Early serum tumor marker levels after fourteen days of tyrosine kinase inhibitor targeted therapy predicts outcomes in patients with advanced lung adenocarcinoma

Objective Image evaluation strategy for lung cancer patients has difficulty obtaining the appropriate quantity of diffuse lung nodules and bone metastases. The study was to demonstrate whether early variations in the levels of serum 4-tumor markers (4-TMs)(carcinoembryonic antigen [CEA], cancer antigen [CA]125, CA19-9, and CA15-3) after TKI targeted therapy were associated with treatment response in patients with lung adenocarcinoma. Methods Patients with stage IIIB-IV lung adenocarcinoma taking epidermal growth factor receptor (EGFR) TKIs or anaplastic lymphoma kinase (ALK) inhibitors were enrolled prospectively from June 2012 to February 2015. According to the variations of the percentage of change in 4-TM levels (4-TMpc), we divided patients into ascending (increases in 4-TMpc over the 7th- 14th day) and descending (decreases in 4-TMpc over the 7th- 14th day) groups. Results 184 patients were enrolled, and 89% had at least one of the pre-treatment evaluable TMs and were further analyzed. An excellent response to the TKI targeted therapy was accurately predicted in the descending group, as determined using receiver operating characteristic curve analysis (an area under the curve, 0.83). Multivariate Cox hazards model analyses demonstrated that the type of 4-TMpc and mutation status were the strongest predictors of progression-free survival (PFS)(descending versus ascending, hazard ratios [HR] 0.30, 95% confidence interval [CI], 0.19–0.47; sensitive mutation versus wide type, HR 0.30, 95% CI, 0.19–0.48). Conclusions Type of 4-TMpc 14 days after TKI targeted therapy is associated with an image response and PFS, without regarding mutation status, in patients with advanced lung adenocarcinoma.

However, although sensitive EGFR mutations and ALK rearrangement are strong predictors of good response to TKIs targeted therapy, not all patients (about 60-70%) respond to the therapy [1][2][3][4][5], although a portion of patients with EGFR wild type mutations and ALK-negative have shown a response [6,7].
Morphologic imaging studies using the Response Evaluation Criteria in Solid Tumors (RECIST) remain the standard tool for evaluating treatment response [8]. However, this image evaluation strategy has several limitations, such as difficulty in obtaining the appropriate quantity of diffuse lung nodules, pleural effusions, and bone metastases [9].
Serum tumor marker (TM) concentration is a reflection of the synthesis potential of the tumor [10]. Assessment of TMs for evaluating treatment responses is clinically objective. Elevated carcinoembryonic antigen (CEA) levels have been observed in 40-80% of patients with non-small cell lung cancer (NSCLC) [11,12]. Nevertheless, a single assessment of CEA levels to evaluate lung cancer treatment response is not sensitive [9].
Thus far, not much is known about TM levels' changes and their genuine relationship to predict prognosis in adenocarcinoma patients receiving TKI targeted therapy [9,13,14]. We assessed 4-TMs, CEA, carbohydrate antigen (CA) 125 (CA125), CA19-9, and CA15-3. The selection of these tumor markers was based on reports [9][10][11][12][13][14] and our previous pilot study results (data not shown). There 4-TMs were also easy to assess in clinical practice and provided the most cost-effective coverage in patients with advanced lung adenocarcinoma. The study aimed to demonstrate whether early variations in serum 4-TMs after TKI targeted therapy were associated with treatment response and PFS in advanced lung adenocarcinoma patients.

