Running title: CEA and TPA in the pre-operative evaluation of NSCLC

From the Cuneo Lung Cancer Study Group at the "S. Croce e Carle" Hospital, Cuneo I-12100, Italy

1. Gianfranco Buccheri, M.D.
Divisione di Pneumologia , Ospedale "S. Croce e Carle" , Cuneo, I-12100, Italy, Tel. 0039.0171.441733, Fax. 0039.0171.441764,  e-mail: buccheri@culcasg.org
2. Domenico Ferrigno, M.D.

Corresponding author: Gianfranco Buccheri, M.D.

Key words:

Acknowledgments:

The authors thank Lorena Gribaudo and Anna Merlo, nurses of their outpatients' unit, for the invaluable help and support.

1. ABSTRACT

Study Objectives: Copious literature shows that in lung cancer many serum markers, especially the cytokeratin degradation products, correlate with the extent of disease. In 1995, we suggested that it should be possible to predict the resectability of non-small cell lung cancer (NSCLC) by measuring the plasmatic level of the tissue polypeptide antigen (TPA), a marker of the cytokeratin family. This study was designed: 1.) To confirm the earlier data in a new prospective evaluation; 2.) To comparatively assess another classic biomarker (i.e., the carcinoembryonic antigen, CEA); and 3.) To incorporate their results into the pre-surgical evaluation of NSCLC.

Design: Analysis of a single-institution database over a 5-year period (1994-1998).

Setting: Community-based Hospital and Second Referral Level Institution for a province of 500,000 people.

Patients: 124 consecutive patients (105 males), with pathologically documented lung cancer (50% with adenocarcinoma) accurately staged, clinically judged operable or potentially operable, and eventually operated upon.

Interventions: Anthropometric, clinical, laboratory data - including the CEA and TPA serum levels -, and the results of a complex staging work-up were prospectively recorded. Receiver-operating characteristic (ROC) curves and diagnostic formulas were used for data analysis.

Measurement and Results: The computed tomography (CT) of thorax, upper abdomen and brain was the most accurate pre-operative method to assess tumor resectability (ROC area: 0.76, 95% CI: 0.67-0.86, p=0.000; accuracy rate 77%, CI: 69-84%). TPA was also predictive for tumor resectability (ROC area: 0.62, 95% CI: 0.51-0.73, p=0.035, accuracy rate at a threshold level of 110 U/L: 65%, 95% CI: 56-73%). CEA was diagnostic only at the extreme values of its distribution (accuracy rate at a level up to 10 ng/ml: 69%, 95% CI: 60-77%). The probability of finding a resectable disease at the time of the operation increased from 78% (baseline CT-based probability) to 83%, when TPA was lower than 90 U/L, and to 85%, when CEA was below 10 ng/mL. The probability of discovering an advanced disease increased from 68% (baseline CT-based probability) to 89%, when also TPA resulted abnormal, and to 100%, when CEA was higher than 10 ng/mL. Conversely, the predictability of CT was diminished by contrasting biomarkers’ results, requiring further clinical investigations.

Conclusions: CT remains the gold standard for the pre-operative evaluation of NSCLC, although it may significantly underestimate the real tumor extension. The addition of the easy and inexpensive TPA test (with or without CEA) is capable to correct this underestimation, and helps to decide whether to completely rely on CT or order additional clinical investigations.

 

2. ABBREVIATIONS

CT = computed tomography; NSCLC = non-small-cell lung cancer; CEA = carcinoembryonic antigen; TPA = tissue polypeptide antigen; ROC = receiver-operating characteristic; CI = 95% confidence interval

 

3. INTRODUCTION

Serum tumor markers are not only of significance to the researcher in developing theories concerning tumor biology, but also to the clinician in treating patients with cancer ³⁴. In oncology practice, serum tumor markers may be helpful in the diagnosis, pathologic classifications, and in the evaluation of stage of disease and prognosis. When measured serially after a diagnosis is established, they may aid in assessing the response to treatment, monitoring the spontaneous course of the illness, and surveilling for tumor recurrences ¹³.

