Serum tumor markers for response prediction and monitoring of advanced lung cancer: A review focusing on immunotherapy and targeted therapies

1 Introduction

The past two decades have witnessed outstanding advances in thoracic oncology. The nature of these advances is three-fold: first, a deeper understanding of the molecular determinants of cancer initiation and progression; second, advances in the development and clinical use of novel anticancer therapies, such as targeted treatments and immunotherapies; and third, the application of increasingly sophisticated technologies and diagnostic tools, such as next-generation sequencing, tissue-based scores, artificial intelligence, and radiomics in lung cancer medicine [1–5].

Immunotherapy and targeted therapy are treatment options for patients with lung cancer, alongside surgery, radiotherapy, and chemotherapy. Immunotherapy uses monoclonal antibodies to interact with cytotoxic T cells or ligands on tumor cells to induce tumor cell apoptosis [3]. Immune checkpoint inhibitors (ICIs) with mechanisms of action against programed death receptor-1 (PD-1; e.g., nivolumab, pembrolizumab), programed death-ligand 1 (PD-L1; e.g., atezolizumab), and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4; e.g., ipilimumab) are available options [3, 4]. Treatment with ICIs alone, or in combination with chemotherapy, is recommended in the first-line setting for patients with advanced non-small cell lung cancer (NSCLC) whose tumors do not have actionable driver mutations, such as alterations in the epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase (ALK) genes [4]. Patients whose tumors do harbor actionable mutations may benefit from targeted therapy with tyrosine kinase inhibitors (TKIs), multi-kinase inhibitors, serine/threonine kinase inhibitors, or other small molecule inhibitors; many such agents are approved for the treatment of NSCLC [6].

Precision medicine and personalized medicine are rightly advocated as the most promising path forward in oncology, although their clinical application remains challenging due to a lack of infrastructure, in-depth fundamental knowledge, and validated data from clinical trials. There has been a focus on molecular markers to stratify patients for targeted or immune treatments and international guidelines recommend that all patients with advanced NSCLC undergo testing for mutations in EGFR, BRAF, Kirsten rat sarcoma virus (KRAS), ALK rearrangements, and PD-L1 expression [7]. However, in this quest for precision and personalization, more dated predictive circulating protein biomarkers have been somewhat neglected, as they have been shown to provide inadequate or insufficient information. Several issues are apparent in the existing literature: small, heterogeneous patient populations that may have concomitant diseases capable of influencing marker levels; use of different methods to assess serum tumor marker (STM) values; and lack of comparison with other existing predictive markers [8, 9].

The rapid evolution of therapeutic options brings the need to revisit and critically assess all tools at our disposal to support and guide clinical decision making. Recent studies suggest that STMs hold therapeutic promise, though published literature around this topic is scarce [10–13]. The European Society for Medical Oncology (ESMO) 2017 and 2019 Clinical Practice Guidelines (European and Pan-Asian adapted) do not recommend routine measurement of STMs for staging and risk assessment in patients with early or advanced NSCLC, and the ESMO 2021 guidelines state that there are currently no validated STMs with predictive value for small cell lung cancer (SCLC) [14–16]. These recommendations, however, are based on historical data, including evidence from a decade old review focusing on measurement of carcinoembryonic antigen (CEA) in patients following surgery, chemotherapy, or radiotherapy only [17]. The 2022 National Comprehensive Cancer Network and American Society of Clinical Oncology guidelines contain no information about STMs [7, 18–20]. Nonetheless, STMs have all the attributes to be considered “companion diagnostics”: STMs reflect tumor size or biochemical activity [21]; they can be assessed quantitatively, quickly, robustly, and with high levels of quality control on automated instruments; assessments are cost-efficient; and testing can be performed in a serial manner before, during, and after therapy to monitor local and systemic tumor control. However, to aid clinical decision making and to interpret individual STM levels and kinetics, several values and measures need to be developed, including defined criteria for appropriate time points, and relevant thresholds or individual value changes over time.

