The association between albumin levels and survival in patients treated with immune checkpoint inhibitors: A systematic review and meta-analysis

1Immune checkpoint inhibitors (ICIs) have bocame a vital part of cancer care in the last decade. First, ipilimumab, an antibody against CTLA-4, and later several monoclonal antibodies against PD-1 or PD-L1, have demonstrated improved survival in almost all tumors and across different treatment lines, either as monotherapy or in combination with chemotherapy or targeted therapy. These agents act by removing the immunosuppression created by the T-cell exhaustive checkpoints in the tumor microenvironment to aid the immune system in fighting tumors more efficiently. This mechanism of action, independent of specific targets on a tumor, permits ICI use in almost all tumors; however, at the same time, it creates a dependence on the tumor microenvironment for efficacy and, therefore, variable outcomes.

2Despite the stunning rate of developments in therapeutic sites with ICIs, biomarker development has been relatively slow. Other than PD-L1 expression in non-small cell lung cancer, gastric cancer, and cervical cancer, and two tumor agnostic markers (microsatellite instability and tumor mutational burden), no biomarker has consistently aided the decision-making process with ICIs. Several issues have factored into the slow biomarker development with ICIs, including the requirement of tissue samples and complex platforms. Furthermore, tissue-based biomarkers have not been able to reflect the status of the immune system, which is the main driver of ICI efficacy. Recently, peripheral blood-based biomarkers have garnered significant interest as the indirect indicators of host immune status, and a significant body of data has been generated with several compound biomarkers measuring the anti-tumor activity of lymphocytes and the uncontrolled inflammatory pressure of neutrophil, platelets, and monocytes, the neutrophil-lymphocyte ratio (NLR), the platelet-lymphocyte ratio (PLR) and the pan-immune-inflammation value.

3Similar to peripheral blood-based biomarkers derived from blood count, albumin levels are significantly affected by inflammatory pressure, and lower albumin levels are seen in cases of chronic inflammatory disorders and cancer as a negative acute phase reactant. Additionally, albumin levels are correlated with nutritional status, with lower albumin levels acting as an important denominator of malnutrition and poor general condition. Due to these features, albumin could be used as a prognostic biomarker in patients with cancer. Several studies in patients treated with chemotherapy, radiotherapy, or targeted therapy have demonstrated a higher risk of mortality or progression with lower albumin levels. Similarly, albumin levels could also be used for prognosis estimation in ICI-treated patients. However, the data on the albumin in ICI-treated patients are limited, with heterogeneous patient cohorts, small sample sizes, and variable cut-offs. Therefore, we conducted a systematic review and meta-analysis to evaluate the association between survival outcomes and albumin levels in patients treated with ICIs.

4We conducted a systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidance (PRISMA). The PRISMA checklist could be found in the supplement. This protocol was registered with the PROSPERO registry (Registration Number: CRD42022337746). We used the Web of Science, PubMed, and Embase databases to systematically filter the published studies from inception to 1 June 2022. The selected MeSH search terms were “albumin” OR “Glasgow prognosis score” OR “prognostic nutritional index” OR “albumin-globulin ratio” AND “immunotherapy” OR “immune checkpoint inhibitors” OR “immune-checkpoint inhibitor” OR “immune checkpoint blockade”. Both albumin and albumin-based compound prognostic markers were selected as search terms to prevent missing relevant studies.

5We included studies that met the following inclusion criteria: 1) prospective and retrospective studies to investigate the prognostic effects of albumin levels in ICI-treated patients with cancer; 2) articles reporting the hazard ratio (HR) for overall survival (OS) and/or progression-free survival (PFS); and 3) articles for which the full text was available in English. The exclusion criteria were as follows: 1) duplicated articles; 2) chapters of books, case reports, editorial letters, review articles, and opinion papers; 3) animal studies; 4) studies including patients without cancer; 5) studies with clinical endpoints other than PFS or OS; 6) studies without data for HRs and confidence intervals (CIs); and 7) studies without a predefined albumin cut-off to define hypoalbuminemia.

6Two authors independently extracted the following data from the available studies (DCG, TKS) following the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines: lead author names, years of publication, study countries, tumor types, total numbers of patients, albumin cut-offs, and HRs with 95% CIs for OS and PFS. Due to the availability of studies in metastatic settings only, we collected PFS data for progression events. Any disagreements between the authors (DCG, TKS) during data collection were resolved by a discussion with the senior author (SY). The individual study qualities and risk of bias were evaluated independently by two authors (DCG, EE) using the Newcastle-Ottawa Scale.

7The study’s primary objective was to evaluate the association between OS or PFS and the presence of hypoalbuminemia in patients treated with ICIs. The secondary objective was to evaluate the association between the OS and the presence of hypoalbuminemia in subgroup analyses according to study location (United States, Europe, Far East), study sample size (<100, 100–200, >200), albumin cut-off (3.5 or other), and tumor type (NSCLC, urothelial cancer, GEC, MPM, HCC, basket cohorts).