Patient selection and treatment
Patients with stage IIIB-IV lung adenosquamous cancer or adenocarcinoma taking EGFR TKIs (gefitinib, erlotinib, or afatinib) or ALK inhibitors (crizotinib or ceritinib) in different lines of therapy were enrolled in a prospective, single-center at the China Medical University Hospital from June 2012 to February 2015. We calculated the sample size needed for the kappa analysis by PASS software (version 20.0.1, NCSS, LLC. Kaysville, Utah, USA) via assuming that the proportion of good response to being around 50% in patients with advanced lung adenocarcinoma treated with targeted therapy. A sample of 161 patients achieves 90% power to

EGFR mutation and ALK immunohistochemistry analysis
The tumor DNA sequences of exons 18 to 21 of EGFR were determined using direct forward and reverse sequencing via the polymerase chain reaction (PCR) product from nested PCR reactions [15]. Sensitizing mutations are defined as G719X in exon 18, in-frame deletions or insertion of exon 19, A763_Y764 insFQEA mutation, and S768I in exon 20 and L858R or L861Q in exon 21 [16][17][18]. ALK immunohistochemistry (IHC) was performed using the Ventana anti-ALK (D5F3) CDx assay. The staining results were evaluated using a binary scoring system: positive or negative following the manufacturer's instructions [19].
Serum CEA, CA125, CA19-9, and CA15-3 level detection and analysis TM levels obtained from peripheral blood samples were measured before TKI targeted therapy and after 7 and 14 days of treatment. To reduce TM levels' influence with inherent intra-individual biological variation and within-laboratory coefficients of variation (TM v ) [20][21][22], we defined a cutoff level for each individual using pre-treatment TM levels of 2-fold over the standard upper limit. Therefore, enrollment criteria included CEA, CA125, CA19-9, and CA15-3 levels at 10.0 ng/mL, 70 units/mL, 70 units/mL, and 76 units/mL, respectively. Patients who did not show an elevation in TMs above this level were regarded as biochemically non-assessable and were excluded from further follow-up.
To evaluate changes in TM levels after TKIs targeted therapy and to account for patients having more than one evaluable TM, we created a formula "percentage of change of 4 TMs (4-TM pc )". Assuming a distinct sub-clone released each TM within the tumor bulk (1-marker (later)/marker(previous)), a reasonable estimate of the proportion of tumor treatment for this sub-clone was made. Our Eq 1 represents the weighted average of the proportion of tumor treatment across different sub-clones.
According to variations in 4-TM pc on days 0, 7, and 14, we divided patients into four groups. Type 1. Ascending: patients who sustained an increase in 4-TM pc . Type 2. Descending-ascending: patients who showed a decreasing trend on the 7 th day, and then showed an increasing trend on the 14 th day. Type 3. Ascending-descending: patients who showed an increasing trend on the 7 th day and then showed a decreasing trend at subsequent time points. Type 4. Descending: patients who showed a persistently decreased 4-TM pc . For minimizing TM v interference, the following definitions were created [20][21][22]: when 4-TM pc was <5% over the 7 th -14 th day, and it was defined as "type uncertain." Confirmed decreases in 4-TM pc over the 7 th -14 th day (types 3/4) were consistent with tumor response. Similarly, increases in 4-TM pc over the 7 th -14 th day (types 1/2) were regarded as tumor progression.

Imaging-based response
Tumor response was assessed on chest radiographs (CXR) and computed tomography (CT) scans, using the RECIST version 1.1 in an independent radiologic review by assessors who did not know the results of 4-TM pc studies and confirmed at least two scans obtained 28 days apart. Long-term follow-up was performed until July 31, 2015.
Therapeutic efficacy was classified as partial response (PR), stable disease (SD) with tumor reduction <30% (SD -30 ), SD with the increase in tumor size <20% (SD +20 ), or progressive disease (PD). Patients who died due to cancer between these CT/CXR procedures were classified as having PD. While analyzing the correlation between 4-TM pc after 14 days TKI targeted therapy and RECIST assessed response, and we combined categories of PR and SD -30 into the "good response group." In contrast, the "poor response group" included cases with PD and SD +20 . This classification's rationales were that: (1) It was the straight forward approach to classify responders vs. non-responders. (2) It was also difficult for PR, SD, or PD to correlate with the ascending or descending of the 4-TMpc from a statistical point of view. The endpoint was PFS. PFS was assessed from the date of the beginning of TKI targeted therapy to the date of PD or death due to cancer. If a patient was lost to follow-up or had no event, time to progression was censored as the date of the last contact date.

