Prognostic value of blood-based protein biomarkers in non-small cell lung cancer: A critical review and 2008–2022 update

1 Introduction

Lung cancer is still the second most frequent cancer type, accounting for 11.4% of all cancers and serving as the leading cause of cancer mortality, with estimated 1.8 million deaths per year worldwide (18%) [1, 2]. Over the last decade, the incidence and mortality of lung cancer have steadily declined [3], mainly due to improvements in both diagnostic and therapeutic areas, such as the introduction of low-dose computed tomography for early lung cancer detection in high risk groups [4] and the approval of novel surgical and systemic treatment approaches including targeted tyrosine kinase inhibitor therapies (TKI) and immune checkpoint inhibitor (ICI) therapies [5, 6]. Consequently, the prognosis for early-stage non-small cell lung cancer (NSCLC) has improved in recent years, with a 5-year survival rate of 72% for adeno-cell (LUAD) and 48% for squamous-cell lung cancer (LUSC) [7, 8]. However, 55% of NSCLC patients continue to be diagnosed with unresectable advanced stages IIIB to IV, which are associated with a 5-year survival rate of only 9.5% [9] and a median survival of 8 to 18 months [10–12]. The advent of targeted and ICI therapies, as well as of new combination regimes [6], has also steadily improved survival in late-stage disease [13]. Notably, for patients ineligible for targeted or ICI therapies, combination chemotherapy regimens remain the recommended systemic therapy for LUSC and LUAD [14, 15].

In addition to molecular classification of lung tumors, for precise patient stratification using predictive “companion diagnostics” that indicate the likelihood of response to specific targeted or ICI therapies [16, 17], patient guidance involves estimating overall prognosis and individually monitoring therapy response as well as post-therapeutic surveillance using radiological and biochemical biomarkers [18, 19].

At present, considerable efforts are devoted to developing predictive molecular diagnostics, such as screening for tumor-specific genomic alterations in EGFR, ALK, ROS1, BRAF, NTRK1/2/3, RET, MET genes, for tumor mutational burden (TMB), mismatch repair and microsatellite instability amongst others, that are assessed in tumor tissue and on cell-free tumor DNA (ctDNA) circulating in the blood plasma [19–26].

To estimate prognosis, clinical markers, such as TNM stage, performance score, weight loss, lymph node involvement, metastases and the histologic subtypes [20, 27], as well as blood-based biochemical markers like routine lab parameters and tumor-associated proteins, provide valuable information in daily clinical practice. In the future, novel molecular markers like mRNA, miRNA, genetic and epigenetic changes in tumor and plasma DNA will further expand the array of prognostic markers [20, 29]. Regarding serum-based protein tumor markers (STM), numerous original studies and reviews have demonstrated prognostic relevance, particularly for cytokeratin-19 fragments (CYFRA 21-1), as well as carcino-embryonic antigen (CEA), neuron-specific enolase (NSE), squamous cell cancer antigen (SCCA), carbohydrate antigens 19-9 and 125 (CA 19-9 and CA 125) in NSCLC patients [30].

The present survey aims to update the findings of our 2010 review [27] which compiled all studies up to 2008 concerning the prognostic significance of serum tumor markers CEA, CYFRA 21-1, NSE, CA 125, CA 19-9, CA 15-3, SCCA, and ProGRP in both early and late-stage NSCLC. In this updated review, we incorporate all prognostic research conducted since 2008 until June 2022, presenting their results and grading the evidence based on criteria established by Hayes et al. [31]. We categorize the examined studies by stage due to the varying prognostic situations and therapeutic implications in early and advanced NSCLC stages. Similar to the previous review, the majority of studies focus on patients undergoing chemotherapy, and the most pertinent tumor markers are discussed individually, with comprehensive and detailed overviews provided in tables. Furthermore, we expanded the search to include the predictive value of STM in advanced stage NSCLC patients treated with targeted or ICI therapies. Finally, we critically address and discuss the limitations in study comparability due to heterogeneity and inconsistencies in the use of prediction and prognosis terminology [28, 32].

2 Methods

A search in the PubMed database was performed using the terms (and corresponding terms) “non-small cell lung cancer” (or “NSCLC”) AND “prognostic value” (or “prognosis” or “survival” or “prediction”) AND serum biomarkers: “CEA” (or “carcinoembryonic antigen”) or “CYFRA 21-1” (or “CYFRA21-1” or “cytokeratin-19 fragment”) or “NSE” (or “neuron-specific enolase” or “neuron specific enolase”) or “SCCA” (or “squamous cell carcinoma antigen” or “SCC-Ag”) or “CA19-9” (or “CA 19-9” or “carbohydrate antigen 19-9”) or “CA15-3” (or “CA 15-3” or “cancer antigen 15-3”) or “CA125” (or “CA 125” or “cancer antigen 125”) since the year 2008 (and three studies from 2007, not included in the last review) until June 2022. We supplemented the structured literature inquiry with a search of the reference lists from the included articles, to find additional eligible studies. Figure 1 displays a flow chart outlining the search process.

