An umbrella review was conducted, based on the Joanna Briggs Institute methodology (Ioannidis, 2009). The predefined protocol, published in PROSPERO International prospective register of systematic reviews (CRD42022299392), adhered to the PRISMA-P guidelines (Moher et al., 2009).

PubMed, MEDLINE (OVID), Embase, Web of Science, Cochrane Library and Emcare were searched from inception to the 15th of April 2022 for studies that reported on physiological and patient reported outcomes following a Hb change. Additionally, we used hand-searching and screened reference lists and guidelines for finding additional potential inclusions for our umbrella review. There were no language or publication year restrictions. The detailed search strategy is reported in Supplementary Material S6.

Studies were eligible if they were published systematic reviews (SRs) or meta-analyses which included primary studies that evaluated the clinical effect of interventions that change the Hb-level. Non-systematic reviews, reviews in which no Hb levels were reported, or in which Hb change was solely due to blood loss in surgery or other forms of acute bleeding (other than intended phlebotomies) were excluded. We aimed to include systematic reviews on all types of anemias, non-anemias and polycythemia. Eligible interventions that change the Hb level are transfusion, phlebotomy, isovolumic hemodilution, altitude training, EPO, and iron and vitamin supplementation.

Outcomes that were recorded in this umbrella review are: VO₂max, heart rate, respiratory rate, heart rate variability, exercise capacity, fatigue, patient reported outcome measures, patient reported experience measures, quality of life and hematological outcomes (Hb/Hct/transfusions). Within-subject designs and placebo/standard of care/other comparator-controlled designs are included. Screening, data extraction and quality assessment was performed independently in duplicate by two authors (RT and LZ). Discrepancies were discussed and when necessary, resolved with the help of a third author.

We screened all titles and abstracts of retrieved citations to assess suitability for inclusion. Two authors (RT and LZ) independently screened the full texts of articles potentially meeting eligibility criteria. Before data extraction, articles meeting eligibility criteria were checked whether they were qualitatively sufficient using the JBI Critical Appraisal checklist for systematic reviews and research syntheses. If articles were scored less than 33% on the JBI checklist, they were excluded. In case of discordant reviews, the Jadad decision algorithm was used to ascertain which meta-analysis represented the best evidence (Jadad et al., 1997). CADIMA software (Julius Kühn Institute, Germany) was used to manage records and data. The data extraction was conducted according to a modified JBI extraction tool (Supplementary Material S6). Author, year, type of participants, control groups, number of participants, description of intervention and control, number of studies, types of studies, databases searched, country of origin of the included studies, outcomes and findings, type of effect model used in the meta-analysis (fixed or random), between study heterogeneity estimates (I² values), any presented measure of publication bias, Grading of Recommendations Assessment Development and Evaluation (GRADE) rating, and sources of funding and conflicts of interest were extracted from the studies. We assessed whether the studies evaluated possible sources of heterogeneity across studies and whether the authors conducted a sensitivity analysis.

Methodological and reporting quality was assessed using the Assessment of Multiple Systematic Reviews (AMSTAR)-2 tool. This tool rates systematic reviews/meta-analysis on literature search, study selection, data extraction and transparency of the studies (Shea et al., 2007). The GRADE score was utilized to score the quality of the evidence found in the included studies (Guyatt et al., 2009). We evaluated the quality on five items: risk of bias, imprecision, inconsistency, indirectness and publication bias using the GRADE Handbook (Schünemann et al., 2013).

We performed a narrative umbrella review. Meta-analyses were not feasible with the included heterogenic data. However, forest plots of PROMs and physiological outcomes were conducted without a meta-analysis for overview purposes. In these forest plots, the effect sizes have been transformed to standardized mean differences (SMD) using R’s “MBESS” and “esc” packages, depending on available data. If sample sizes for intervention and control groups were not available separately, we assumed equal groups to estimate the confidence interval for the SMDs. Moreover, with these computed SMDs, we produced funnel plots for PROMs and physiological outcomes separately, to evaluate the possibility of publication bias.