Lung cancer does not make exception to this rule and the expression of serum biomarkers, in this particular tumor, is various and abundant ⁴. Lung tumor markers fall into several categories including oncofetal proteins, structural proteins, enzymes, cell membrane components, secreted peptides, hormones, and other tumor-associated antigens ⁴. In non-small cell lung cancer (NSCLC), cytokeratin-derived molecules and the carcinoembryonic antigen (CEA) are probably the most helpful markers and certainly the most used ones ^{4; 15; 16}.

In 1995, we suggested that it should be possible to predict the resectability of NSCLC by simply measuring the plasmatic level of tissue polypeptide antigen (TPA), the oldest marker of the cytokeratin family. We were aware that such evidence was very preliminary and needed confirmation. However, the possible equivalence of a blind serum test to the more complex, time-consuming and expensive multiorgan-computed tomography (CT) was a truly exciting perspective. Given the importance of the issue, we decided to continue the pre-operative biomarker testing, for a new confirmatory study.

In this report, we describe the Cuneo Lung Cancer Study Group’s (CuLCaSG) most recent experience with CEA and TPA obtained in subjects with potentially resectable NSCLC. As already mentioned, this study was done mainly to confirm our earlier data. Secondary aims were: 1.) The comparative assessment of TPA and the "classic" carcinoembryonic antigen (CEA), and 2.) The incorporation of marker test results into the conventional pre-surgical evaluation of NSCLC.

3. PATIENTS AND METHODS

Patients' database and study design

In the fall of 1982, a group of chest physicians decided to devote their professional activity to the study of lung cancer. The group, who later became known as CuLCaSG, is still active at the department of Pulmonary Medicine of the "A. Carle" hospital, in the city of Cuneo, Piedmont, Italy. The former "A. Carle" hospital for chest diseases merged with the "S. Croce" general hospital and the two hospitals, now named "S. Croce e Carle," were then designated as Hospitals of National Importance. They serve the whole Cuneo Province as Second Referral Institution. Among the prime acts of the CuLCaSG, was the creation of a clinical database for patients with carcinoma of the lung, effective in January 1983. All lung cancer patients, referred to a physician of the Group, were managed uniformly. Data regarding 44 variables was collected for each new patient with a cytologically or pathologically documented diagnosis of lung cancer ³⁷ and recorded on computer. Such database included anthropometric and clinical characteristics, routine lab tests and serum tumor markers, TNM descriptors and a computer-derived stage of disease. Since 1983, TNM definitions have changed radically in 1986 and minimally in 1997, because of two consecutive revisions to the International Staging System for Lung Cancer (ISSLC) ^{29; 30}. Therefore, to allow for homogeneous comparisons, we had to review the patients’ charts and upgrade their TNM variables. This work was done as soon as the revision of the ISSLC was formalized. Every 4 to 5 years, the structure of the database was modified and upgraded to new software, while the number of variables progressively increased to hundreds. However, the core variables of the early database remained unchanged, allowing for careful analyses and time-related comparisons. In particular, out of 1296 new lung cancer patients seen consecutively during the years 1983-1998, 1136 underwent a pre-treatment CEA test, while 1115 had a pre-treatment TPA. A large part of this population has been the object of a number of prior publications concerning mainly TPA ^{6; 7; 9-11}. Updated descriptive statistics of the entire population was obtained during the preparation of this article. It confirmed prior data, showing that both CEA and TPA levels increased significantly (CEA Rs=0.161, p=0.000; TPA Rs=0.322; p=0.000), paralleling the stage of disease ³⁰. In the entire sample of the 1115 patients tested, median values of TPA were: 70, 120, 70, 115, 114, 130, and 180 U/L for, respectively, the stages Ia, Ib, IIa, IIb, IIIa, IIIb, and IV. Corresponding median values for CEA were 2, for the stages Ia-IIa , and 3 for more advanced stages.

All patients, seen in 1994 and afterwards, were eligible for this study if they had a pathologic diagnosis of NSCLC ³⁷ and had undergone each of the following: 1.) Complete and accurate evaluation of disease extent which had indicated an almost certain or likely tumor resectability; 2.) Pre-treatment CEA and TPA serum tests; 3.) Thoracotomy made with curative intent, which, at least, had resulted in an accurate mediastinal exploration and pathological confirmation of the T and N status.