OPEN IN VIEWER

Fig. 1

Flow-diagram of the literature search in PubMed. NSCLC (non-small cell lung cancer).

Inclusion criteria were: article in English (or German) language, no double publication, NSCLC patients identifiable, no mixed histology investigations with SCLC, minimum number of participants N > 40, appropriate “prognostic” study design and statistical survival analysis evaluation, relevant serum biomarkers, no case reports. The following items were listed in the Tables 1–3: study type, number of patients, tumor stage, histology, therapy, endpoint investigated, STMs investigated, analytics and analyzer used, evaluation of results, the level of evidence and statistically significant prognostic STMs and additional investigated markers.

Grade of evidence was rated according to the criteria suggested and adapted by Hayes et al. [31]:

Evidence from single, high-powered, prospective, controlled study that is specifically designed to test marker, or evidence from meta-analysis, pooled analysis or overview of level II or III studies

Figure 2 presents the number of investigations, rather than the number of studies or patients, as in some studies multiple endpoints or baseline and additional kinetics of STMs were investigated. Consequently, in some studies, several investigations were conducted and considered separately.

OPEN IN VIEWER

Fig. 2

Results of tumor marker investigations in non-small cell lung cancer for all stages. The size of circles reflects the number of investigations, since baseline values, values post therapy or kinetics are investigated separately in some studies. Hence, the size of circles does not represent the number of studies but the number of investigations of the tumor marker. Positive predictive (low tumor marker levels reflect longer endpoint), negative predictive (high tumor marker levels reflect longer endpoint), NS (not significant), CEA (carcinoembryonic antigen), CYFRA21-1 (cytokeratin-19 fragment), NSE (neuron-specific enolase), CA125 (cancer antigen 125), SCCA (squamous cell carcinoma antigen), CA19-9 (carbohydrate antigen 19-9), ProGRP (pro-gastrin releasing peptide), CA15-3 (cancer antigen 15-3), TT (targeted therapy), ICI (immune checkpoint inhibitor), ChT (chemotherapy), OS (overall survival), PFS (progression-free survival).

Since 2008, numerous prognostic protein biomarker studies have been published. One thousand sixty nine articles were identified in the Pubmed database searched for publications between 2008 and June 2022. Eight hundred twenty two articles were excluded in the abstract screening as they did not fulfil the inclusion criteria. In full text screening of the remaining 247 articles, further 133 were found not to be eligible. Finally, a total of 114 studies were included in the review. For the evaluation of all stage NSCLC, 16 papers were identified, 36 papers for early-stage NSCLC and 62 for advanced stage NSCLC (Fig. 1). Among patients with advanced stages who were treated with either tyrosine kinase inhibitors (TKI) or immunotherapy (ICI), further studies were identified that claimed predictive value and conducted survival analysis. These studies investigated the same endpoints, primarily OS and PFS, making it difficult to differentiate them from studies on prognostic value. These studies are discussed in a separate section.

The majority of prognostic studies were single-center (102 out of 114), retrospective (86 out of 114) observations of single or multiple marker combinations at baseline (98 out of 114), before the initiation of therapy. Tumor marker kinetics during the course of treatment were considered more frequently (25 out of 114), especially in advanced stage NSCLC (20 out of 62). The primary endpoint for predicting prognosis was overall survival (OS; 95 out of 114) followed by the surrogate endpoints, progression-free survival (PFS; 45 out of 114) and disease-free survival (DFS; 9 out of 114) (Tables 1–3).

The most frequently reviewed tumor markers were CEA (98 out of 114), CYFRA 21-1 (72 out of 114), and NSE (33 out of 114), while other markers such as SCCA, CA 125, CA 19-9, CA 15-3, tissue polypeptide specific antigen (TPS) were investigated in single studies (Tables 1–3). Furthermore, routine blood parameters like C-reactive protein (CRP) [33–35], natrium [36], albumin [37, 38], ferritin [39], neutrophil-lymphocyte ratio (NLR) [40–42] and lactate dehydrogenase (LDH) [43] were identified as independent prognostic factors in studies investigating serum tumor markers in NSCLC (Tables 1–3). Over 90% of studies provided evidence levels 3 and 4, according to Hayes et al. [31].

