Acute normovolemic hemodilution (ANH) can trace its roots to the early days of cardiopulmonary bypass. Hemodilution with crystalloids instead of priming the pump with blood was shown to be a safe, blood-sparing technique as early as 1960.1 In the 1970s, the practice of “hemospasia,” or the sequestering of whole blood into bags separate from the circuit and then reinfusing after the procedure, took the idea 1 step further.2 Perhaps, this history is why ANH piques the interest of so many cardiac anesthesiologists.
In this issue of the Journal, Barile et al3 report on a meta-analysis of ANH use specifically in cardiac surgery. Their primary end point was the number of allogenic red blood cell (RBCs) units transfused. Utilizing the data from 21 randomized trials involving 1852 total patients, they calculated that ANH patients received about ¾ of a unit fewer RBCs compared with controls. Clinicians will likely interpret this finding according to their own personal bias. Proponents of ANH will point out that the technique does in fact have benefit, while detractors will likely find the overall impact to be less than impressive.
If debating the results of an ANH meta-analysis gives Anesthesia & Analgesia readers a sense of déjà vu, it would be completely understandable. This month’s Barile article is actually the third meta-analysis on the subject that has been published in Anesthesia & Analgesia to date. Most recently, Zhou et al4 identified 62 studies with 3819 total patients that utilized ANH in a wider variety of surgical procedures. Interestingly, in a primary aim identical to that in the Barile article, they demonstrated a similar finding: ANH was good for saving a little less than 1 unit of RBCs. However, Zhou et al4 also noted significant heterogeneity between studies and publication bias that likely overestimated the true effect. In an accompanying editorial by noted blood conservationist Steven Frank and colleagues, cardiac surgery was suggested as being the ideal venue for assessing the efficacy of ANH.5 Thus, we have the efforts of Barile et al.3
It is a fair question to ask how different 2 ANH meta-analyses published a little over a year apart could possibly be, particularly since each included only randomized controlled trials. Many readers will likely be surprised to find that for purposes of comparing the number of RBC units transfused between groups, the 2 articles have only 10 studies in common. Given that the focus of Barile et al3 was specifically on cardiac surgical patients, this is not necessarily unexpected. However, it does illustrate that data selection in a meta-analysis is not an exact process. Even with similar initial search strategies, altering criteria like language restrictions and publication years, as well as the effectiveness of searching bibliographies (ie, “backward snowballing”), can produce different data sets. A comparison between the studies in Zhou et al4 and Barile et al3 is provided in Table 1.
An important point demonstrated by both articles is that the degree of heterogeneity among ANH studies is high no matter what patient population is focused upon. In plain English, this means that the inconsistency of effect seen by the included studies is not simply because of chance. There are likely differences in study design, patient populations, and other variables that influence the aggregate result. This is the aspect of a meta-analysis that typically draws the most criticism—blending apples with oranges. The appropriateness of combining studies with different covariates is a judgment call, and this limitation must be kept in mind. Readers should ask themselves if controls benefited from the same additional measures as the ANH groups; if not, was decreased utilization due to something besides the use of ANH?
Important sources of variation across studies that were identified by Barile et al3 include type of surgery, the amount of blood removed, year of publication, and the presence of a transfusion protocol.3 Stratification by these variables can be done (and was) in a secondary analysis. However, that does not necessarily produce a completely unconfounded answer since interaction between variables is not considered. Removing more ANH blood may be more effective in certain types of procedures, or transfusion protocols may be more effective over time. It becomes very difficult to sort out all of the apples and oranges in a fruit salad.
Meta-analyses are inherently Bayesian since scientifically and statistically we assume that there will be natural variation in the treatment effect across studies, even if very similar protocols are followed. So, the question from the scientific and statistical perspective is not whether there was variability in the effects across studies, but whether the observed variability is more than expected by chance. This is typically measured with the “Q statistic” or “I2 statistic.” In the case of both Zhou et al4 and Barile et al3 ANH meta-analyses, there is indeed greater observed heterogeneity in treatment effects than expected by chance.
