Fax: (978) 371 2445
Anemia as an independent prognostic factor for survival in patients with cancer
A systematic, quantitative review
Article first published online: 13 JUN 2001
Copyright © 2001 American Cancer Society
Volume 91, Issue 12, pages 2214–2221, 15 June 2001
How to Cite
Caro, J. J., Salas, M., Ward, A. and Goss, G. (2001), Anemia as an independent prognostic factor for survival in patients with cancer. Cancer, 91: 2214–2221. doi: 10.1002/1097-0142(20010615)91:12<2214::AID-CNCR1251>3.0.CO;2-P
- Issue published online: 13 JUN 2001
- Article first published online: 13 JUN 2001
- Manuscript Accepted: 12 MAR 2001
- Manuscript Revised: 1 MAR 2001
- Manuscript Received: 10 JUL 2000
- Janssen Ortho, Inc.
- prognostic factors;
Anemia is common in cancer patients, although the prevalence is influenced both by the type of malignancy and the choice of treatment. Individual studies have compared the survival of patients with and without anemia and have shown reduced survival times in patients with various malignancies, including carcinoma of the lung, cervix, head and neck, prostate, lymphoma, and multiple myeloma. The objective of this study was to systematically review, to summarize, and to obtain an overall estimate of the effect of anemia on survival in patients with malignant disease.
A comprehensive literature review was carried out using the MEDLINE data base and reviewing the reference lists from published studies. Two hundred papers were identified. Of these, 60 papers that reported the survival of cancer patients according to either hemoglobin levels or the presence of anemia were included. Among these papers, 25% related to patients with lung carcinoma, 17% related to patients with head and neck carcinoma, 12% related to patients with multiple myeloma, 10% related to patients with prostate carcinoma, 8% related to patients with cervicouterine carcinoma, 7% related to patients with leukemia, 5% related to patients with lymphoma, and 16% related to patients with other types of malignancies.
The relative risk of death increased by 19% (95% confidence interval, 10–29%) in anemic patients with lung carcinoma, by 75% (37–123%) in anemic patients with head and neck carcinoma, by 47% (21–78%) in anemic patients with prostate carcinoma, and by 67% (30–113%) in anemic patients with lymphoma. The overall estimate increase in risk was 65% (54–77%).
Anemia is associated with shorter survival times for patients with lung carcinoma, cervicouterine carcinoma, head and neck carcinoma, prostate carcinoma, lymphoma, and multiple myeloma. Cancer 2001;91:2214–21. © 2001 American Cancer Society.
Anemia is a common finding in cancer patients.1 Up to 30% of patients with tumors have been reported to suffer from anemia.2 This can be related to the disease process itself or to its treatment, whether it is chemotherapy or radiotherapy. The factors associated with anemia itself encompass disorders of iron metabolism,3 reduced numbers of erythroid progenitor cells in the bone marrow, increased levels of inflammatory cytokines,4 extracorpuscular hemolysis, catabolism of patients with tumor burden, and relative deficiency of erythropoietin.2
Generally, the anemia is considered within the category of anemia of chronic diseases as normocytic and normochromic.3 The association of increased hemoglobin levels with increased quality of life has been demonstrated by randomized, controlled trials and large, community-based studies.5–8 Some authors have suggested that maintenance of hemoglobin levels may improve survival, mainly in patients with extensive spread of disease.
Some authors have suggested that the maintenance of hemoglobin at normal levels would improve the quality of life9, 10 and survival, mainly in patients with extensive spread of disease.4 Individual studies have compared the survival of patients with and without anemia and have found reduced survival times in patients with a wide range of malignancies, including those of the lung, cervix, head and neck, and prostate; lymphoma; and multiple myeloma. Despite the number of individual studies addressing the topic, no quantitative, comprehensive review has been published examining the correlation between anemia and survival in cancer patients and no overall estimate has been published.
MATERIALS AND METHODS
Two researchers independently searched the MEDLINE data base using the keywords hemoglobin, hematocrit, and anemia and survival, prognosis, survival rates, survival, hazard, death, mortality, hazard rate ratio, Cox proportional hazard and cancer, neoplasia, and oncology in various types of malignancies. The search was carried back 30 years without any language limitations. The references from all reviews or original papers also were searched to complete the list of potential papers for assessment.