In early stage NSCLC, most studies investigated tumor markers in patients undergoing surgery with or without additional chemotherapy (Table 1). Patients in studies investigating all stages were mainly treated with chemotherapy; however, treatment strategies were highly heterogeneous (Table 2). Reflecting the therapeutic advancements in late-stage NSCLC, chemotherapy regimens (18 out of 64) have been increasingly supplemented or substituted by tyrosine kinase inhibitor (TKI) (32 out of 64) or immune checkpoint inhibitor (ICI) (12 out of 64) therapies (Table 3).

4 Discussion

Many efforts have been made to assess the clinical significance of STMs for predicting monitoring therapy response, as well as for prognosis of NSCLC patients. Although many studies provide strong evidence of the high relevance of STMs for prognosis and prediction in both traditional chemotherapy and new targeted and immune therapies, none have been incorporated into guidelines or routine clinical practice. This may be due to the often retrospective nature of the studies –particularly in the chemotherapy era –and the lack of randomized controlled trials. As a result, many studies only attained evidence levels of 3 or 4 [31], while only a few high quality-pooled or meta-analyses reached higher levels. Noticeable efforts have been made since 2008 to adhere to the existing guidelines for reporting prognostic biomarkers, known as the REMARK recommendations [112], which were first introduced in 2005. In addition, improvements in study design and harmonization of study populations through subgroup investigations, particularly with respect to stage and histology, have been observed. However, the approval and introduction of new therapies that offer diverse treatment options and drug combinations have contributed to increased heterogeneity within patient cohorts. This, in turn, has made study reports heterogeneous, inconsistent, and sometimes conflicting, thereby complicating direct comparisons.

The most commonly investigated STMs were CYFRA 21-1, CEA, and NSE. Especially CYFRA 21-1 demonstrated high prognostic relevance across various therapeutic settings, stages and histologic subgroups. While elevated STM levels were often associated with poor prognosis, the relationship with CEA in TKI therapies was more controversial, as high CEA levels also predicted longer OS [113] and PFS in several studies [99–104]. A growing number of studies considered the inclusion of established clinical prognostic markers such as performance score, TNM stage, and histology, which were also the most important clinical parameters with prognostic relevance. However, data on factors potentially affecting STM levels, like concomitant diseases, were seldom provided. Although many studies adhered to REMARK guidelines, the reporting of pre-analytical specimen handling was often inadequate, while the documentation of analytical methods saw improvement. Cut-off levels for the STM were primarily determined based on manufacturer instructions, or own cut-off values were defined through receiver operating characteristic (ROC) curve analysis, leading to a range of inconsistencies. Concerning statistical survival analysis, most studies employed a multivariate Cox proportional model, but often relied on one-sided stepwise variable selection methods, including univariate prognostic variables without (nested) cross-validation. In some instances, the prognostic impact may have been overestimated due to selective variable assessment and subgroup evaluations that did not account for significant prognostic variables, small sample sizes, and too many events per variable in the multivariate analysis. There was a notable absence of control group investigations and of large, prospective, multicentric studies.

With the advent of new targeted and immune therapies and the definition of various first-, second- and even third-line therapy sequences, there is an increasing need for predictive and prognostic biomarkers that inform treatment decisions and long-term outcomes. Particular attention has to be drawn to the distinction between predictive and prognostic biomarkers, when evaluating outcomes in patients receiving specific treatments [18]. Predictive markers interact with treatment and directly affect patient outcomes by distinguishing responders from non-responders, while prognostic biomarkers are associated with differential disease outcomes regardless of the treatment applied [18, 114]. Unfortunately, inconsistent and interchangeable use of the terms “prediction” and “prognosis,” particularly, when progression-free survival is the study endpoint, has led to confusion. To establish a biomarker as predictive for a specific treatment’s benefit, a control group receiving a different treatment must be included to rule out the possibility that the biomarker is merely prognostic, indicating survival in both cohorts [18, 115]. Ideally, prognostic and predictive value should be validated simultaneously, as the presumed therapy benefit and, consequently, predictive value could merely reflect the prognostic significance of the marker [18, 115].