Of course, how much statistical and clinical heterogeneity is “too much” to conduct a meta-analysis as part of a good systematic review is somewhat of an unresolved question. Researchers are challenged to make sure their meta-analyses are focused enough to be interpretable by the interested clinician but must balance the inevitable variation encountered across studies. To this end, we commend Barile et al3 for making considerable efforts to narrow the focus to cardiac surgeries in an attempt to reduce the observed heterogeneity. Nevertheless, with an I2 of 95% (close to the maximum of 100%!), readers still need to interpret findings in a very cautious manner.
Another key feature of any meta-analysis is the assessment of publication bias—in essence, the underreporting of negative studies. This pervasive problem should be assessed by visually inspecting a funnel plot of variability versus treatment effect, as well as analytical appraisal based on the Begg adjusted rank correlation test6 or the Egger linear regression test.7 When publication bias is detected, there are novel statistical methods to try and “fill in” the missing studies.8 While not foolproof by any means, these method are an important step in efforts to remove publication bias from a meta-analysis. Both Zhou et al4 and Barile et al3 have done this well, not only assessing for it but also conducting sensitivity analyses to “fix” observed publication bias.
Perhaps, the most surprising aspect of comparing Barile et al3 with Zhou et al4 is that, despite sharing less than 50% of the data, the 2 publications both concluded a similar effect size for ANH. Consistency in the presence of variability probably says something. Per the 2011 Society of Thoracic Surgeons/Society of Cardiovascular Anesthesiologists blood conservation guidelines, ANH is a class IIb recommendation.9 Although their meta-analysis will not change that, Barile et al3 should be congratulated for taking the logical step of compiling the currently available evidence…even if it is a mixture of apples and oranges.
Name: Roman M. Sniecinski, MD.
Contribution: This author helped write the manuscript.
Conflict of Interest: Research support from Grifols and Shire ViroPharma.
Name: Edward J. Mascha, PhD.
Contribution: This author helped write the manuscript.
Conflict of Interest: None.
This manuscript was handled by: Jean-Francois Pittet, MD.
1. Neptune WB, Bougas JA, Panico FG. Open-heart surgery without the need for donor-blood priming in the pump oxygenator. N Engl J Med. 1960;263:111–115.
2. Lawson NW, Ochsner JL, Mills NL, Leonard GL. The use of hemodilution and fresh autologous blood in open-heart surgery. Anesth Analg. 1974;53:672–683.
3. Barile L, Fominskiy E, Di Tomasso N, et al. Acute normovolemic hemodilution reduces allogeneic red blood cell transfusion in cardiac surgery: a systematic review and meta-analysis of randomized trials. Anesth Analg. 2017;124:743–752.
4. Zhou X, Zhang C, Wang Y, Yu L, Yan M. Preoperative acute normovolemic hemodilution for minimizing allogeneic blood transfusion: a meta-analysis. Anesth Analg. 2015;121:1443–1455.
5. Grant MC, Resar LM, Frank SM. The efficacy and utility of acute normovolemic hemodilution. Anesth Analg. 2015;121:1412–1414.
6. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics. 1994;50:1088–1101.
7. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315:629–634.
8. Duval S, Tweedie R. Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics. 2000;56:455–463.
9. Ferraris VA, Brown JR, Despotis GJ, Hammon JW, Reece TB, Saha SP, Song HK, Clough ER, Shore-Lesserson LJ, Goodnough LT, Mazer CD, Shander A, Stafford-Smith M, Waters J, Baker RA, Dickinson TA, FitzGerald DJ, Likosky DS, Shann KG; Society of Thoracic Surgeons Blood Conservation Guideline Task Force, Society of Cardiovascular Anesthesiologists Special Task Force on Blood Transfusion, International Consortium for Evidence Based Perfusion. 2011 update to the Society of Thoracic Surgeons and the Society of Cardiovascular Anesthesiologists blood conservation clinical practice guidelines. Ann Thorac Surg. 2011;91:944–982.
10. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327:557–560.