Value of variation index of inferior vena cava diameter in predicting fluid responsiveness in patients with circulatory shock receiving mechanical ventilation: a systematic review and meta-analysis

Relevant studies up to June 2017 were searched in the PubMed, Embase, and Cochrane Library databases with the following terms and their combination: “fluid therapy,” AND “inferior vena cava,” AND “mechanical ventilation,” AND (“acute circulatory failure” OR “shock”). All scanned abstracts, studies, and citations were reviewed. Moreover, references of the retrieved manuscripts were also manually cross-searched for further relevant publications.

All the studies on the requirements for ΔIVCD were evaluated by two independent authors, and any disagreement was resolved by group discussion until a consensus was reached. The inclusion criteria were as follows: (1) studies on patients with shock receiving mechanical ventilation; (2) studies with a reference gold standard for diagnosing fluid responsiveness; (3) studies published in any language; and (4) studies providing sufficient data for constructing two-by-two tables, including true-positive (TP), false-positive (FP), true-negative (TN), and false-negative (FN). The exclusion criteria were as follows: (1) studies that used the same population or overlapping databases and (2) studies on animal models.

All the available data were extracted from each study by two investigators independently according to the aforementioned inclusion criteria, and any differences were resolved by discussion with a third investigator. The following data were collected from each study: (1) basic characteristics of studies, including name of the first author, publication year, country where the research was performed, sex, mean age, number of patients, tidal volume, index test, reference standard measurement, reference standard threshold, and reference standard device; and (2) diagnostic performance, including cutoff value, sensitivity, specificity, area under the ROC curve (AUROC), TP, FP, FN, and TN. The quality of included studies was scored independently by two reviewers using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria [12]. The quality of studies was assessed using RevMan (version 5.3, 2014; The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark), with four key domains: patient selection, index test, reference standard, and flow and timing. Seven questions were used to evaluate the quality of included studies. Each question was answered with “yes,” “no,” or “unclear.” An answer of “yes” meant that the risk of bias could be judged as low, whereas an answer of “no” or “unclear” meant that the risk of bias could be judged as high. In the case of conflict, a third reviewer was consulted, and disagreement was settled through multilateral discussion.

All analyses were performed using Stata 14.0 software (StataCorp, College Station, TX, USA). The bivariate meta-analysis model was used to summarize sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and diagnostic OR (DOR) [13, 14]. The sensitivity and specificity of each included study were used to plot the summary ROC (SROC) curves and calculate the area under the SROC curve (AUC). The AUC could be statistically interpreted as the probability of distinguishing patients from normal control subjects correctly. The between-study heterogeneity was evaluated using Q test and I² statistics. A P value less than 0.10 for the Q test or I² value ≥ 50% indicated substantial heterogeneity, and then the random-effects model was applied. Confirming the stability of the present study, the sensitivity analysis and outliner exclusion were also performed. Because publication bias is a concern for meta-analyses, the Deeks’ funnel plot asymmetry test was used, with P < 0.10 indicating statistical significance [15].

This meta-analysis yielded 145 primary studies after the initial independent review, comprising 144 published studies identified through electronic database searches and 1 published study identified through a manual search. Figure 1 shows the study selection process. A total of 36 studies were initially excluded owing to duplicate records; 94 studies were excluded owing to the source not being related to the research topic or being a letter, review, or meta-analysis; and 9 studies were excluded because they included surgical, cardiac, or mixed patients or did not present usable data. Finally, six studies [9, 10, 16‐19] fulfilled all the inclusion criteria and were considered for analysis. The main characteristics of the eligible studies are shown in Table 1. The quality of the included studies was assessed using QUADAS-2, as shown in Fig. 2.

Study data and individual diagnostic estimates are summarized in Table 2. Overall, 603 patients were included in this review, 324 (53.7%) of whom were fluid-responsive. The cutoff values of ΔIVCD varied across studies, ranging from 8% to 21%. The AUROC of individual studies ranged from 0.43 to 0.91. Heterogeneity between studies was assessed with an overall Q = 0.069, I² = 0%, and P = 0.483. The pooled sensitivity and specificity for the overall population were 0.69 (95% CI, 0.51–0.83) and 0.80 (95% CI, 0.66–0.89), respectively (Fig. 3). The DOR was 9.28 (95% CI, 2.33–36.98) (Fig. 4). AUROCs were reported in five studies [9, 16‐19]. In four studies [9, 17‐19], the AUROC was > 0.5, and in the remaining study [16], ΔIVCD had no diagnostic value. Overall, the pooled AUROC was 0.82 (95% CI, 0.79–0.85) (Fig. 5).

Goodness-of-fit and bivariate normality analyses (Additional file 1: Figure S1a and b) showed that the bivariate model was robust for calculating the pooled estimates. Influence analysis and outlier detection (Additional file 1: Figure S1c and d) did not reveal outlier studies, suggesting the robustness of the present meta-analysis.

The publication bias of the studies was assessed using the Deeks’ funnel plot asymmetry test. The slope coefficient of the six studies was associated with a P value of 0.19 (Additional file 2: Figure S2). The aforementioned results indicated symmetrical data and no significant publication bias.