Predictive markers are defined as those that are measured at baseline and are indicative of therapeutic efficacy; they can discriminate between patients who will or will not respond to a specific therapy [22, 23]. Prognostic markers indicate the likelihood of disease recurrence, progression-free survival (PFS), or overall survival (OS), and are independent of therapy received as they reflect innate tumor behavior [23]. In this review, “prognosis” is considered after the initiation of immunotherapy or targeted treatment. In addition, some biomarkers have monitoring value, in that they can be measured serially to evaluate disease status or inform on the effect of a medical or biologic agent over time [23]. Importantly, some biomarkers are both predictive and prognostic, whilst also providing monitoring value (Fig. 1) [22, 23].

OPEN IN VIEWER

Fig. 1

Summary of predictive, prognostic, and monitoring STMs for lung cancer. Biomarkers may have multiple clinical applications during the course of the disease, depending on the timing of the test and the exact measurements and comparisons that are to be made. The arrow represents the time from initial histologic diagnosis and staging in arbitrary units. CT: computed tomography, NSCLC: non-small cell lung cancer, SCLC: small cell lung cancer, STMs: serum tumor markers.

We conducted a literature search to gather current evidence around the predictive, prognostic, and monitoring value of circulating STMs, including: CEA, cytokeratin 19 fragment antigen (CYFRA21-1), carbohydrate antigen 19-9 (CA19-9), cancer antigen 125 (CA125), cancer antigen 15-3 (CA15-3), squamous cell carcinoma antigen (SCCA), neuron-specific enolase (NSE), progastrin-releasing peptide (ProGRP), human epididymis protein 4 (HE4), and antibodies against tumor protein 53 (as described in Table 1), for patients with advanced stage lung cancer, including NSCLC and SCLC, receiving immunotherapy or targeted therapy.

Using PubMed and Cochrane CENTRAL databases, we conducted literature searches with a cut-off date of May 9, 2022. A team of medical professionals advised on the design of the search strategies, and these were adapted for each database. The Boolean operators “AND", “OR", and “NOT” were used to combine keywords related to the STMs of interest (Table 1), NSCLC, or SCLC, and a set of “Core terms” that defined the outcomes of interest and treatment, including immunotherapy and targeted therapy specific for NSCLC and SCLC. Searches were restricted to records of papers published after 2011 and non-human studies were excluded. Full search strategies for both PubMed and Cochrane CENTRAL are provided in Appendices 1–4.

Search results were imported into a shared spreadsheet and duplicate records removed. Two team members screened each abstract for eligibility. For records marked for potential inclusion, full texts were evaluated by the wider review team. Of 362 publications identified, 229 titles and abstracts were screened and 72 full texts were of potential interest and assessed for eligibility. In total, 36 publications were included in the review, of which only one concerned SCLC. Overall, 326 publications were excluded due to being duplicates (n = 132) or not fulfilling the inclusion criteria as follows: irrelevant study design (n = 134), case reports (n = 36), wrong patient population/setting (n = 15), not English language (n = 7), irrelevant outcome(s) (n = 2) (Fig. 2).

OPEN IN VIEWER

Fig. 2

Flow diagram of search results. NSCLC: non-small cell lung cancer, SCLC: small cell lung cancer, STMs: serum tumor markers.

A key objective of identifying predictive STMs in patients with lung cancer is to enable appropriate use of immunotherapy or targeted therapies and to personalize treatment by predicting response prior to treatment initiation, using baseline samples. Several STMs have demonstrated predictive value for treatment response in patients with advanced NSCLC receiving immunotherapy or targeted therapy (Table 2 and Fig. 3A and B).

OPEN IN VIEWER

Fig. 3

Summary of STM results across reviewed literature by A) predictive value of STMs in patients treated with immunotherapy; B) predictive value of STMs in patients treated with targeted therapy; C) prognostic/monitoring value of STMs in patients treated with immunotherapy; D) prognostic/monitoring value of STMs in patients treated with targeted therapy. a: Patients in the test set were treated with second-line nivolumab or atezolizumab; patients in the validation set were treated with first-line pembrolizumab. b: Correlation is based on a positive predictive value of non-response rather than on outcome. An elevation of CYFRA21-1, CEA, or NSE (≥50% compared to baseline) gave a positive predictive value of 90.3% (95% CI: 40.7–59.3) of non-response. c: Treatment modalities included: targeted treatment, immunotherapy, chemo-immunotherapy, chemotherapy, chemo-radiotherapy, radiotherapy, surgery + adjuvant chemotherapy. CA125: cancer antigen 125, CA15-3: cancer antigen 15-3, CA19-9: carbohydrate antigen 19-9, CEA: carcinoembryonic antigen, CYFRA21-1: cytokeratin 19 fragment antigen, EGFR-TKI: epidermal growth factor receptor-tyrosine kinase inhibitor, HE4: human epididymal protein 4, IT: immunotherapy, NSE: neuron-specific enolase, PD-L1: programmed death-ligand 1, ProGRP: progastrin-releasing peptide, SCCA: squamous cell carcinoma antigen, STMs: serum tumor markers, TT: targeted therapy.