8We performed the meta-analyses with the generic inverse-variance method with a random-effects model due to the high degree of heterogeneity across the studies. The principal summary measure was HRs with 95% two-sided CIs, and the heterogeneity within each subgroup was reported with I-square statistics. Furthermore, we conducted additional analyses witha fixed-effects model after the exclusion of studies that caused a high degree of heterogeneity. The possibility of publication bias was assessed by the visual inspection of funnel plots. The meta-analyses were conducted with Review Manager software, version 5.4 (the Nordic Cochrane Center, the Cochrane Collaboration, Copenhagen, Denmark), and p-values below 0.05 were regarded as statistically significant.

9Our systematic search retrieved a total of 2965 records. After removing the duplications (n = 1419), we screened the titles of the remaining 1546 records and excluded 1419 more records due to the following reasons: topic irrelevance (n = 944); review, case report, commentary, meta-analysis, or editorial (n = 379); animal studies (n = 70); and studies not in English (n = 26). We screened the titles and abstracts of the remaining 127 articles and excluded 91 records due to the lack of HRs for survival (n = 28) and the lack of separate reporting for albumin levels in studies evaluating albumin-based prognostic scores (n = 63). We evaluated the full texts of the remaining 36 articles and included these 36 studies in the meta-analyses. The PRISMA diagram for article selection is included in the supplement (Supplementary Figure S1).

10Thirty-six studies encompassing 8406 patients were included in the meta-analyses. Almost half of the studies were conducted in NSCLC cohorts (n = 15), followed by urothelial (n = 5), gastroesophageal (n = 5), and hepatocellular cancers (n = 3) (Table 1). In addition, three studies included basket cohorts. The sample sizes were variable and spanned from 44 to 1714. More than half of the studies were reported from the Far East (n = 15), followed by Europe (n = 12). A 3.5 gr/dL albumin cut-off was most frequently used in the included studies (n = 20), while eight studies evaluated the effect of albumin levels on survival by using albumin as a continuous measure. All but two studies reported HRs for OS, while data for PFS was available in 15 studies. Most studies tested prognoses in patients treated with ICI monotherapy, and nivolumab was the most commonly used ICI in the studies (n = 21) (Table-1). Most studies had a low or intermediate risk of bias according to the Newcastle-Ottawa Scale (Table 2).

11Thirty-four studies were included in the analyses for OS. In the meta-analysis, patients with lower albumin levels had a significantly increased risk of death (HR: 1.65, 95% CI: 1.52–1.80, p < 0.0001) than patients with higher albumin levels (Figure 1). The included studies had high degrees of heterogeneity (I2 = 89%). The important portion of the heterogeneity stemmed from the studies including albumin as a continuous parameter in the analyses. After the exclusion of these studies (n = 8), heterogeneity was decreased to 47%, and a fixed-effect meta-analysis after the exclusion of these studies demonstrated a consistent negative effect of low albumin levels on OS (HR: 2.17, 95% CI: 1.99–2.36, p < 0.0001) (Supplementary Figure S2). Sensitivity analyses, with the exclusion of individual studies, also demonstrated a significantly higher risk of death in patients with lower albumin levels.

12Subgroup analyses across several tumor types demonstrated a consistently higher risk of death in patients with lower albumin levels than in patients with higher albumin levels (Figure 2A). Similarly, subgroup analyses according to study location (HR: 2.48, 95% CI: 1.93–3.18, p < 0.0001 for the United States, HR: 1.68, 95% CI: 1.37–2.05, p < 0.0001 for Europe, and HR: 2.30, 95% CI: 1.63–3.24, p < 0.0001 for Far East) (Figure 2B) and sample size (HR: 2.87, 95% CI: 1.83–4.50, p < 0.0001 for sample sizes of <100, HR: 1.58, 95% CI: 1.33–1.87, p < 0.0001 for sample sizes between 100 and 200, and HR: 1.69, 95% CI: 1.51–1.89, p < 0.0001 for sample sizes over 200) (Figure 3A) demonstrated a negative association between lower albumin levels and OS. Additional subgroup analyses according to different cut-offs to define hypoalbuminemia demonstrated a significantly increased risk of death in patients with lower albumin levels (HR: 2.00, 95% CI: 1.74–2.30, p < 0.0001 for a cut-off of 3.5 gr/dL, and HR: 2.79, 95% CI: 2.30–3.39, p < 0.0001 for other cut-offs) (Figure 3B). Furthermore, in the subgroup analysis of eight studies using albumin levels as a continuous prognostic factor, every 1 gr/dL decrease in albumin levels was associated with a significantly increased risk of death by a factor of 10% (HR: 1.10, 95% CI: 1.05–1.16, p = 0.0002) (Figure 3B). The funnel plot evaluation demonstrated the possibility of publication bias, especially in studies with a sample size below 100 patients (Supplementary Figure S3).

