane and Web of Science databases for relative studies published in English from inception to July 2016. The key words we used consist of term related to IVC (“inferior vena cava”, “caval index”, “collapsibility” and “distensibility”) and terms related to volume status (“fluid or volume or preload responsiveness”, “fluid or preload challenge”, “preload dependence or independence or dependency or independency”, “functional haemodynamic monitoring” and “fluid therapy or management”). These key words were searched separately by two groups using different combination strategy. We also looked for relevant articles cited in review articles, commentaries and editorials. The search was performed repeatedly until no new studies could be found.

Study identification was performed in two steps. Step 1 comprised screening for titles and abstracts, and step 2, review of full texts of studies obtained in step 1. We only included studies investigating the accuracy of the △IVC that were published in full text or accepted for publication in indexed journals. Excluded criteria were 1) studies using central venous pressure or right atrial pressure as the reference standard, because these static parameters cannot predict fluid responsiveness accurately; 2) studies measuring IVC with techniques other than ultrasonography; 3) studies involving animals and healthy volunteers. Two reviewers process searching independently, disagreement was settled by a third opinion. The quality of the included studies was evaluated by using the QUADAS-2 scale [11] . The meta-analysis was performed according to the PRISMA statement.

Important information was extracted from the included articles using a standardized data form by two reviewers. Extracted data include the name of the first author, publication year, characteristics of the investigated population, sample size, respiratory pattern, the device for IVC measurement, formula for the calculation of ΔIVC, definition of fluid responsiveness and volume challenge strategy, the number of true positives, true negatives, false positives and false negatives, sensitivity, specificity, the area under the receiver operation characteristics curve (AUROC) and the best threshold of ΔIVC which is used to predict the fluid responsiveness.

Included studies were assessed for their quality based on the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) protocol. QUADAS-2 scale [11] was made up of 4 domains: patient selection, index test, reference standard, and flow and timing. Each domain is assessed in terms of risk of bias, and the first 3 domains are also assessed in terms of concerns regarding applicability. For the ‘‘patient selection’’ domain, we examined whether patients were consecutively included and whether inappropriate exclusions were avoided. For the ‘‘index test’’ domain, we examined whether the threshold used to define volume responsiveness was pre-specified. For the ‘‘reference standard’’ domain, we examined whether the result of VE on SV, CO or surrogates was assessed without knowledge of ΔIVC result. Finally, for the ‘‘flow and timing’’ domain, the authors examined whether there was an appropriate interval between IVC measurement and VE, whether patients received the same VE and whether all patients were included in the analysis. For each criterion, the risk was judged as high, low and unclear.

We performed a meta-analysis in order to determine the pool sensitivity, specificity and diagnostic odds ratio (DOR). In addition, the pooled area under the ROC curve (AUROC) and threshold for ΔIVC as a predictor of fluid responsiveness was also evaluated. To investigate a threshold effect, we calculated the Spearman correlation coefficient between sensitivity and specificity. Homogeneity between studies was tested by the Chi squared test and I² index. According to heterogeneity, we adopted a random effect model by using the method of DerSimonian-Laird from the Mantel-Haenszel model. We compared studies with ICU setting versus non-ICU setting making the hypothesis that ΔIVC could be more reliable in ICU patients. We compared studies with adults versus children making the hypothesis that ΔIVC could be more reliable in adults. We compared studies with different devices for measuring IVC making the hypothesis that one device is better than the others. We compared studies with three different formulas for the calculation of ΔIVC making the hypothesis that one formula is better than the others. We compared studies with patients on mechanical ventilation versus studies with spontaneously breathing patients, testing the hypothesis that the reliability of ΔIVC is better in patients on mechanical ventilation. We compared studies where fluid responsiveness was defined by an increase in SV, CO or surrogate ≥ 15% versus studies with other definitions of fluid responsiveness, testing the hypothesis that the reliability of ΔIVC is better when fluid responsiveness is defined by a larger increase. We compared studies where SV, CO or surrogate were measured by echocardiography versus studies where they were measured by other methods, testing the hypothesis that the reliability of ΔIVC is better when SV, CO or surrogate were measured by echocardiography. Finally, we compared studies where VE was performed with versus studies where it was performed by colloids, testing the hypothesis that the reliability of ΔIVC is better when VE is performed with colloids. Causes of heterogeneity were also investigated by meta-regression based on the Littenberg and Mose linear model.

Results are expressed as mean (95% confidence interval) or as mean ± standard deviation. The meta analysis was performed with Meta-Disc v.1.4 (Universidad Complutense, Madrid, Spain). The additional statistical analysis was performed with MedCal 15.2.2 (MedCal Software, Mariakerke, Belgium). A two-tailed p < 0.05 was considered to statistical significance.

A flow chart of the study selection is provided in Figure 1. Our initial search identified 399 citations. 379 of them were excluded: 320 for not relating to the subject, 49 for being reviews, letters, guidelines, case reports and editorials, 3 for not writing in English, 4 for not using proper reference standard, 3 for being animal experiments. Finally, a total of 20 studies [12] - [31] reported the ability of △IVC to predict fluid responsiveness were included in our analysis.