Though a formal sensitivity analysis was not possible without a meta-analysis, we assessed whether outcomes were notably different when selecting only high-quality SRs, to reduce confounding and bias, or when leaving out SRs that evaluated ESAs. ESAs, a group of drugs that has been evaluated in larger studies with more pharmaceutical company sponsorship, may give biased results due to conflicts of interests, or contrastingly, less-biased more adequate results due to proper financing for larger, methodological superior trials.

The search strategy yielded 4,120 results. After screening of titles and abstracts, the full texts of 93 articles were further evaluated. Of these 33 studies with 169,878 participants met the inclusion criteria (Figure 2). The included studies were published from 2009 to 2020. The majority of primary studies were conducted in Europe and North-America and 24% (8/33) of the included reviews declared pharmaceutical company sponsorship. We included quantitative reviews with meta-analyses for the interventions ESAs, iron supplementation, altitude training and blood donation. The included patient groups comprising patients with cancer, hematological malignancies, bone marrow failure, chronic kidney disease, chronic heart failure, autoimmune disorders, critically illness in the ICU, iron-deficiency and healthy subjects. We did not encounter suitable SRs that covered hemochromatosis, polycythemia, thalassemia, sickle cell disease and other hemoglobinopathies and subtypes of anemias. For red cell transfusion only narrative reviews were included. No relevant reviews were encountered on the subjects of vitamin supplementation and phlebotomy. Details of studies included for review are shown in Supplementary Material S1. A full reference list of the excluded studies with reasons for exclusion is provided in Supplementary Material S2.

The individual assessment for each SR with the AMSTAR-2 critical appraisal tool is shown in Figure 3. The quality was found to be high in seven SRs, two SRs scored low and 24 SRs were appraised as of critically low quality. Furthermore, we plotted all quantitative data in funnel plots to assess possible publication bias (Supplementary Figures S1, S2). For physiological outcomes, we found no signs of publication bias (Egger’s test: p = 0.277). Patient reported outcome measures, however, showed clear signs of publication bias. The funnel plot in Supplementary Figure S1 has a very pronounced asymmetry, with less powerful studies finding larger effect estimates, suggesting substantial bias. This is supported by the Egger’s test: p < 0.001.

An overview of the baseline Hb-levels and the achieved Hb-increment by intervention per study is presented in Figure 4. The mean increase in Hb after ESA-treatment lies between 1 and 3 g/dL. Iron treatment and altitude training show mean increases in Hb of <1 g/dL. The colored areas in this figure indicate per intervention what increment per baseline-Hb has been studied. These areas do not overlap, creating a patchwork-like image. While the lack of overlap shows that we cannot accurately compare the interventions with one another, this supports the combining of these data: to get a more complete picture of the effect of various increments at different baseline-Hb’s.

Twenty citations that investigated the effect of ESA treatment were eligible for evaluation (Supplementary Materials S1, S3). The SRs were published between 2003 and 2020, and AMSTAR-2 quality of reporting ranged from critically low to high, the GRADE quality of evidence was generally low, ranging from very low to high. ESAs were compared with placebo, lower targets or no ESAs. Results are depicted in Figures 5, 6. Due to heterogeneity of the investigated subjects, we summarized the outcomes per patient group (Supplementary Figures S3, S4).