The two researchers also independently reviewed the list of articles for potential inclusion. All disagreements were then resolved by consensus. The decision to include an article was not related to the study's results. The articles selected met the following inclusion criteria: diagnosis of cancer and survival data stratified by hemoglobin levels or anemia or otherwise reported in a manner that permitted the results to be examined in this way. Also, articles that did not report the total number or proportion of anemic patients were excluded. All studies that were reported only as abstracts were excluded from the analyses.
Studies were reviewed independently by the two investigators, and any differences were resolved by consensus. Information was retrieved on study design, patient age, gender, types of malignancy, hemoglobin levels, hematocrit levels and/or anemia, survival times or rates at particular intervals or overall, hazard rates, rate ratios (unadjusted and adjusted estimates with their corresponding variables), and confounding or modifier variables, such as disease severity, treatment, and blood transfusions. We accepted the primary study authors' definitions for each malignancy.
All data were entered into an electronic data base for the statistical analyses. Because the studies reported survival in different ways, we transformed cumulative survival over a given period to hazard rates using
where λ is the hazard over the period, and S(t) is the survival (or 1 − M(t); the mortality) over time t. To transform the median survival to hazard, a constant hazard over time was assumed.
We calculated the hazard rate ratio (HRR; hazard with anemia and/or hazard without anemia) for each study. The analysis of HRR was carried out according to the information available. If the authors reported HRR obtained from a univariate analysis (or if we calculated the HRR from survival estimates) that did not consider other factors, then these estimates were labeled unadjusted HRRs (uHRRs). If the authors reported an HRR in light of other factors (using a Cox proportional hazards model) or reported a β coefficient for anemia to derive the HRR, then these estimates were considered adjusted HRRs (aHRRs). The aHRRs were extracted from articles irrespective of the prognostic factors for each malignancy.
For the uHRR, the hazard rates were pooled using the proportion of the total number of patients as the weight for each study. Because the articles did not consistently report any measure of variation, we could not estimate the confidence intervals of unadjusted estimates. To reduce the heterogeneity among studies, we stratified by disease, although it was not possible to stratify by severity because of insufficient data.
To estimate the aHRR and corresponding 95% confidence interval (95% CI) by study, we used the published information of β coefficient and standard error. In this case, it was possible to use the inverse of the variance as source of weights to pool the aHRR. The 95% CI for the overall aHRR (âHRR) was calculated using
To assess the possibility of publication bias, the uHRRs were plotted against the sample size (inverted funnel plot) and the year of publication. Also, a cumulative meta-analysis, a type of meta-analysis in which the combined point estimate of an effect is computed sequentially by adding one study at a time in a prespecified order,13, 14 was done by type of disease.
Two hundred articles were identified, with the majority published in English. We excluded 30 reviews, 108 articles with insufficient data (no hemoglobin data or no survival data), and 2 articles that reported only disease free survival.2 Sixty studies were included, and, among them, two Japanese,15, 16 one Chinese,17 one Polish,18 and one French study met the inclusion criteria.19 Twenty-five percent of the studies focused on patients with carcinoma of the lung, 17% focused on patients with carcinoma of the head and neck, 12% focused on patients with multiple myeloma, 10% focused on patients with prostate carcinoma, 8% focused on patients with cervicouterine carcinoma, 7% focused on patients with leukemia, 5% focused on patients with lymphoma, and 16% focused on patients with other malignancies (Table 1).
|Type of disease||Reference(s)||No. of studies (%)||Unadjusted HRR||Adjusted HRR (95% CI)|
|Lung||15–17, 21–32||15 (25.0)||1.54||1.19 (1.10–1.29)|
|Head and neck carcinoma||18, 33–41||10 (17.0)||2.35||1.75 (1.37–2.23)|
|Multiple myeloma||19, 42–47||7 (12.0)||4.47||NR|
|Prostate carcinoma||48–53||6 (10.0)||1.78||1.47 (1.21–1.78)|
|Cervicouterine carcinoma||20, 54–57||5 (8.0)||2.61||NR|
|Lymphoma||62–64||3 (5.0)||3.74||1.67 (1.30–2.13)|
|Renal carcinoma||65||2 (3.0)||1.90||NR|
|Ovarian carcinoma||1, 66||2 (3.0)||1.40||NR|
|Colorectal carcinoma||67, 68||2 (3.0)||1.83||NR|
|Total||—||60 (100.0)||Overall, 2.33||Overall, 1.65 (1.54–1.77)|
In relation to the study type, we found 39 cohort studies, 19 randomized clinical trials, and 2 case-referent studies. The majority of the participants were receiving current chemotherapy and/or radiotherapy treatment at the time that the hemoglobin or anemia status was established. Generally, the authors did not report whether a patient received transfusion.
