This study was the first attempt to explore the safety of a restrictive transfusion strategy in surgical patients with seven types of cancer using observational data. Our results showed that when heterogeneity was well controlled using tailored designs and analyses, perioperative RBC transfusion in surgical patients with cancer was no longer associated with death and complications. A transfusion threshold of 8 g/dL was potentially feasible for total cancers, specifically 6 g/dL for colorectal cancer, at least in terms of short-term outcomes. These values are lower than the liberal transfusion threshold of 9 g/dL suggested for patients with cancer having major abdominal surgery [6]. Our results suggest a substantially lower RBC requirement; for instance, according to a rough estimate based on the transfusion pattern in this study and national data [27], the overall RBC demand in China could be reduced by 0.1 million units annually if a transfusion threshold of 8 g/dL rather than 9 g/dL were applied for all patients with cancer in surgery.
Three main flaws affect the validity of evidence from existing observational transfusion studies. First, most observational studies on transfusion are retrospective, leading to a passive study design regarding patient selection, study variable collection, and sample size [11]. Second, despite loud calls to control for the effects of heterogeneity, many studies have neglected to assess heterogeneity or have misused the P value approach to assess heterogeneity [28]. The use of a P value to quantify heterogeneity can be affected by the sample size; this is not the case with the use of the SMD. Third, many studies are now using PS analysis as post-hoc randomization to balance confounding, but the conditions for the use of this method have rarely been considered, which leads to still great patient heterogeneity (SMD: 0.2–53.2%) even after using PS analysis [29, 30]. This is because the PS analysis requires that the overlapping range of the PS distribution in the two groups is sufficiently large; in other words, when there is a systematic bias between groups, this is hard to balance only via the statistical method itself [31]. In this case, other approaches must be applied beforehand until the heterogeneity is properly addressed. In this study, the exclusion prior to the statistical analysis offers an effective way to reduce patient heterogeneity (SMD: 0.8–6.3%) because patients with strong indications for transfusion (severe anaemia and intraoperative massive haemorrhage) and irrational transfusion at higher concentrations (haemoglobin ≥ 10 g/dL) were naturally excluded [19].
In our study, 59.37% of the 6055 surgical patients with cancer had perioperative anaemia (haemoglobin < 12.0 g/dL), which was similar to the proportion of patients after various cancer treatments in the European Cancer Anaemia Survey (haemoglobin < 12.0 g/dL, 42–63%) [32]. Anaemia is a common side effect of cancer treatment as well as cancer itself. A plummeting RBC count after cancer treatment with chemotherapy and radiation is not unusual. Bone marrow hematopoietic dysfunction directly impacts the body’s ability to produce RBCs, and several cytokines (e.g., tumour necrosis factor α and interferon γ) inhibit the synthesis of erythropoietin in the kidney, leading to diminished erythropoiesis [33]. In colorectal cancer, chronic microscopic blood loss into the digestive tract is a common symptom and is considered to be the primary cause of anaemia. The blood in the stool has time to degrade, so the stool often looks normal, and blood may not be noticed during defecation. However, the blood loss can build up slowly over time, leading to chronic anaemia. Thus, chronic anaemia may be one of the first signs of the development of colorectal cancer, more so than with any other type of cancer [34]. Patients with chronic anaemia may tolerate a low haemoglobin concentration through compensation. If the compensatory mechanism can maintain oxygen homeostasis and there are no obvious symptoms of anaemia, blood transfusion may be withheld, which may make a restrictive transfusion policy at a threshold of 6.0 g/dL possible for surgical patients with colorectal cancer [35]. However, previous studies have also reported the negative impact of preoperative anaemia on long-term cancer prognosis [36, 37]. Additionally, patients often require follow-up adjuvant radiotherapy and chemotherapy after tumour resection and are likely to be anaemic prior to or during adjuvant treatment. Previous studies have shown that anaemia may have a negative influence on radiotherapy and chemoradiation outcomes, and correction of anaemia is an important component of treatment for optimizing long-term outcomes, such as disease-free survival rates [38, 39]. Thus, a closer look at anaemia tolerance in surgical oncology patients is required, preferably regarding long-term and short-term outcomes under various management strategies (e.g., allogeneic red blood cell transfusion, iron therapy, erythropoiesis-stimulating agents).
Evidence is still lacking regarding the safety of a restrictive transfusion policy in the surgical cancer context [6, 25]. It is ethically unfeasible to conduct RCTs among certain patient subgroups of special clinical concern. Moreover, searching for a minimum tolerable threshold in a trial-and-error fashion hampers the speed of an RCT. In 2021, a registered RCT among postoperative oncology patients in critical care set its restrictive threshold to 7 g/dL [40], whereas in the present study, we identified a potentially feasible threshold of 6 g/dL for patients with colorectal cancer. Observational studies using daily healthcare practice data provide a unique opportunity to complement RCTs [41, 42]. However, a divergence exists between both the objective and findings of observational studies and those of RCTs regarding RBC transfusion [43]. We transferred the objective of this study to that of an RCT (the indication for blood transfusion) using a tailored design [20]. The core strategy involves refining how to define study participants. Rather than defining study participants according to a specific disease or surgical procedure, as in previous studies, we defined the study population according to a trigger haemoglobin range (6–10 g/dL). This new definition allows for greater flexibility to explore possible haemoglobin thresholds. Most importantly, this definition unites the objectives of observational studies and RCTs, paving the way for more efficient collaborative research. Another cornerstone of the study design is the use of a stable haemoglobin level in defining the exposure and control groups because the selection of controls is decisive for effect estimation [44]. Among the changing haemoglobin levels, the choice of the indicative one (last prior to transfusion; nadir without transfusion) was made to maximize the likelihood of reflecting the role of haemoglobin in the transfusion decision [45]. Overall, this design could efficiently cover patients with different types of cancer in our study, despite the varied patterns of perioperative transfusion, because the indication for transfusion is the same, namely, haemoglobin. This allowed us to propose different feasible transfusion haemoglobin thresholds simultaneously for patients with different cancers. Although haemoglobin is not the only trigger in determining the need for transfusion (physiological transfusion triggers, such as tissue oxygenation or microcirculation are also involved), a haemoglobin-based transfusion trigger is still the most commonly used trigger for blood transfusion in the current transfusion guidelines and daily clinical practice [46].
Several limitations of the present study should be noted. First, the data on each specific cancer were limited; therefore, we could not perform independent analysis for some types of cancer, and results of stomach cancer showed low precision, but the results of the different perspectives of stomach cancer do not contradict each other. Second, the study outcomes were combined (death and complications) owing to the low rate of specific outcomes; thus, the impact of transfusion on specific outcomes could not be assessed. Limited data availability also limited our analyses of long-term outcomes, but there is evidence showing that long-term outcomes, such as disease progression and survival, are less likely to be influenced by perioperative transfusion [47]. Finally, data on tumour staging and grading were unavailable in this study, which has a great impact on patients’ prognosis. Further studies with larger sample sizes and more detailed tumour information are needed to validate the findings of this study.