Five SRs evaluating iron treatment were eligible (Supplementary Materials S1, S3) (Gurusamy and Richards, 2013; Pasricha et al., 2014; Burden et al., 2015; Clevenger et al., 2016; Houston et al., 2018). Treatment with intravenous, oral and intramuscular iron was compared with placebo, iron oral, or no iron in anaemic and non-anaemic patient groups. The articles were published between 2013 and 2018. AMSTAR-2 scores ranged from critically low to high and the evidence was graded very low to moderate. In Figures 5, 6 the effects of iron on various outcomes are depicted. Clevenger et al. investigated the effect of iron on QoL in anaemic patients without chronic kidney disease and found that these patients benefit significantly from parental iron (Hb MD = 1.04 g/dL, CI: 0.52, 1.57, n = 1744), but oral application (Hb MD: 0.91 g/dL, CI: 0.48, 1.35, n = 851) did not lead to a significant difference in QoL. Pasricha et al. (2014) evaluated studies with women of reproductive age, anaemic and non-anaemic together, and concluded that oral iron supplementation increased the Hb-level and had a significantly positive effect on VO₂max (MD = 2.35, CI: 0.82, 3.88) and the heartrate (MD −4.05 CI:−7.25, −0.85). Pasricha et al. (2014) also mentioned the role of iron in the mitochondrial respiratory chain, suggesting that improved tissue iron might improve exercise capacity, independent of a Hb-modulation. The exercise capacity showed a positive trend, although this was not significant. Two SRs evaluated iron treatment in iron deficient non-anaemic groups. These groups significantly benefited from the iron-induced Hb-change for the outcomes VO₂max and fatigue. The exercise capacity however, again did not change significantly.

Two narrative SRs were eligible for evaluation with regard to transfusion (Supplementary Materials S1, S3). Nielsen et al. (2017) evaluated the effect of transfusion on tissue oxygenation in critically ill ICU patients with anemia compared to similar controls. They found no significant effect on tissue oxygenation (Figure 5). However, nine of the primary studies revealed a consistent pattern—patients with abnormal tissue oxygenation prior to transfusion improved with transfusion, whereas those with normal pre-transfusion indices did not. The authors conclude that this supports a strategy for future clinical trials which focuses on abnormalities of tissue oxygenation or microcirculatory parameters rather than hemoglobin levels as the trigger for transfusion. Solheim et al. (2019) evaluated the effect of autologous transfusion on the VO₂max and exercise capacity in trained and untrained adults, compared to controls that did not receive an autologous blood transfusion. They reached the conclusions that the magnitude of change in Hb explains the increase in VO₂max, and reinfusion with as little as 135 mL packed red blood cells increases time-trial performance by elevating CaO₂. The AMSTAR-2 scores were critically low for both SRs, the GRADE quality of evidence was very low.

Of the retrieved citations evaluating restrictive (7.0–9.0 g/dL) and liberal thresholds (9.6–12.0 g/dL), two SRs were eligible for review (Supplementary Materials S1, S3). The SRs were published in 2015 and 2017, and both evaluated the effect of transfusion thresholds on quality of life. Estcourt et al. (2017) concluded to be very uncertain whether a restrictive threshold reduces quality of life in a population with hematological malignancies: one trial, 89 participants, fatigue score: restrictive median 4.8 (IQR 4–5.2); liberal median 4.5 (IQR 3.6–5) (very low-quality evidence). Gu et al. (2015) aimed to evaluate fatigue in a myelodysplastic syndrome population, but reported only one study which was terminated early due to poor recruitment and reported no results for their primary outcome measure. Both SRs are of high quality as scored with the AMSTAR-2 tool, but produced very low GRADE quality evidence due to small sample sizes and low quality primary studies.

Two SRs, published in 2012 and 2013, evaluated altitude training in athlete populations (Supplementary Materials S1, S3). Whereas Saunders et al. (2013) investigated several types of altitude training together, Lancaster and Smart (2012) focused only on live-high train-low altitude training. The AMSTAR-2 scores for both SRs were critically low, and GRADE quality of the evidence was low. Both SRs reported a small increase in Hb (0.57–0.69 g/dL and ∼3%). Lancaster reported an improvement in maximal heart rate (MD = 1.77, CI:0.50, 3.03) and both SRs concluded that altitude training improves VO₂max significantly (MD = 1.51, CI: 0.44, 2.58 and slope = 0.48, CI: 0.30, 0.67) in athletes, compared to normoxic training (Figures 5, 6).