Anthropometric and clinical characteristics of the 124 assessable patients are shown in Table 1.

CEA and TPA assays

Sera for CEA and TPA were stored at -20° C and assayed three times per week in the central laboratory of the "S. Croce and Carle" hospital. The laboratory is located in the "S. Croce" hospital. It receives blood samples from many medical and surgical wards, including the lung division at the "A. Carle" hospital. Since we provide no clinical information, biologists have no means of knowing even the disease for which a particular test is required.

Assays were performed using commercial kits (CEA test, CIS Bio International, France; Prolifigen R, TPA IRMA, AB Sangtec Medical, Bromma, Sweden), and following the manufacturer's instruction. Normal reference values for CEA and TPA were up to 5 ng/ml and 90 U/L, respectively.

Computed tomography: technique and reading

All patients included in this report were studied with a CT of thorax, upper abdomen and brain. Up to October 1998, CT scans were performed on a conventional scanner (GE 9800, General Electric, Milwaukee, WI, USA); since then, a spiral-CT machine (CT twin flash, Elscint Ltd., Haifa, Israel) has been used. Ten millimeter-thick sections of the brain were obtained at 1-cm intervals, during suspended inspiration, from the lung apices to the upper abdomen. In selected cases, five mm-thick sections at five mm-intervals were acquired through the region of interest. Iodinated intravenous contrast (150-c³ bolus, plus 100-c³ in slow infusion) was injected prior to all studies. Appropriate windows were used for viewing both lungs and soft tissues.

Mediastinal nodes were labeled as abnormal if they were 1.5 cm or larger (transverse diameter). All CT scans were interpreted with no restriction to the clinical information available at the time of the exam.

Other staging procedures

Other diagnostic and staging techniques did not vary considerably during the 5 years of recorded data; furthermore, the frequent coexistence of experimental protocols, aimed to optimize diagnostic and staging procedures, ensured an overall accurate clinical assessment. All patients received a technetium-99m methylene diphosphonate bone scan of the entire body. In addition to this, the baseline clinical evaluation included also physical examination, routine lab tests, bronchoscopy, and functional respiratory tests. In half of the sample, the baseline work-up was supplemented by non-routine imaging studies, such as the anti-CEA monoclonal antibody scintigraphy ⁵). Other imaging tests, including x-rays, CT and magnetic resonance imaging of the bones, ultrasonographic studies of the abdomen, and other organ-specific investigations were optional, and performed as clinically indicated. Any information obtained in this way was considered part of the final clinical evaluation. The pre-operative staging evaluation was particularly reliable in 12 patients (10% of the cohort), who had a pathologic stage assessment made by mediastinoscopy (11 subjects) or CT-guided biopsy of a suspected (and unconfirmed) bone metastasis in one case. Since we considered still investigative both anti-CEA immunoscintigraphy and the marker assay, no clinical decision was made on the sole basis of their results.

All staging tests were obtained within a 3- to 4-week period and no thoracotomy was performed later than 30 days after the first physical examination.

Data analysis and statistical considerations

Diagnostic capabilities were calculated for the final clinical assessment, CT reading, CEA, TPA and a CEA-TPA combined variable (mean of the 2 marker values expressed in percentage of their reference of, respectively, 5 ng/mL and 90 U/L). For CEA, TPA and CEA-TPA, one or multiple threshold levels were chosen to describe a positive or a negative test. In this study, diagnostic capabilities are not intended to show the presence or absence of disease, but the presence (or absence) of the condition of full resectability, i.e. the pathological post-operative documentation of a stage Ia through IIb. Accordingly, a marker level below a given threshold or a CT result suggestive for stage Ia-IIb was declared true positive (TP) when the actual pathological stage was Ia-IIb, false positive (FP) when the latter was IIIa or more. Markers over the considered threshold or more advanced CT-stages were considered true negative (TN) or false negative (FN), when the corresponding pathological stage was, respectively, IIIa-IV or Ia-IIb.