13Fifteen studies were included in the meta-analyses for PFS. Nine studies reported significantly lower PFS in patients with lower albumin levels, while the association between albumin levels and PFS did not reach statistical significance in the remaining six studies. In the meta-analysis of fifteen studies, patients with lower albumin levels had an increased risk of progression or death compared to patients with higher albumin levels (HR: 1.76, 95% CI: 1.40–2.21, p < 0.001) (Figure 4). The included studies had a significant degree of heterogeneity (I2 = 84%). Sensitivity analyses conducted by the subtraction of individual studies demonstrated consistent results. Heterogeneity was reduced to 31% after the exclusion of studies using albumin levels as a continuous measure in the analyses. An additional fixed-effect meta-analysis conducted after the exclusion of studies causing heterogeneity demonstrated a significantly increased risk of progression or death in patients with lower albumin levels (HR: 1.75, 95% CI: 1.58–1.95, p < 0.001) (Supplementary Figure S4). No additional subgroup analyses were conducted due to the limited number of studies evaluating the PFS.

14In this meta-analysis, we observed a significantly increased risk or progression in ICI-treated patients with hypoalbuminemia. The adverse effect of low albumin levels in OS and PFS was consistent across tumor type, albumin cut-off for hypoalbuminemia, study country, and study sample size. To the best of our knowledge, the present study is the first meta-analysis evaluating the association between albumin levels and survival in patients treated with ICIs.

15Nutritional status is an essential denominator of patients’ general conditions and is closely related to survival outcomes with anti-cancer treatments. Laboratory parameters like albumin are inexpensive and readily available measures of cancer patients’ nutritional status. In addition, albumin is a negative acute phase reactant, and its levels decrease in chronic inflammatory states like cancer. While albumin production could be increased in the early stages of exposure to harmful insults like carcinogens, albumin production is significantly decreased in advanced cancer due to malnutrition and the inhibitory pressure of cytokines and chemokines, like C-reactive protein and interleukin-6 on the liver. Therefore, albumin levels could be a biologically plausible biomarker with the potential to reflect both the nutritional status and inflammatory pressure on patients with cancer and decreased levels of albumin could reflect a more advanced cancer stage and a more dismal prognosis. Furthermore, the measurement of albumin levels could be preferable to the readily available liver enzymes as a prognostic biomarker. The lower albumin levels were independently associated with the presence of cancer, while the levels of bilirubin, alanine transaminase, and aspartate transaminase were similar in patients with or without cancer in the pivotal Glasgow Inflammation Outcome Study. Pavo et al. demonstrated that only albumin and butyrylcholinesterase levels were associated with all-cause mortality in cancer patients with non-liver primaries and patients without hepatic involvement. These data support the robust role of albumin levels as a prognostic biomarker in patients with cancer compared to other candidate biomarkers related with hepatic function.

16Several studies in patients treated with chemotherapy, surgery, targeted therapy, or radiotherapy demonstrated lower OS and PFS in cancer patients with lower albumin levels. Furthermore, compound scores created by mixing albumin levels with positive acute reactants like CRP, as in the Glasgow prognostic score or globulin in albumin-globulin ratio, could also aid in prognosis estimations in patients with cancer and should be thoroughly investigated. In contrast, the data on the association between albumin levels and prognosis are still limited in ICI-treated patients. The available literature on ICI-treated patients is mainly focused on NSCLC cohorts and includes only patients with advanced-stage disease, while ICIs have become standard adjuvant treatment options for melanoma, NSCLC, bladder cancer, and esophageal cancer. Additionally, most available studies on the prognostic role of albumin levels in ICI-treated patients have limited or absent data regarding previously established ICI-efficacy biomarkers like PD-L1 and tumor mutational burden levels. Therefore, additional data are needed to delineate the benefit of albumin levels for treatment selection in clinical practice and clinical trial stratification. Furthermore, available studies were significantly heterogeneous regarding albumin level cut-offs. They used various dichotomous cut-offs, as well as albumin levels as a continuous biomarker, limiting the generalizability of albumin levels as a biomarker. Future studies should focus on the performance of variable, pre-defined cut-offs and the possible use of albumin levels as a continuous biomarker. Additionally, separate reporting for albumin levels as a continuous biomarker, in addition to dichotomous cut-offs, should be encouraged.

17The present meta-analysis is subject to several limitations inherent to the methodology used and the characteristics of the included studies. First, we collected the reported HRs from the studies rather than conducting an individual patient data meta-analysis, so the presence of single-patient variables was not reported. The included studies were heterogeneous regarding tumor type, sample size, and albumin cut-offs, although we conducted subgroup analyses for these factors and observed consistent results. We did not conduct Bonferroni corrections for p-values, although the significance of most results was not expected to change, considering most of the analyses had a p-value of <0.0001. In addition, publication bias due to the higher publishing possibility of studies with positive results could not be excluded. Lastly, most of the included studies were conducted on patients treated with ICI monotherapy, while ICI-ICI, ICI-chemotherapy, and ICI-targeted therapy combinations have become standard care for most tumors, especially as first-line treatment.

18In conclusion, the available evidence demonstrates that albumin levels could be a promising prognostic biomarker in patients with metastatic cancer treated with immunotherapy. Further research is needed to delineate the role of albumin levels in patients treated with ICIs in adjuvant settings and patients treated with ICI-based combinations, as well as the possible benefit of therapeutic approaches to improve hypoalbuminemia in ICI-treated patients.