Prognostic biomarkers are typically defined by evaluating various survival endpoints such as overall survival (OS), disease free survival (DFS) and progression free survival (PFS) [20, 116]. However, each outcome measure’s limitations must be considered. While OS is objective and considered as the gold standard, it requires larger sample sizes and longer follow-up periods but can be accurately assessed due to its definite endpoint of death or disease-related death [118]. In times of multiple sequential therapy options, ‘surrogate endpoints’ like PFS and DFS are used to expedite drug approval or therapy changes in the event of treatment failure, often with shorter follow-up periods and smaller sample sizes [116]. Challenges with these surrogate endpoints include i) the necessity of frequent radiological controls, ii) at well-defined time intervals, iii) controlling evaluation bias due to interobserver variations, and iv) precise, clinically meaningful definitions of tumor response (complete response, partial response, stable disease etc.) or ‘progression event’ [118, 119]. The response evaluation criteria in solid tumors (RECIST 1.1) [117, 120], serve as the foundation for surrogate endpoint determinations, particularly for evaluating cytotoxic chemotherapy responses [121]. However, atypical response patterns (pseudoprogression and hyperprogression) observed in patients undergoing immune therapies [122], have made disease monitoring challenging using this measure, leading to the introduction of immune-based RECIST criteria [123]. Moreover, non-measurable lesions, asymmetrical tumor size changes, multiple metastatic lesions, differing dynamics of tumor size versus tumor activity, and the critical definition of “stable disease” present additional challenges for applying RECIST criteria in evaluating treatment outcomes [123, 124]. Some of these issues may be addressed by emerging developments like the metabolic imaging (e.g. 18-FDG PET –PERCIST and iPERCIST criteria) [121, 125] or radiological image pattern analyses (Radiomics), where medical imaging analysis and data mining methods are combined [126]; furthermore the combination with clinical aspects [127] and liquid biopsies [25] or more futuristic approaches, including deep learning mechanisms (artificial intelligence algorithms) [121] could result in personalized disease profiles and individualized therapy strategies [125, 128].

In recent years, molecular liquid profiling of cell-free tumor (ct)DNA in the blood plasma opened a whole new field of biomarkers and gained more and more interest. Several studies have explored the potential of liquid biopsies for prognosis, prediction and monitoring therapy response and detecting disease progression in lung cancer. These studies have investigated the additional use of circulating tumor cells (CTCs) [129] and cell-free tumor (ct)DNA [130] with STMs. Results have shown that changes in these biomarkers over time may correlate with longer progression-free survival (PFS) and overall survival (OS) in certain cancer types. In addition, newer approaches such as the use of cell-free RNA (cfRNA) in addition to STM testing have also shown promising results for early detection and monitoring of NSCLC [131]. These findings suggest that joint liquid profiling and STM investigations have the potential to be valuable tools in the clinical management of cancer patients.

In general, it has to be stated, that a noticeable heterogeneity in study designs, patient characteristics, analytical methods, pre-analytical methods, and statistical evaluations made it difficult to confidently assess the prognostic validity of STMs. Moreover, due to the non-comparability of these studies, it is currently not possible to provide concrete recommendations on how to use STMs for prognostic approaches, including clinically significant timepoints in early and late-stage therapies, absolute value thresholds, and kinetics in serial evaluations, preventing their timely incorporation into existing lung cancer protocols.

To address earlier and the above mentioned [27, 133] unresolved and fundamental issues in future studies, we suggest creating a core set of study criteria to conduct consistent, comprehensive, and comparable studies, that yield reliable clinical and biomarker data, thereby producing a more robust evidence base for specific tumor marker testing. A standardized core set could assist in the planning, the correct and sufficient evaluation of generated data and, especially, reporting of the results. Our proposal is to create such a core set through a Delphi panel, with an overview provided in Fig. 3.

OPEN IN VIEWER

Fig. 3

Checklist for prognostic and predictive serum tumor marker studies. STM (serum tumor markers), NPV (negative predictive value), PPV (positive predictive value).

The present survey updates and reaffirms the significant prognostic value of individual STMs and their combinations, particularly CYFRA 21-1 and CEA, in both early and advanced NSCLC patients undergoing chemotherapy, despite the considerable heterogeneity in study design and reporting. Furthermore, the clinical utility of STMs for prognosis and prediction in novel TKI and ICI therapies is demonstrated. To achieve higher evidence level of STM studies, it is recommended to include STMs in translational biomarker substudies of randomized phase III trials. These trials should include a large number of patients in both treatment and control groups, adhere to well-regulated (post)-treatment protocols, employ standardized outcome measures, establish well-defined blood collection schedules, and maintain standardized preanalytics, biobanking, analytics, and statistical evaluations. There remains substantial work to be done to fully harness the potential of protein-based blood biomarkers in traditional and emerging targeted and immune therapies.

Author contributions

Conception: IT, SH.

Interpretation or analysis of data: IT and SH.

Preparation of manuscript: IT and SH.

Revision for important intellectual content: IT and SH.

Supervision: SH.

Conflict of interest

IT has declared no conflict of interest.

SH has received research funding or honoraria from Roche Diagnostics, Bristol Myers Squibb, Merck KgaA, Sysmex Inostics and Volition SPRL. SH is also an editorial board member of Tumor Biology but had no involvement in the peer review process of this article.