Increased baseline levels of serum CYFRA21-1 and CEA may predict immunotherapy treatment response in patients with advanced NSCLC or SCLC. Shirasu et al. [47] described pre-treatment CYFRA21-1 levels ≥2.2 ng/mL as an independent predictor of prolonged PFS in patients with advanced NSCLC (n = 50) receiving second- or later-line nivolumab (median PFS 155 vs. 51.5 days for CYFRA21-1 <2.2 ng/mL) [47]. In contrast, Dall’Olio et al. [48] reported that patients with stage IIIB/IV NSCLC who received anti PD-1/PD-L1 treatment and had baseline CYFRA21-1 levels >8 ng/mL had shorter OS than patients with baseline CYFRA21-1 levels ≤8 ng/mL, whilst baseline CEA levels >8 ng/mL were not correlated with OS [48].

In a retrospective study, Kataoka et al. [49] found that high CEA expression (≥13.8 ng/mL) was independently associated with poorer PFS in patients who received nivolumab as a second- or later-line treatment for advanced NSCLC (n = 189) [49]. Similarly, in patients treated with anti-PD-1/anti-PD-L1/anti-CTLA-4 therapies (n = 61), Huang et al. [50] reported that elevated pre-treatment CEA levels ≥5 ng/mL were associated with significantly shorter PFS [50]. Thus, measurement of pre-treatment serum CYFRA21-1 or CEA levels may aid prediction of response to immunotherapy; however, the appropriate cut-off must be examined further.

Lang et al. [51] examined the predictive performance of a panel of STMs, including CEA, CA19-9, CYFRA21-1, and NSE in patients with advanced NSCLC treated with ICI monotherapy (n = 84) [51]. Using receiver operating characteristic curve analyses, baseline levels of CYFRA21-1 were found to have the best predictive power for disease progression or death, followed by CA19-9, CEA, and NSE [51]. Chai et al. [52] developed an OS probability nomogram to evaluate individual risk for patients prior to PD-1 inhibitor treatment (n = 110). The study reported that several of the parameters examined, including baseline CEA and CYFRA21-1 levels, were positively correlated with the individual’s risk score, such that higher baseline levels were associated with a higher risk score and hence a shortened OS [52]; however, this study has several limitations. Given the small sample size and, most strikingly, the absence of a comparison against data from patients receiving standard or no therapy, the predictive role of the analyzed biomarkers emerging from this study can only be speculated.

In patients with advanced SCLC receiving first-line PD-1/PD-L1 inhibitors plus chemotherapy (n = 102), Li et al. [53] noted that increased serum NSE levels at baseline and at 3 weeks post-treatment were correlated with poorer clinical outcomes [53]. Further investigation into the value of baseline serum NSE levels for predicting therapy response in a larger population of patients with advanced SCLC appears warranted.

3.2 Predictive STMs in patients treated with targeted therapy

Several studies have reported an association between high pre-treatment CYFRA21-1 levels and shorter PFS in patients with advanced NSCLC treated with TKI therapy. Tanaka et al. [54] found that high CYFRA21-1 levels (>2 ng/mL) were associated with significantly shorter PFS in patients with EGFR-mutated NSCLC (n = 160) receiving EGFR-TKIs [54]. Fiala et al. [55] also reported shorter PFS in patients with advanced NSCLC treated with erlotinib (n = 144) who had high vs. low pre-treatment CYFRA21-1 levels (≥2.5 vs. <2.5μg/L) [55]. Takeuchi et al. [11] noted that patients with advanced NSCLC treated with EGFR-TKIs (n = 95) and presenting elevated baseline serum CYFRA21-1 levels had shorter PFS than those with normal baseline serum CYFRA21-1, regardless of EGFR status [11]. Conversely, Zhao et al. [8] and Yanwei et al. [9] found no significant association between PFS and baseline serum CYFRA21-1 levels in patients with advanced NSCLC treated with EGFR-TKIs (n = 177 and n = 200, respectively) [8, 9]. Yanwei et al. [9] also reported no significant difference in PFS between patients with normal and high serum CA125 levels [9]. Discrepancies between the results of these two studies and most other studies regarding CYFRA21-1 may be due to differences in ethnicity of the patient population, as both studies were conducted in China, or might indicate that the cut-off value chosen (3.3 ng/mL) was not sufficiently discriminatory.

