Cyfra 21-1 as a Biologic Marker of Non-small Cell Lung Cancer*

Transcription

1 Cyfra 21-1 as a Biologi Marker of Non-small Cell Lung Caner* Evaluation of Sensitivity, Speifiity, and Prognosti Role Bram Wieskopf, MD; Christine Demangeat, PhD; Ashok Purohit, MD; Rodolphe Stenger, MD; Pasal Gries, MD; Harvey Kreisman, MD; and Elisabeth Quoix, MD Bakground: Cytokeratins are epithelial markers whose expression is not lost during malignant transformation. Cyfra 21-1 is a ytokeratin-19 fragment that is soluble in serum and may be a useful irulating tumor marker. Study objetive: The aims of this study were (1) to onfirm sensitivity and speifiity of Cyfra 21-1 in deteting non-small ell lung aner (NSCLC) and espeially the squamous ell subtype, (2) to assess the potential relationship between Cyfra 21-1 and disease stage of the disease in NSCLC, and (3) to evaluate prognosti effet of Cyfra 21-1 in NSCLC. Methods: An immunoradiometri assay of serum Cyfra 21-1 was performed in 161 patients with lung aners and 71 others with benign lung diseases. The ability of Cyfra 21-1 to detet different histologi subtypes of lung aner vs benign lung diseases was assessed through reeiver operating harateristi (ROC) urves and omparisons with other tumor markers suh as arinoembryoni antigen, neuron-speifi enolase, and squamous ell arinoma antigen. Comparisons of Cyfra 21-1levels aording to histologi subtype and disease stage were done using Kruskal-Wallis test. Independent prognosti value of Cyfra 21-1 was studied with a multivariate analysis of survival (Coxs model). Results: Using a threshold of 3.3 ng/ml for Cyfra 21-1, sensitivity and speifiity were, respetively, 0.59 and 0.94 in NSCLC, 0.68 and 0.94 in the subgroup of the squamous ell arinoma, and 0.19 and 0.94 in small ell lung aner. Cyfra 21-1levels were signifiantly higher in advaned NSCLC than in early-stage disease. All 29 patients with serum onentrations >32 ng/ml had stage IIIB-IV and only one of 14 patients with stage I-II disease had Cyfra 21-1 level > 18 ng/ml. In the multivariate analysis of survival, Cyfra 21-1 was an independent prognosti fator along with performane status and disease stage in NSCLC. Conlusions: Cyfra 21-1 is a sensitive and speifi tumor marker of NSCLC, espeially of squamous ell subtype. It also reflets the extent of the disease and has an independent prognosti role along with performane status and disease stage in NSCLC. Impliations: A high level of Cyfra 21-1 in apparently early-stage NSCLC should be an indiation for more extensive workup before thoraotomy. The independent prognosti role of Cyfra 21-1 level may be useful in stratifying populations with advaned NSCLC or earlystage reseted NSCLC as elevated Cyfra 21-1 levels might identify those patients at high risk for treatment failure. (CHEST 1995; 108:163-69) CEA =arinoembryoni antigen; FN =false-negative; FP= false-positive; LDH =latate dehydrogenase; mab=monolonal antibody; NSCLC=non-small ell lung aner; NSE=neuron-speifi enolase; ROC=reeiver operating harateristi; SCC=squamous ell arinoma antigen; SCLC=small ell lung aner; TN =true-negative; TP=truepositive Key words: Cyfra 21-l; non-small ell lung aner; prognosti fator; squamous ell lung aner; tumor markers Formerly relatively rare, lung aner is now the most ommon fatal malignany in both men and women. There are over 150,000 new ases diagnosed every year in the United States. 1 *From the Pavilion Laenne, Hopitaux Universitaires de Strasbourg, Strasbourg Cedex, Frane (Drs. Purohit, Stenger, and Quoix); the Sir Mortimer B. Davis Jewish General Hospital and MGill University, Montreal, Canada (Drs. Wiesko_pf and Kreisman); the Institut de Biophysique, Faulte de Medeine, Strasbourg Cedex, Frane (Dr. Demangeat); and the Servie de Medeine Nuleaire, Centre Hospitalier Regional, Haguenau, Frane (Dr. Gries). Manusript reeived August 15, 1994; revision aepted November 4. Reprint requests: Dr. Quoix, Pavilion Laenne, 1 Plae de l Hopital Strasbourg Cedex, Frane The four most important histologi ategories in terms of both frequeny and linial signifiane are squamous ell arinoma, small ell arinoma, adenoarinoma, and large ell arinoma. Approximately 90% of all lung arinomas are inluded in these four main ategories. 2 The prognosis of lung aner is generally poor and only slight improvement in survival has been obtained during the last deades. In non-small ell lung aner (NSCLC), 5-year survival varies from 50% in reseted stage I disease to 5% in stage IIIB and virtually 0% in stage IV. 3 In small ell lung aner (SCLC), median survival is 10 to 18 months in the limited disease stage and 7 to 12 months in extensive CHEST / 108/1 I JULY,