Characteristics of included studies are listed in Table 1. Sample sizes were

IVCmax and IVCmin = maximum and minimum diameter of inferior vena cava during a complete respiratory cycle; CI = cardiac index; CO = cardiac output; VTI = velocity-time index; SV = stroke volume; SVI = stroke volume index; SBP = systolic blood pressure; TV = tidal volume.

small, ranging from 14 to 50 patients. A total of 635 patients were included. 16 studies [12] [13] [14] [16] [17] [18] [19] [21] - [27] [30] [31] enrolled adults, and 4 studies [15] [20] [28] [29] enrolled pediatric patients. 14 studies [12] [13] [14] [15] [16] [20] [21] [23] [25] [26] [28] [29] [30] [31] enrolled patients on mechanical ventilation, and 6 studies [17] [18] [19] [22] [24] [27] enrolled spontaneously breathing patients. The formulas for the calculation of ΔIVC during the respiratory cycle were different. (IVCmax − IVCmin)/IVCmin was used in 11 studies [12] [14] [16] [21] [23] [25] [26] [28] [29] [30] [31] , (IVCmax − IVCmin)/IVCmax was used in 5studies [17] [18] [22] [24] [27] and (IVCmax − IVCmin)/[(IVCmax + IVCmin)/2] was used in 4 studies [13] [15] [19] [20] . Interestingly, the 11 studies using the formula (IVCmax − IVCmin)/IVCmin allfocused on mechanically ventilatedpatients, and the 5 studies using the formula (IVCmax − IVCmin)/IVCmaxall focused on spontaneously breathing patients. In the 4 studies using (IVCmax − IVCmin)/[(IVCmax + IVCmin)/2] as the formula, one study focused on spontaneously breathing patients, while the other three studies focused on mechanically ventilated patients. With respect to reference standard, fluid responsiveness was defined as an increase in SV, CO or surrogate by more than 15% in 14 studies [12] [13] [14] [15] [18] [19] [21] [22] [23] [25] [26] [29] [30] [31] , 10% in 5 studies [16] [17] [20] [27] [28] , and increase in SBP by more than 10 mmHg in 1 study [24] . 16 studies [12] [13] [15] [16] [18] - [23] [25] - [31] used echocardiography to measured SV, CO or surrogate, 2 studies [14] [26] used transpulmonary thermodilution technique to measure CO, 1 study [17] used bioimpedance to measure cardiac index (CI) and the last study [24] used arterial catheter to measure SBP. VE was performed by crystalloids in 8 studies [16] [19] [22] [24] [27] [29] [30] [31] , by colloids in 10 studies [12] [13] [14] [15] [18] [20] [21] [23] [25] [28] , passive leg raise in 2 studies [17] [26] . Quality assessment according to QUADAS-2 criteria is outlined in Figure 2.

The diagnostic performance of ΔIVC in each study is showed in Table 2. The

TP = true positive; FP = false positive; FN = false negative; TN = true negative; AUROC = area under the receiver operating characteristic curve; CI = confidence interval.

sensitivity and specificity was reported in 14 studies [12] [13] [14] [16] [18] [19] [22] [23] [24] [26] [27] [29] [30] [31] . The pooled sensitivity, specificity and DOR was 0.68 (0.62 - 0.75), 0.80 (0.75 - 0.85) and 14.2 (6.0 - 33.6), respectively. (Table 2, Figure 3). The area under the corresponding ROC curve was reported in 17 studies [12] [14] - [24] [27] [28] [29] [30] [31] . In 9 studies [12] [14] [15] [16] [18] [22] [24] [29] [31] , the AUROC of ΔIVC were more than 0.7, and in the other 8 studies [17] [19] [20] [21] [23] [27] [28] [30] , ΔIVC showed low diagnostic value. The pooled AUROC was 0.86 (0.78 - 0.93) (Table 2, Figure 4). The threshold of △IVC was reported in 13 studies [12] [13] [14] [16] [18] [22] [23] [24] [26] [27] [29] [30] [31] , the values varied across studies, ranging from 12% to 42% (Table 2).

The Spearman correlation coefficient between sensitivity and specificity was 0.323 (p = 0.260), indicating no threshold effect. The heterogeneity Chi-squared was 56% for sensitivity and 39% for specificity. The I² statistics was 77% for sensitivity, 66% for specificity.

Meta-regression shows none of the covariates included were the significant source of heterogeneity. However, the comparison between studies with mechanical ventilation versus studies with spontaneously breathing, and between

studies with different devices and formulas for the calculation of ΔIVC had influence on sensitivity and specificity. Diagnostically, ΔIVC performed better in patients on mechanical ventilation than in spontaneously breathing patients with higher sensitivity (0.75 vs.0.56), specificity (0.82 vs. 0.78), DOR (22.9 vs. 7.9), and AUROC (0.9 vs.0.8) (Table 3). In addition, 9 studies [12] [13] [14] [16] [23] [26] [29] [30] [31] with mechanical ventilation reported the threshold ranging from 12% to 23.5%, the average was 17% ± 4%; the average of the other 4 studies [18] [22] [24] [27] with spontaneously breathing was 33% ± 12%.