Low-quality Trials as a Source of Heterogeneity in the Association of Hydroxyethyl Starch 130/0.4 or 0.42 with Excess Mortality in Sepsis.

To the Editor: In their meta-analysis, Ma et al.[1] aimed to determine the association of mortality with hydroxyethyl starch 130/0.4 or 0.42 (HES) with HES dose and daily fluid balance in sepsis patients. Sensitivity tests were performed to analyze reasons for heterogeneity in mortality data. These tests focused on “risk of bias.”

In the methods section, they stated that risk assessment was performed as advised by the Cochrane collaboration. Assessment includes “randomization sequence generation,” “allocation concealment,” “blinding,” “incomplete outcome data,” and “selective outcome reporting.” In Table 2 of Ma et al.[1] however, assessed domains differed with “intention-to-treat analysis” and “no loss to follow-up” replacing “incomplete outcome data” and “selective outcome reporting.” The Cochrane collaboration's tool for assessing the risk of bias recommends that assessment of “other potential bias” be included, however, no such attempts were made, but bias assessment appears incomplete and flawed.

The study of Zhu et al. 2011[2] was categorized as “intermediate risk” because blinding was identified as high risk; other unclear domains were not identified: “Random sequence generation” is unclear because not stated; “allocation concealment” is not described; and “incomplete outcome data bias” is unclear since it is not stated that there were no dropouts. With correct assessment, the study should have been categorized as “high risk.”

The study of Annane et al. 2012[3] did not restrict the use of various HES products and the number of patients actually receiving HES 130/0.4 or 0.42 is unknown; therefore, the study should have been excluded from the meta-analysis. It was unblinded and allocation of patients occurred by envelope randomization using fixed block size, which may not have ensured allocation concealment. This is supported by marked baseline imbalance in the numbers of patients that received crystalloids and colloids in the 12 h prior to randomization.[4] Other bias includes the secondary 90-day mortality endpoint which was used in this meta-analysis, that was exploratory only, added midway through the 9-year trial.[5] The correct risk of bias assessment in Table 2 should have been “high risk” since high-risk bias was identified at least for two domains (“blinding” and “allocation concealment”).

With two corrections in the risk of bias assessment, results of the sensitivity analysis are different [Figure 1]. The mixed effects analyses of trials with low risk and high risk of bias showed a relative risk of 1.12 (1.01–1.23; χ2 = 2.24, P = 0.02) and 0.77 (0.54–1.08; χ2= −1.51, P = 0.13), respectively, and the test for subgroup difference between trials with low versus high risk of bias was significant (χ2 = 4.23, P = 0.04).

Figure 1

Forest plot of all-cause mortality in relation to risk of bias in trials. The size of squares for Mantel–Haenszel (M-H) risk ratio reflects the weight of trial in pooled analyses. Horizontal bars represent 95% confidence intervals. Results of a mixed effect analysis are shown. Pooled effect size and subgroup analyses are based on a fixed effects model using Comprehensive Meta-analysis version 2.2.064.

Despite wrong inclusion of the Annane et al. 2013 study, which puts doubt on the study selection process, other weakness includes the restriction to HES 130/0.4 or 0.42 because side effects are share among different HES preparations; several of the included studies reported nonstratified subgroup data on sepsis patients only; and in the meta-analysis of Ma et al.[1] the Siegemund et al. 2012 study is still classified as “low risk of bias” although still not fully published.

On the basis of studies selected for this biased meta-analysis, the better conclusion is that pooled analysis of mortality showed neither benefit nor harm, but this analysis may be influenced by trials of low-quality, since trials with a low risk of bias suggested an excess mortality of 12% of sepsis patients assigned to HES 130/0.40 or 0.42 versus crystalloid or albumin. It is important to communicate this to the readers of the Chinese Medical Journal.

Financial support and sponsorship

Nil.

Conflicts of interest

The author has received fees for speaking and travel reimbursements from manufacturers of plasma-derived therapies (Grifols, Kedrion, CSL Behring, Baxter).