Two SRs evaluating blood donation were eligible for review (Supplementary Materials S1, S3). The AMSTAR-2 scores of both articles were critically low. The GRADE quality of the evidence ranged from very low to low. (Van Remoortel et al. (2017) evaluated several studies that compared healthy adults that underwent blood donation before and after donation, and one study that used sham bleeding as comparator. Johnson et al. (2019) performed a similar SR with comparable eligibility criteria. The results and conclusions presented by Johnson et al. (2019) were not completely understood, and Johnson et al. (2019) could unfortunately not be reached for comments. Intriguingly, Johnson and Van Remoortel reached contrasting conclusions. With much overlap in primary studies, we selected the review with the best data synthesis on the basis of the JADAD decision algorithm. Hence, we selected Van Remoortel’s SR. The latter concluded that the VO₂max and time to exhaustion decreased significantly after blood donation. Maximum power showed no significant difference pre- and post-donation (Figures 5, 6).

Systematic reviews evaluating ESAs make up for a large part of the included evidence. Therefore, we decided to additionally evaluate what the outcome of this umbrella review would have been without these SRs evaluating ESAs. A forest plot, funnel plot and an overview table of results, all without ESA-related trials, are shown in Supplementary Material S5a (Supplementary Figures S5–S8). Remarkably, very little evidence remains regarding PROMs and regarding anaemic patients in general. Too few reviews could, however, be included to create a funnel plot for PROMs in this sensitivity analysis. Non-etheless, there is no evident change of significance direction notable.

Furthermore, for a second sensitivity analysis, we selected only high-quality SRs. In doing so, no data for non-anemia were left (Supplementary Figures S9–S12). For anaemic patients, however, the outcomes again remained similar in this analysis.

The reporting of evidence is generally scored by the AMSTAR-2 checklist to be of critically low quality, and meta-analyses with substantial sample sizes are mostly restricted to one intervention: ESAs. We did not impose any restrictions by language or publication year and used very wide eligibility criteria to decrease the chance of selection bias. However, because of the wide eligibility criteria for this overview, the included populations are very heterogeneous. It is therefore emphatically not our intention to compare the included reviews with each other. Likewise, all quantitative outcomes—from VO₂max to PROMs—have been presented together in the forest plots in Figure 6. The intention here is to give a visual overview of the standardized effect sizes of all results, not for meta-analytic purposes. Moreover, due to incomplete reporting in several reviews, multiple assumptions had to be made to compute SMDs from the reported effect sizes, limiting possibilities and accuracy of deeper analyses. Non-etheless, we believe this gives a good overview of the available evidence. Furthermore, it was commonly noted by the included narrative SRs that studies and outcomes were often too heterogenous to be compared in meta-analyses. International cooperation and standardization of methods for producing and reporting outcomes would thus be helpful to reach more meaningful conclusions and to compare outcomes with other trials. As per custom for umbrella reviews, we chose to only include systematic reviews and meta-analyses to compare and contrast published reviews and to provide an overall examination of a body of information that is available for this topic (Hartling et al., 2012). Therefore, primary studies may exist that have assessed areas that we identified as gaps in evidence. Reviewing primary studies in these identified gaps, preferably in a meta-analysis, may be a next step for future research. Furthermore, the Hb-increment was higher in groups with a lower baseline Hb than in groups with a higher baseline Hb (Figure 4). This is important to realize when interpreting the forest plots in Figure 6: the effect of a Hb-change seems more pronounced at lower baseline-levels. However, as the increment is higher at lower baseline Hbs, a dose-dependency effect might have confounded this result. Other bias may have been introduced by side-effects of the investigated therapies. With current techniques it is impossible to determine whether an improvement in QoL is caused solely by a Hb modulation. It is conceivable that other metabolic effects of iron supplementation or treatments like ESAs may cause this change in QoL.