2 disease. 4 Some of the prognosti fators ommon to NSCLC and SCLC are the stage and performane status. 5 6 Tumor markers, inluding arinoembryoni antigen (CEA), neuron-speifi enolase (NSE), and squamous ell arinoma (SCC) antigen have been investigated for putative diagnosti and prognosti value. Many lung adenoarinomas ontain immunoreative CEA, but there is no speifiity at all for lung origin. 7 NSE, a neural form of a glyolyti enzyme present in all human ells, is found to be elevated in a majority of patients with SCLC. 8 Serial NSE levels tend to mirror the patients linial ourse, falling with remission, rising with relapse. 9 Unfortunately, NSE laks speifiity and like CEA, oasionally is elevated in nonmalignant onditions. se is a tumor antigen purified from a ervial squamous ell arinoma and has a sensitivity between 33 and 61%. 10 Cytokeratin 19 is an aidi protein of 40 kd that is part of the ytoskeleton of epithelial ells. 11 The distribution of the intermediate filament is exlusive to ells of simple and pseudostratified epithelium (suh as the bronhial epithelium). Cytokeratin 19 is speifially reognized by two monolonal antibodies (mabs), KS 19-1 and BM Serum levels of Cyfra 21-1 have been determined in healthy individuals and those with benign lung disease. Of 711 healthy blood donors, 99.8% had a Cyfra 21-1level below 1.2 ng/ ml. In a group of 546 patients with benign lung diseases, 96% had levels below 3.3 ng/ ml.l 2 Reent results suggest that Cyfra 21-1 has a good sensitivity and speifiity for the diagnosis of squamous ell arinoma as well as prognosti value. 13 The aims of the urrent study are to (1) investigate the sensitivity and speifiity of Cyfra 21-1 for squamous ell arinoma and other subtypes of NSCLC, (2) identify any relationships between elevated Cyfra 21-1levels and other tumor markers, and (3) evaluate the putative independent prognosti value of Cyfra 21-1 in NSCLC. Patients METHODS One hundred sixty-one patients with lung aner admitted to the Department of Chest Disease (University Hospital, Strasbourg, Frane) between Marh 1989 and June 1993 were prospetively entered in the study (Table 1). All had histologially and/ or ytologially onfirmed lung aner. There were 72 with squamous ell arinomas, 29 with adenoarinomas, 15 with large ell arinomas, and 45 with SCLC. Karnofsky performane status was noted and staging 3 of NSCLC was done by linial examination; hest radiography; bronhosopy; omputed tomographi san of hest, upper abdomen, and brain; abdominal ultrasonography; and bone san. Following idential staging for SCLC, limited disease was defined as any disease limited to the hemithorax, inluding mediastinal and/ or ipsilateral supralav- iular lymph nodes. Extensive disease was defined as any disease extending beyond the hemithorax by invasion or metastasis. 14 In addition to Cyfra 21-1, CEA, NSE, and total latate dehydrogenase (LDH) serum levels were performed. SCC levels were determined only in patients with squamous ell arinoma. Control Subjets The serum Cyfra 21-1 was measured in 71 ontrol subjets with nonmalignant pulmonary disease (saroidosis, COPD, asthma). CEA, NSE, and SCC levels were performed in 38 of the 71 nonmalignant ontrols. Cyfra 21-1 Immunoradiometri Assay A speial kit (ELSA-Cyfra 21-1) is used for the immunoradiometri assay of Cyfra 21-1 in human serum 15 Serum samples were separated and stored at -30 C until tested. Cyfra 21-1 (CIS Biointernational; GIF / Yvette, Frane) is a solid-phase "sandwih" immunoradiometri assay. The ytokeratin 19 was reognized by two mouse mabs, KS 19-1 and BM 19-21, direted against two different epitopes present on the soluble fragment of the ytokeratin, namely serum Cyfra 21-l. These mabs had been obtained by immunization against the MCF-7 ell line. KS 19-1 oated ELSA tubes were inubated with 100 JLL of the patients serum, a ontrol serum, and a standard urve (omposed of the following onentrations of ytokeratin 19: 0, 2.1, ll.5, 34, and 65 ng/ ml) and 300 JLL of iodine-1251abeled BM (18.5 kbq/ ml) for 20 h at 8 C. Afterwards, the ELSA tubes were washed with a solution of 9 ml of polysorbate 20 (Tween 20) in 3 L of distilled water. Radioativity was ounted in a gamma sintillation ounter alibrated for iodine-125 (LKB ounter) and expressed in nanograms per milliliter. The amount of radioativity bound to the ELSA is proportional to the amount of Cyfra 21-1 present at the beginning of the assay. The CEA, SCC, and NSE assays were performed, respetively, with different analyzers (Boehringer ES 600 [ELISA assay], IMx Abbot [MEIA assay], and Pharmaia kit [double-antibody Table!-Charateristis of the Patients with NSCLC and SCLC Charateristi M/ F Mean age, yr (range) NSCLC Karnofsky performane status >70 :::;70 Unknown Histologi features Squamous ell Other NSCLC Stage I-II Ilia Illb IV SCLC Disease extent Limited disease Extensive disease Karnofsky performane status >70 :::;70 Unknown 146/ (35-82) Clinial Investigations