Statistical analysis was performed using the SPSS package for Windows, Version 9.0 (SPSS Inc., Chicago, IL, USA). Continuous variables were described by medians and ranges because in many instances, such as in the case of CEA and TPA, their distribution was not Gaussian ³⁵. Sensitivity, specificity, accuracy rates, along with predictive positive and negative values, were obtained using standard diagnostic formulas ¹⁸. Diagnostic proportions were given along with their 95% confidence interval (CI) ¹². To compare diagnostic capabilities, we used the receiver-operating characteristic (ROC) curves ³², whose circumscribed areas (the area under the curve) give an estimate of the diagnostic efficiency ²¹. Correlations and differences were tested for statistical significance using the Spearman rank test and the Kruskall-Wallis ANOVA ³⁵. A probability (p) level <0.05 was considered statistically significant. All statistical tests were two-sided.

4. RESULTS

Descriptive Statistics

Table 1 shows the anthropometric and clinical characteristics of the patients’ population, with descriptive statistics of sex, age, history of weight loss, performance status (Eastern Oncology Group scale ³⁸), CEA and TPA serum levels. Also reported are tumor cell type, the post-operative pathologic stage of disease, the correlated parameters of disease extension, the type of surgical treatment, the survival duration and the patients’ status at the last follow-up re-assessment. As per May 1999, 64 out of 124 patients (52%) were still alive after a median follow-up time of 19 months (range 1-64). Most recruited patients had an early stage of disease and experienced favorable surgical outcomes. There were 88 completely operable patients (post-operative stages Ia/Ib/IIa/IIb), another 18 subjects who might have had benefit from the operation (stage IIIa), and 18 others who had a non-surgical condition (stage IIIb/IV). This resulted in 104 pulmonary resections and 20 explorative thoracotomies.

Median values (ranges) of CEA and TPA in the post-operative pathologic stage Ia-IIb were, respectively, 2 (0-60) and 70 (30-360). Corresponding values for the post-operative stage IIIa-IV were 2.5 (1-22) and 100 (40-790). The correlation between serum levels of TPA and the pathologic stage of disease was statistically significant (Spearman R=0.3, p=0.001), while that of CEA did not reach the significance level (Rs=0.162, p=0.072). No significant difference in the distribution of the serum markers’ concentration among different histotypes was also observed (Kruskall-Wallis statistic).

Diagnosis of post-operative pathologic stage, made by means of a 3-organ CT and the best clinical assessment

Table 2 shows a cross-tabulation between the 3-organ CT-stage and the best clinical assessment, on one side, and the post-operative pathologic stage, on the other. Based on CT, 105 patients were pre-operatively classified in stage Ia-IIb. Only 88 of them were post-operatively confirmed in stage Ia-IIb; on the contrary, there were 19 cases with CT-diagnosed stage IIIa-IV and 36 truly inoperable patients. This means a considerable CT underestimation of the real extent of disease, partially corrected by the completion of the full staging work-up. The final and best clinical assessment, however, showed no greater diagnostic accuracy, because of an increased overestimation of the real disease (Table 2).

Diagnosis of resectable disease (post-operative pathologic stages Ia-IIb) made by 3-organ CT findings, CEA and/or TPA serum levels

TPA and CEA differed remarkably in their capability to recognize early stages of disease. In particular, the diagnosis of post-operative stage Ia-IIb was only in part correct with TPA and inaccurate with CEA (Fig. 1a-1b). The area under the ROC curve was 0.621 (p=0.035) for TPA and 0.57 (p=0.201) for CEA (Table 3). For a TPA of 90 U/L or less, the diagnostic sensitivity was 67%, while specificity and accuracy rates were, respectively, 56% and 64% (Table 4). A normal level of TPA was properly associated with a resectable disease in 79% of the patients (95% CI: 69-88%), while a higher level was diagnostic of advanced disease in 41% (95% CI: 27-55%). TPA was capable to diagnose resectability also when the T or the N factors alone were taken into consideration (Fig. 2a-2b). For a CEA level up to 5 ng/mL, sensitivity, specificity and accuracy rates were, respectively, 82%, 14% and 62%; while the probability of properly diagnosing a condition of resectability was 70% (CI: 61-79). The combination of CEA and TPA (CEA-TPA combined variable) was unable to increase the overall diagnostic accuracy of TPA alone (Table 4).