High pre-treatment levels of CEA have also been found to be an independent predictor of outcomes in patients with advanced NSCLC treated with EGFR-TKIs, although the results are again conflicting, possibly due to differences in trial design, patient characteristics, or the differing cut-off thresholds used. Fiala et al. [55] observed significantly shorter PFS in patients with a high vs. low pre-treatment CEA level (≥3 vs. <3μg/L; n = 144) [55]. Zhao et al. [8] reported that baseline CEA expression ≤10 ng/mL predicted favorable outcomes in patients with advanced NSCLC and EGFR mutations (n = 177) [8]. Chen et al. [56] investigated the predictive value of CEA in patients with advanced EGFR-positive NSCLC treated with first-line EGFR-TKIs (n = 241) and found that patients with elevated baseline CEA levels had reduced PFS [56]. Wu et al. [57] also reported a significant association between higher pre-treatment CEA levels (≥5μg/mL) and shorter PFS in patients with EGFR mutations treated with first-line EGFR-TKI therapy [57]. In a subset of patients with EGFR-positive NSCLC with brain metastases (n = 66) who had undergone whole-brain radiation therapy prior to EGFR-TKI treatment, Wei et al. [58] noted that pre-treatment CEA levels ≤10μg/mL were associated with significantly longer PFS and OS compared with CEA >10μg/mL [58]. Conversely, Romero-Ventosa et al. [59] reported that patients (n = 58) with CEA levels ≥5 ng/mL prior to treatment with the EGFR-TKI erlotinib had significantly longer OS compared with patients who had lower CEA levels, but changes in PFS were not significant [59]. The authors concluded that pre-treatment CEA levels may provide similar predictive information to EGFR mutation status. The study also reported no significant correlation between pre-treatment CYFRA 21-1 or SCCA levels and PFS or OS.

Facchinetti et al. [60] examined the correlation between survival and baseline CEA levels (≥5 vs. <5 ng/mL) in patients with EGFR mutations or wild-type/unknown EGFR status and found no significant relationship between OS and CEA levels in the total patient population (n = 79) [60]. Similarly, Yanwei et al. [9] found that in patients with advanced EGFR-mutated NSCLC who received EGFR-TKI treatment (n = 200), patients with higher baseline CEA levels (≥20 ng/mL) had significantly prolonged PFS as compared to those with lower CEA levels [9]; however, PFS did not differ significantly between the patients when using a cut-off of 5 ng/mL. This may be due to an inappropriate threshold being used. Ding et al. [61] reported that for patients with advanced EGFR-positive NSCLC (n = 22) treated with gefitinib or erlotinib, high baseline CEA levels (>3μg/L) did not predict higher metastatic disease burden or reduced survival times [61]. In addition, no correlation between an early decrease (within 4 weeks of beginning of treatment) in CEA levels and treatment response or survival was observed. Han et al. [62] examined whether CEA levels could predict acquired EGFR-TKI resistance in patients enrolled for palliative care with EGFR-TKI therapy (n = 100). Whilst the study found that pre-treatment CEA levels did not predict acquired EGFR-TKI resistance, patients with high pre-treatment CEA levels (>10 ng/mL) benefited more from treatment in terms of PFS and OS than patients with low (5–10 ng/mL) or normal (<5 ng/mL) CEA levels [62].