A key issue was the criteria used by the authors to diagnose anemia. Some authors reported only anemia versus no anemia without providing the criteria used, others stratified by hemoglobin levels, and the cut-off point varied from 8.5 g/dL to 14.0 g/dL. Some authors used different hemoglobin cut-off points for male patients and female patients but did not include the gender distribution in the article.
The median survival time varied according to the disease as well as the presence of anemia. For patients with anemia, the median survival ranged from 4 months (lung carcinoma and multiple myeloma) to 96 months (prostate carcinoma) compared with 7 months (lung carcinoma) to 120 months (prostate carcinoma) for patients who were not anemic. In all studies, the median survival was longer for patients without anemia compared with the anemic patients (Fig. 1).
When the uHRR was calculated for each study, we found that it varied according to the disease, but it was always higher for patients with anemia, regardless of the type of malignancy. For example, the uHRR ranged from 1.2 to 2.6 in lung carcinoma studies, from 1.3 to 1.4 in ovarian carcinoma studies, and from 1.5 to 9.0 in multiple myeloma studies. For some types of malignancies, like multiple myeloma and lymphoma, the average uHRR was > 5. It is possible that this reflects the involvement of bone marrow and other reticuloendothelial sites in these diseases. Another possibility may be that it reflects disease severity; however, because the articles did not report a correlation between anemia and survival by disease severity, we could not examine this possibility. We pooled all unadjusted hazard rates to obtain the overall uHRR by disease weighted by sample size (Table 1).
For the studies that reported the aHRR obtained from a Cox proportional hazards multivariate model or the corresponding β coefficient, we also compared the aHRR and 95% CI by disease. The aHRRs were significantly > 1.0 in the majority of the studies, ranging from 1.1 to 2.5 for the individual studies. Some studies had wide 95% CI levels because of the small number of participants in the study. The overall aHRR and 95% CI was estimated for diseases that were reported in more than one study, and they were statistically significant for all four of the diseases studied: lung carcinoma, head and neck carcinoma, prostate carcinoma, and lymphoma. The overall aHRR was 1.65 with a 95% CI of 1.54–1.77 (Table 1).
The cumulative HRRs showed a positive relation between anemia and survival by type of disease. The cumulative HRRs ranged between 3.18 for lymphoma and 7.04 for lung carcinoma.
The estimates also were plotted against sample size to observe the effect on the distribution of the HRR estimates (inverted funnel plot) (Fig. 2). There were no differences between small studies and large studies in relation to either the uHRR or the aHRR. There was no evidence of positive publication bias, because the smaller studies were not skewed toward a higher HRR estimate and covered a wider range of values than the larger studies.
The aHRR estimates were published only for models in which anemia was a significant predictor. Nevertheless, because the main objective of the majority of papers was not related to the anemia-survival relation, publication bias may not have arisen for the articles that reported aHRR. In addition, the uHRRs were estimated for 39 studies, and these also showed no evidence of a substantial publication bias.
Anemia is common in cancer patients,1, 20 and it has been associated with lower survival. In our study, ≈33% of patients were diagnosed as anemic, and the median survival was reduced by 20–43%. Individual studies have reported this association in patients with a variety of diseases. Therefore, this meta-analysis was carried out to combine the survival information from individual studies to get an overall estimate of the association between survival and anemia.
From the 60 articles that were included, the majority reported median survival, and 21 articles (35%) published the impact of anemia on survival adjusted for other factors, such as clinical stage and performance status. The type of disease did influence survival time; therefore, the results were reported separately for each type of malignancy.