Stratifying the CT-based stage of disease by marker results (Table 4), it was possible to recognize two different situations. The first was the presence of concordant findings (which increased both positive and negative predictive values of CT); the second was the occurrence of bi-directional discordant data (this lowered CT predictability). The TPA-stratified CT-stage was the best possible combination of CT and any one single marker (Table 4). As shown, the presence of a normal TPA level increased the chance, based on CT, of correctly predicting tumor resectability from 78% (CI: 70-86) to 83% (CI: 74-92). The opposite situation was associated to a diagnostic improvement of 21%, from the 68% rate of negative predictability of a sole CT suggesting non-resectability (CI: 48-89) to 89% (CI: 68-109). In the uncertainty area of an abnormal TPA and tumor resectability suggested by CT (or, conversely, of normal TPA and presence of advanced disease on CT), the CT diagnosis of resectable disease was correct in about 50-70% of the cases. The combination of CT, TPA and CEA (CT-stage Ia-IIb, TPA up to 90 U/L, and CEA up to 10 ng/ml) increased further the likelihood of correctly predicting a resectable disease (85%, CI: 76-94%). On the other end of the spectrum, a CT-diagnosis of stage IIIa-IV, associated with a CEA level above 10 ng/ml, was 100% diagnostic of unresectability (or 89% diagnostic in case of a sole TPA elevation). However, the area of the uncertainty increased from 40% of the sample (50 patients) to 52% (64 subjects) when two markers, rather than one, were used to stratify the CT-results (Table 4).

5. DISCUSSION

This study aimed to assess, in resectable NSCLC, the exact diagnostic capability of two serum biomarkers, CEA and TPA, which are commonly measured in many European countries ³⁶, but are still ignored by important medical societies ^{1; 27}. We can summarize our current findings as follows: 1.) The pre-operative assessment of tumor resectability, based on a CT scan of brain, thorax and upper abdomen, is acceptably accurate (accuracy rate: 77%); 2.) Its 20-25% margin of error is not appreciably reduced by the addition of any other pre-operative investigation; 3.) In patients considered for operation, a blind single serum test of TPA is diagnostic of full resectability (post-operative stages Ia-IIb) at an overall accuracy rate of 65%; 4.) Very elevated levels of CEA (above 10 ng/mL) are also capable of recognizing a post-operative resectable disease (accuracy rate: 69%); 5.) The stratification of CT-readings by TPA (serum test results up to 90 U/L or more) allowed to individuate a group of 9 patients, out the 19 supposed inoperable, who had a particularly high risk of unresectable disease (89%) and another 65, out of the 105 judged operable, whose tumors were more often resectable (83%). The remaining 50 subjects were left in an area of uncertainty that required further clinical testing; 6.) In slightly smaller groups, the stratification of CT data by both CEA (cut-off: 10 ng/ml) and TPA (cut-off: 90 U/L) made it possible to accurately discriminate between resectability and non-resectability increasing the negative predictive efficiency to 85-100%.

CEA is an oncofetal protein found normally in the embryonic and fetal gut, produced sometimes by malignant cells. It was discovered in 1965 by Gold and Freedman in the sera of subjects with adenocarcinoma of the colon ¹⁹. Raised CEA concentrations may be detected in smokers, in patients with benign tumors, and in 15-20% of subjects with inflammatory disorders such as ulcerative colitis, pancreatitis, liver disease, and pulmonary infections ³¹. In patients affected by lung cancer, abnormally elevated concentrations of CEA can be found in 30-70% of patients ^{9; 11; 22; 23; 28}. Raised levels of CEA are particularly common in adenocarcinoma ^{28; 33}; however, they can be present in any histologic type ^{22; 23}. Studies have shown that increased concentrations of CEA occur more frequently in advanced cancers ^{11; 23}, although reported differences are not always statistically significant ⁹.