Our meta-bias analysis suggests that reviews analyzing PROMs are prone to publication bias, whereas physiological outcomes do not show meta-biases in the funnel plot or sensitivity analysis. Indeed, PROMs are a subjective tool, and therefore easily biased if the methodology of the primary study lacks blinding and such. Possibly, larger primary studies were methodologically sounder than those with small sample sizes, in which case, the funnel plot is a result of information bias. The other option is that non-significant meta-analyses were not published. Either way, as this issue did not occur with the physiological outcomes, we recommend for future primary studies to at least incorporate physiological outcomes, preferably in combination with PROMs. With fast emerging wearables, eHealth and remote monitoring options (Tonino et al., 2019), incorporating physiological outcomes has rapidly become more feasible and may provide intriguing new insights.

Our sensitivity analyses underline the fact that there is a lack of high-quality evidence to support guidelines. Selecting high quality SRs only, the results for anaemic patients were similar. However, there were no high quality SRs that assessed Hb-changes in non-anaemic patients. After removing ESA-related reviews there were hardly any reviews left that assessed PROMs and very little evidence for anaemic patients in general. Since ESA-studies are predominantly funded by pharmaceutical companies, this demonstrates the importance of funding by pharma for research. Even though pharma-funded studies may be prone to conflicts of interest, they do contribute significantly to current knowledge. Non-etheless, independent high-quality evidence is very scarce.

In search of evidence based guidance for hemoglobin optimalization, whether for transfusion thresholds, ESA-targets, hyperviscosity management or optimal athletic performance, we found that for a quantitative benefit-harm assessment, we lack the data to draw meaningful conclusions. In some cases, as in ESA treatment in chronic kidney disease, there are sufficient data for a moderately certain conclusion that there is benefit in increasing the Hb-level up until about 12 g/dL for both patient reported and physiological outcomes. With the heterogeneity, confounding and scarce high quality evidence, it is difficult to compare and/or extrapolate these findings to Hb-levels >12 g/dL. Additionally, due to data suggesting an increased thrombotic risk above 12.0 g/dL in ESA-treated chronic kidney disease-patients, we would advise against Hb-targets >12.0 g/dL in this patient group (Palmer, 2010). For other treatments and patient groups, and the subjects of donation and phlebotomies, the evidence is thin or non-existent, and no conclusions about optimal Hb-levels can be drawn.

Critics may state that the Hb level is an inadequate measure to begin with. Indeed, the total Hb-mass is not affected by changes in plasma volume and would be more stable and render more accurate insights (Prommer et al., 2008). However, most studies use the Hb level as a proxy for total Hb-mass due to its ready availability. And while less accurate, it can be insightful in larger samples. In the future, however, we need to look at alternative objective parameters for oxygen transport optimization. With current technology, wearable devices are able to remotely measure basic physiology parameters like heart rate, heart rate variability, respiration rate, tissue oxygenation, capillary oxygenation and many more. Collecting these parameters and using big data algorithms may be a step towards a more accurate assessment of the patients’ treatment requirements when it comes to oxygen transport optimization. As this umbrella review shows, we still have many data gaps to fill.

In conclusion, this overview has revealed many knowledge gaps due to a lack of high-quality evidence. For chronic kidney disease patients, a clinically relevant benefit of increasing the Hb levels up until at least 12 g/L on PROMs and physiological outcomes was found. However, a personalized approach of patient care remains a necessity due to the many patient-specific factors that can affect the tolerance to deviating Hb-levels. In non-anaemic patients, the evidence suggests that increasing the Hb-level improves physiological and patient reported outcomes, but the effect is less pronounced. Decreasing the Hb-level has a clear inverse effect on physiological outcomes in non-anemic subjects. For supraphysiological Hb-levels, no eligible reviews were found. Because of the many knowledge gaps in general in the reported data sets, we strongly encourage future trials to incorporate physiological outcomes as objective parameters together with subjective, but still very important, patient reported outcome measures.