In the current study, we found that anemic patients, compared with nonanemic patients, with carcinoma of the lung, prostate, and head and neck as well as lymphoma had a significant reduction in survival. The association remains significant even after adjusting for other factors. For example, there is a 19% greater risk of death if a patient with carcinoma of the lung has anemia (95% CI, 1.10–1.29). Is the anemia an independent predictor of survival, regardless of disease severity, or does it modify the relation between disease severity and survival? Anemia may be an independent predictor of survival, because the published Cox proportional hazards models were adjusted for disease stage and/or severity. This strongly implies that anemia is an independent predictor of survival. In this meta-analysis, however, it was not possible to determine whether anemia modifies the relation between disease severity and survival.
The aHRR, although it still was significantly greater than 1.0, was less than the unadjusted rate, suggesting that some of the effect was due to imbalances in other factors. Nevertheless, some of the regression models used to control for other factors included variables like performance status that actually may reflect the anemic status. If this is true, then the models are partitioning the effect of anemia into two variables, thus underestimating the true HRR.
The small number of studies addressing this topic for each type of malignancy is the major limitation of this meta-analysis. We included both randomized clinical trials and observational studies. The randomized clinical trials were not designed primarily to examine the effect of anemia, but they retrospectively analyzed the prognostic factors in patients with cancer. The observational studies were mostly of the cohort design, but there were two case-referent studies, a design commonly used to study prognosis in the oncology setting.
The criteria used to classify a patient as anemic varied between individual studies. For this meta-analysis, hemoglobin levels < 8.5–14.0 g/dL were used as well as a requirement for transfusion to classify patients as anemic. Therefore, this analysis pools the impact across a range of hemoglobin levels. The information in the reports was too limited to analyze the relation between different hemoglobin levels and survival.
Most of the studies were observational; therefore, a certain degree of heterogeneity among populations and study design may have influenced the results. One limitation of the unadjusted hazard rate estimates was that they were derived from median survival times assuming a constant hazard. Another issue related to our results may have been publication bias, because the majority of authors who reported on anemia in the aHRR for anemic patients did so because it was a statistically significant factor. We explored this possibility by estimating the uHRR for each study and pooling by type of disease, with results and direction similar to those in the aHRR. Also, we plotted both aHRR and uHRR by year of publication and sample size without any clear evidence of publication bias. The potential mechanisms accounting for this reduction in survival are multiple and include reduced free radical formation in the tumor leading to radioresistance and modification of hypoxia-inducible genes (e.g., vascular endothelial growth factor, p53, HIF-1, etc.).
One limitation of this study is that it was not possible to determine whether anemia is the cause of decreased survival or a surrogate for other adverse factors. It is possible that anemia may be due to direct interference by the malignancy in red blood cell production, like what is seen in patients with multiple myeloma or lymphoma. Alternatively, it may be an indirect effect of high tumor burden and/or biologic effects, for example, through cytokine interference with red blood cell production. Furthermore, patients were receiving chemotherapy or radiotherapy; therefore, it is difficult to implicate anemia as a direct cause of decreased survival. However, from this study, it was clear that patients with anemia–as either a direct cause or an indirect cause–have lower survival.
Given the serious reduction in patient survival implied by anemia, it appears that the potential impact of its correction should be considered strongly by clinicians, although which treatment to employ for anemic patients is not obvious. An apparent choice is transfusion, but this presents problems, such as availability of blood, risk of infections, cost, and (very important for some patients) a poor prognosis in and of itself. Another treatment option that has become available recently for anemic patients with malignancies is recombinant human erythropoietin.
In conclusion, this study showed that anemia is a strong predictor of poorer survival in cancer patients. Although additional studies are necessary to determine whether treating anemia improves survival, its impact should be considered when planning and analyzing the impact of novel treatments for patients with malignant disease.
- 2Erythropoietin therapy in cancer patients. Cas Leuk Ces 1995; 20: 647–50., .
- 8Once-weekly dosing of epoetin alpha is similar to three times weekly dosing in increasing hemoglobin and quality of life. Proc Am Soc Clin Oncol 1999; 18: 574A., , , , .
- 12Point estimation of overall effect and interval estimation of effect. In: KleinbaumDG, KupperLL, MorgensternH, editors. Epidemiologic research. Principles and quantitative methods, 1st edition. New York: Van Nostrand Reinhold, 1982: 340–76., , .
- 33The impact of blood hemoglobin content on the outcome of radiotherapy. Strahlenther Onkol 1998; 174(Suppl IV): 31–4., , .