Statistical analysis
To obtain a descriptive analysis, we resumed each continuous variable as median and 25-75, and categorical variables as proportion. We performed receiver operating characteristic (ROC) curve analysis for 4-TMs to predict TKI targeted therapy's response. The agreement between the 4-TMs and the image-based morphologic response was evaluated using the kappa statistic. PFS was analyzed according to the Kaplan-Meier method and was compared with the log-rank test. Cox proportional hazards model was used to evaluate independent predictive factors associated with PFS. Data were analyzed using SPSS-17 (IBM SPSS Statistics. Inc. Chicago, IL, USA). For all analyses, two-sided P<0.05 was taken as statistically significant.

Results
In all, 195 patients with a diagnosis of stage IIIb-IV adenocarcinoma (one with adenosquamous carcinoma) were screened for 4-TM levels before the start of TKI targeted therapy, and 11 were excluded: 1 because the patient had severe interstitial lung disease after taking erlotinib and ten because the standard protocol was not followed. Seven patients were recruited more than once because patients accepted re-challenge TKI targeted therapy. Therefore, 184 patients and 191 patient-times were enrolled in this study, including 29 accepted the diagnostic procedure of computed tomography-guided core needle biopsy, 68 accepted transbronchial biopsy, 20 accepted ultrasound-guided biopsy, 27 accepted thoracentesis or pericardiocentesis, and 40 accepted operative procedures. Detailed baseline characteristics are summarized in Table 1.
Activating mutations were documented in 73% cases, and 3% had an unknown mutation status because no sufficient pathological material was available. Further, 98% of patient-times received EGFR TKIs, and 2% received ALK-inhibitors Of the 184 enrolled patients, baseline serum CEA levels, CA125, CA19-9, and CA15-3 were 2-fold over the standard upper limit in 71%, 62%, 27%, and 23% patients, respectively. In all, 89% of patients with at least one of the pre-treatment evaluable TM were further analyzed. Of

PLOS ONE
Tumor markers Predict TKIs Efficacy in Lung cancer these, 7% had all 4 TM levels elevated, 26% had 3, 33% had 2, and 34% had 1 TM elevated. The summarized data is shown in Table 2.
However, 24% of patients had baseline CEA< 10 ng/mL. This group of patients could not be biochemically assessed using CEA levels alone. The AUC was only 0.51 (95% CI, 038-0.63) for predicting TKI targeted therapy response with the type of CEA. The Kappa value in 117 cases was 0.449 (Fig 2).
One hundred and forty patients in whom first-time TKI targeted therapy with the type of 4-TM pc was used, were further analyzed using Kaplan-Meier curves for PFS. As shown in Fig  3A, the median PFS had no significant difference between types 1 and 2 (30 and 28 days), nearly the same as types 3 and 4 (252 and 245 days). However, PFS was significantly longer in Table 2 types 3/4 than in types 1/2 (P<0.001). Analysis for subgroups stratified according to mutation status found that PFS was still longer in patients with types 3/4 than in patients with types 1/2 (activated mutation group, P = 0.016 ( Fig 3B); EGFR and ALK-negative/unknown group, P<0.001 (Fig 3C)).