Nearly half a century ago, B. Bjö rklund discovered a new antigen, by mixing many different tumors and producing an immune serum against the mixture ². He called it TPA. TPA has been subsequently identified as a degradation product of the cytoskeleton, formed by the cytokeratins 8, 18 and 19 ³. The cytoskeleton is a complex network that influences the dynamic morphology of all eukaryotic cells in their tissue environment ²⁵. It is composed of microfilaments (7-9.5 nm in diameter), microtubules (25 nm), and intermediate filaments (10-12 nm) ²⁵. Cytokeratins are the major components of the intermediate filaments ²⁴. Cytokeratins may be divided into 20 different types, according to molecular weight and isoelectric point ¹⁴. The expression of a single cytokeratin or a combination of certain cytokeratins is typical of a specific tissue. Elevated serum concentrations of TPA (i.e., of cytokeratins 8,18,19) have been observed in different types of epithelial cancers²⁰. In lung cancer, pretreatment levels of TPA are often elevated, particularly in the later stages of disease ^{6-11; 17; 26}.

As already mentioned, this is not our first attempt to define a clinical strategy capable of exploiting the well-known correlation between tumor burden and serum marker concentration. In 1995, in the pages of this journal, we reported a study specifically designed to investigate the use TPA as a common test of pre-operative assessment ⁸. We considered 104 patients with NSCLC, who had undergone either thoracotomy, or mediastinoscopy, or biopsy of suspected metastatic deposits in addition to an extensive non-invasive evaluation of the stage of disease. We made several retrospective evaluations, but the two most pertaining to this study were those regarding TPA and the readings of a computed tomography of brain, thorax, and upper abdomen. Among the 20 threshold values considered, ranging from 45 U/L to 450 U/L, we identified two thresholds for detecting the post-operative stage of disease with the highest rate of success. Then, based on the pathologic reference, we determined the sensitivity, specificity, accuracy, and predictive capabilities of both CT and TPA at the thresholds of 110 U/L, and 160 U/L. It was found that CT and TPA had a diagnostic accuracy of, respectively, 79%, and 68% for stage I and II; 69%, and 77% for stage IIIa; 77%, and 76% for stage IIIb and IV (1987 UICC classification). Impressed by those results we postulated the equivalence of TPA and CT in predicting surgical resectability. This study starts from that premise, but limits the focus to what we think the most clinically important information. Is the patient under assessment a surgical candidate or is he partially or completely inoperable? In other words, can we make a pre-operative diagnosis of post-operative pathologic stage Ia-IIb using a blind serum test? Based on the previous experience, the answer to that particular question is the least successful clinical application of TPA. In fact, CT was favored again, showing a 12% accuracy advantage over TPA (11% in the 1995 study). However, we felt it important to give priority to the clinical question, even if this would have implied a minor diagnostic capability of the marker test. Given this for granted, what are the answers to the posed questions? First, we obtained the confirmation we were looking for (the TPA predictability for full resectability was 65%, as compared to the previous one of 68%). Second, it was shown that CEA is a weak test "to mark" the burden of NSCLC, and that TPA is globally superior. However, using a cut-off level of 10 ng/mL, this classic biomarker also becomes useful and its diagnostic yield approaches that of TPA. Finally, the incorporation into the conventional pre-surgical evaluation of NSCLC of one or two inexpensive serum tests (i.e., TPA with or without CEA) helps to decide whether to rely on CT alone or to order additional clinical investigations.

6. CONCLUSIONS

Evidence from this study suggests obtaining a routine TPA test (and possibly a CEA test) in all potentially operable patients with NSCLC. Computed tomography remains the gold standard for the pre-operative evaluation of NSCLC. However, it may significantly underestimate the real extension of tumor. The TPA test is capable to correct such an underestimation, and helps to decide the next steps. We believe that a 3-organ CT, showing a resectable tumor (stages Ia-IIb) and a TPA level up to 90 U/L in an asymptomatic subject are clear indication to proceed straight to the operation. On the other hand, a higher TPA value (or a very high level of CEA), that is associated to CT findings of non-resectability (stages IIIa-IV), virtually eliminates any surgical approach. In case of conflicting data, the patient is still a surgical candidate, but an intensification of the pre-operative evaluation is mandatory. This could be obtained by ordering a bone scan or performing a mediastinoscopy, even in the absence of symptoms and signs.

 

7. REFERENCES