In a retrospective cohort study, Pan et al. [63] analyzed the correlation between pre-treatment STM levels and response to EGFR-TKIs in patients with stage IIIB/IV lung adenocarcinoma (n = 48). Patients with elevated pre-treatment levels of serum CEA (≥5 ng/mL) and CA19-9 (≥35 U/mL) had a greater disease control rate (DCR; defined as a partial or complete response or stable disease ≥6 weeks) and prolonged survival time, but pre-treatment serum CA15-3 and CA125 levels were not related to outcomes [63]. Furthermore, in a randomized controlled trial, Ramalingam et al. [64] evaluated a panel comprising CEA and CYFRA21-1 for its predictive response to treatment with linifanib (a multi-targeted receptor TKI), in combination with carboplatin and paclitaxel as first-line therapy for patients with advanced, non-squamous NSCLC (n = 138) [64]. A high CEA (>3 ng/mL) and low CYFRA21-1 (<7 ng/mL) signature at baseline was associated with significantly improved PFS in both treatment arms (linifanib 7.5 and 12.5 mg) [64]. Conversely, Takeuchi et al. [11] reported no association between elevated pre-treatment CEA levels (>5 ng/mL) and either PFS or OS in patients with advanced NSCLC who were treated with first- and second-generation TKIs (n = 95) [11].

An association has also been reported between baseline plasma NSE levels and response to treatment in patients receiving targeted therapy. Fiala et al. [65] reported significantly shorter PFS in patients with high (≥12.5μg/L) vs. low (<12.5μg/L) pre-treatment NSE levels receiving EGFR-TKIs (erlotinib or gefitinib; n = 163) [65]. Suh et al. [66] also found that increased baseline NSE levels were associated with significantly shorter PFS following treatment with erlotinib or gefitinib (n = 151) [66]. However, Zhao et al. [8] found no significant difference in PFS between patients with low (≤13.7 ng/mL) or elevated (>13.7 ng/mL) serum NSE levels (n = 177) [8]. The predictive value of CA19-9 was also evaluated in this study (cut-off: 35 U/mL), but no significant difference in PFS was observed between low and high expression of this marker [8]. In gefitinib-treated patients with EGFR-mutated NSCLC and available baseline data (n = 22), Inomata et al. [67] found that higher pre-treatment NSE levels were significantly associated with shorter PFS and OS; however, there was no association between pre-treatment serum ProGRP level (n = 31) and PFS or OS in this setting [67].

4 Monitoring STMs (measured serially)

5 Discussion

This scoping review gathered current evidence around the predictive and monitoring value of existing circulating STMs for patients with advanced NSCLC or SCLC receiving immunotherapy or targeted therapy. The search retrieved 362 publications, of which 36 relevant articles were included in this review. Of the 36 included, 29 described retrospective studies, two were observational studies, three were non-randomized experimental studies, one was a randomized controlled study, and one was a diagnostic test accuracy study. Only one SCLC study was identified by the search and included in this review (Li et al. [53]), highlighting a clear knowledge gap for this indication. The majority of studies assessed individual biomarkers, and the STMs most often examined were CEA, CYFRA21-1, and NSE. Measuring a panel of STMs also appears to be a promising approach, based on both the published articles found in our literature search and data presented at recent international congresses [77–81]. Overall, we observed that i) STMs may add value for predicting treatment response in patients with advanced lung cancer receiving immunotherapy (Fig. 3A), and ii) STMs may be useful as a monitoring tool for patients with advanced lung cancer receiving either targeted therapy or immunotherapy (Fig. 3C and D). The predictive value of STMs for patients treated with targeted therapy (Fig. 3B) is less clear, with several studies reporting conflicting results, possibly due to differences in the timing of baseline measurements, patient ethnicity, and the STM level cut-off thresholds used [8, 11]. Another explanation might be that for targeted treatments, the mechanism of treatment resistance and the onset thereof has more of an influence on PFS and OS than STM levels prior to treatment.

It is important to note that many of the studies evaluated did not discriminate between predictive STMs (those measured at baseline and being indicative of therapeutic effect) and prognostic STMs (those indicative of patient outcome, independent of therapeutic agent received). Instead, these studies only showed a relationship between baseline STM measurements and patient outcome. Moreover, separating prognostic significance and the prediction of treatment effects remains difficult, particularly in studies lacking a control arm, or when there is no direct mechanistic connection between a given STM and a therapeutic agent. In light of this, we must consider whether high STM levels at baseline could be indicative of a more advanced or aggressive disease at treatment initiation (compared with lower STM levels), rather than specifically being predictive of an unfavorable response to treatment.