Discussion
This prospective study was the first to provide CEA, CA125, CA19-9, and CA15-3 permutation and combination in patients with advanced lung adenocarcinoma. We compared the AUC of CEA and 4-TMs to predict clinical responses of TKI targeted therapy. Our study demonstrated that 14 days of 4-TM pc type could be an early predictor of TKI targeted therapy efficacy. The descending type of 4-TM pc over the 7 th -14 th days had longer PFS in mutated or non-mutated adenocarcinoma patients. Although squamous cell lung cancer and adenocarcinoma are a subset of NSCLC, they have different driver mutations and treatment [23]. CEA is more frequently reported in patients with adenocarcinoma than squamous lung cancer [11,12]. Therefore, we focused on lung adenocarcinoma. Furthermore, not all lung adenocarcinoma patients had elevated CEA levels [9,13,14]. Advanced lung adenocarcinoma has other potentially valuable TMs in addition to CEA. In our series, CA125, CA 19-9, and CA15-3 levels reached evaluable criteria in 62%, 27%, and 23% patients, respectively. While combined with 4-TMs, only 11% of patients had 4-TMs below the evaluable levels ( Table 2).
There is insufficient evidence to support a conclusion concerning the standardized combination of TMs to evaluate tumor status. Different intra-tumor sub-clones may release different TMs. We presumed "one TM, one evaluable clone" and combined TMs as "4-TM pc ".
Zhang et al. indicated that the descending type of CEA within one month correlated with PR and SD of EGFR-TKI in patients with lung adenocarcinoma [13]. However, the CEA type can be affected by TM v and can influence the clinician to make a wrong decision [20,22]. In our 170 patient-times with evaluable TM cohort, "type uncertain" was classified when 4-TM pc   [13,24]. Otherwise, type 2 with fluctuation in 4-TM pc and then an ascending pattern the over 7 th -14 th day is not logical if TKI targeted therapy is considered effective (Fig 1). We evaluated the type of 4-TM pc within 14 days of TKI targeted therapy and the relationship with imaging-based response and PFS. The ROC curve analysis showed that using 4-TMs for predicting the efficacy of TKI targeted therapy response had a higher AUC (0.83) than that using CEA (0.51). The Kappa value for the agreement analysis between 157 cases with the type of 4-TM pc and radiographic results was "good" (0.762) (Fig 1). However, using CEA levels, 24% of patients were biochemically non-assessable in this 170 patient-time series. The kappa value was 0.449 only (Fig 2). These findings showed that using 4-TMs to predict TKI targeted therapy response was more accurate than using CEA alone.
The results of sensitizing mutations cannot guarantee a clinical response to TKI targeted therapy [6,7,25]. It is essential to develop a new strategy for early prediction of the effect of TKI targeted therapy. In our series, patients with types 3/4, 4-TM pc had a longer PFS than those with types 1/2. Regarding the ascending (type 3) or descending pattern (type 4) over the 0 th -7 th day, patients' outcomes were similar. On the contrary, similar PFS was observed in patients with types 1 and 2 ( Fig 3A). Regarding activated mutation ( Fig 3B) and EGFR and ALK-negative/unknown group (Fig 3C), PFS was also significantly longer in patients with types 3/4 than in patients with types 1/2. Similarly, in multivariate models, our results demonstrate that 4-TM pc and mutation status continues to be the strongest predictors of PFS (Table 3).
Our study's strengths include its prospective design, and a large number of patients included compared to previous TM studies in NSCLC patients under TKI targeted therapy [9,13,14]. However, certain drawbacks should be considered. Firstly, it is a single-center study. Secondly, there is insufficient evidence to define the best cutoff level for minimizing the interference of TM v. Thirdly, we did not extend our study after Feb 2015 because the enrolled cases and the follow-up time are sufficient to reflect the study results in which we did find high-level agreement [kappa = 0.762] by planned accrual [around 190 cases]. Furthermore, the investigated scenario (gefitinib, erlotinib, afatinib, crizotinib or ceritinib for lung adenocarcinoma patients) was still valid in Taiwan in 2020, although whether our finding was applicable for some new inhibitors (such as osimertinib [26] or alectinib [27]) deserved to be studied.

Conclusion
In conclusion, image evaluation strategy with RECIST for patients with lung cancer has difficulty in obtaining the appropriate quantity of diffuse lung nodules, pleural effusions, and bone metastases. The type of 4-TM pc after 14 days TKI targeted therapy is associated with image response and PFS without accounting for mutation status in advanced lung adenocarcinoma patients. Our study results could help early therapeutic decision making by identifying patients who may benefit from gefitinib, erlotinib, afatinib, crizotinib, or ceritinib 14 days after TKI targeted therapy.