Whilst the predictive benefit of STMs is unclear, particularly for patients receiving targeted treatments, one could compare them to other indicators of therapeutic response such as PD-L1 staining [82], tumor mutational burden [83], and EGFR-mutation status [84]. These indicators are not overly precise or specific, nor do they have widely accepted cut-off thresholds, yet they are still useful for guiding therapeutic decision making. Provided STMs give clear predictive information on the effectiveness or non-effectiveness of immunotherapies and targeted therapies for a proportion of patients (e.g., for 30% of patients with 95% specificity), they could have real clinical value, particularly as more treatment options become available and predictive STMs become better characterized and validated.

When serial STM measurements were used to monitor response to treatment for both immunotherapy and targeted therapy, all studies evaluated presented a negative correlation between STM concentration change and clinical outcome (Fig. 3C and D), meaning that increases in STM levels after the start of treatment were associated with worse clinical outcomes and vice versa. Since all studies showed this relationship, this demonstrates the potential utility of STMs as a tool to monitor targeted treatment and immunotherapy in patients with advanced lung cancer. As cancer treatment becomes ever more personalized, based on the precise genetic alterations detected in each tumor, continued research into the most appropriate STM for use during therapeutic monitoring will also be necessary. Increasing coordination between drug and diagnostics development arms is likely to be paramount in the future, to develop both a mutational test and a specific monitoring assay to correlate with each precisely targeted treatment agent.

Based on this review, there is currently no consensus on i) the optimal STM sampling time points for baseline and serial measurements, ii) the cut-off thresholds and change in STMs that can be used to predict or monitor response/non-response, iii) the relevant clinical events to predict or monitor therapeutic efficacy, and iv) the required sensitivity, specificity, or positive/negative predictive values to enable clinical application. An additional relevant consideration concerns the unmet clinical need and for what clinical purpose the STM can be of most value. For monitoring, this could be accurate detection of response or non-response to treatment to support either continuation or discontinuation of treatment. For instance, monitoring STMs may provide clinical utility in patients treated with immunotherapy where radiologic follow-up may be difficult to interpret due to phenomena like pseudo-progression or the absence of target lesions.

Measurement of STMs has several advantages, including low costs, general availability, low risk to the patient, and short turn-around times. STM monitoring may qualify as informative companion diagnostics during treatment and over the course of the disease. A major challenge and complication, however, is that STM tests are not well harmonized, with various methodologies and systems used across the studies evaluated herein, and it is important to note that results obtained for one measurement system might differ from another [85]. For the time being, therapeutic decisions should not and cannot be based on STM measurements alone but must always be considered in the context of other factors, including a patient’s tolerance to treatment and imaging results. Stepwise diagnostic pathways, incorporating STMs with other clinical and radiologic measures should be the focus of future development, with the ultimate aim of i) improving patient guidance; ii) limiting the harm caused by treatment side effects; and iii) reducing economic burden of lung cancer treatment.

Our review emphasizes that evidence for the clinical application of STMs is not sufficient for high-grade clinical guideline recommendations and further research is needed; however, given the current therapeutic landscape, using STMs for monitoring purposes seems most promising, with the majority of evidence concerning CEA and CYFRA21-1. There is a clinical need for better monitoring of response to immunotherapies on a biochemical level. STMs that correlate with tumor mass or tumor cell activity may be accurate and cost-efficient tools for at least a portion of patients and therefore could be of added value, particularly when combined with other molecular or radiologic exams. Inclusion of such STMs, particularly in prospective immunotherapy studies, is therefore recommended. In clinical practice, there is currently little need for predictive STMs as treatment options for advanced lung cancer are limited and therapeutic decisions are based on histopathologic and molecular tumor characteristics, radiologic staging, and patient performance status. Given their inherent limitations, STMs alone are unlikely to change current predictive practices; however, there may be a place for them in combination with other biomarkers in future predictive models.

In conclusion, we found that baseline and dynamic changes in STM levels may be of added value in the context of prediction of treatment outcome and monitoring of patients with advanced lung cancer during and after treatment with targeted or immunotherapies. However, further research into the utility of STMs in routine clinical care of patients with advanced lung cancer is warranted.