3 radioimmunoassay]). Upper limit of normal values are 5 ng/ ml for CEA, 1.5 ng/ ml for SCC, and 12.5 ng/ ml for NSE Statistis Sensitivity and speifiity of Cyfra 21-1 as a serologi marker were determined. Sensitivity is defined as the probability of testing positive if the disease is truly present, whereas, speifiity is the probability of sreening negative if the disease is truly absent. Sensitivity=TP / TP+ FN and speifiity=tn / TN+ FP where TP is true-positive, TN is true-negative, FP is false-positive, and FN is false-negative 19 To evaluate the ability of tumor markers to predit the histologi type, reeiver operating harateristi (ROC) urves were plotted. As an example, in squamous ell arinoma for different utoffs of Cyfra 21-1 levels, the ROC urve shows the proportion of patients with a TP test (ie, patients with squamous ell arinoma and a serum Cyfra 21-1 above a given level) as ompared with the proportion with a FP test (ie, patients with benign lung disease and a serum Cyfra 21-1 above the given level). If a test is of no use, then both proportions are roughly equal for all values and the ROC urve is a straight line with a slope of l. A useful test is indiated by a ROC urve that rises rapidly then reahes a plateau; the point of infletion represents the value of the test giving the best ompromise between the TP and FP. 20 Results of the serum markers were expressed in eah subset of patients as median and variation as interquartile range. To ompare quantitative variables, nonparametri statistial analysis was used. Differenes between more than two groups were evaluated by means of the Kruskal-Wallis one-way analysis of variane. Correlation between quantitative variables was established through Spearmans oeffiient. Survival analysis was performed for NSCLC. Survival was defined as the time from the date of serum sampling to the date of death or to the date of analysis, on Otober 31, Probability of survival was estimated through the Kaplan-Meier method. Univariate omparisons of survival urves were done by means of the log rank test and multivariate analysis of survival was performed using the Cox model. 21 Before inlusion in the model, quantitative variables were ategorized to allow a baseline hazard ratio (Karnofsky performane sore ::570 or >70, LDH ::5 upper limit of normal or > upper limit of normal, Cyfra 21-1 ::53.3 ng/ ml or >3.3 ng/ ml, and ACE :55 ng/ ml or >5 ng/ ml). All statistial alulations were performed using a software pakage (BMDP, Statistial Graphis System). RESULTS Reprodutibility of Cyfra 21-1 Radiometri Assay Reprodutibility of determination of serum on , /... / 88. lie I. I.Ji. CEll ~ 58 ~ z >.. I.a ;: 48 lli 5.35 ~ IL FP <188 - ~ i f i in 7.) i t y FIGURE l. ROC urves of all NSCLC vs benign lung diseases. The areas under the ROC urve for Cyfra 21-1, NSE, and CEA were 0.85, 0.78, and 0.66, respetively. CHEST /108/1 I JULY,

4 1ee _a=- + - ~ ~ ~ f. ae 2.0 _8 X.:.. >.. ~ ~ 70 Sll 50 4ll UJ-- lie. se o lli ; 1.2,. i i : NSE tillli /... -!, 4. 6.f. f.cea f. +.j.... CL l ll FIGURE 2. ROC urves of squamous ell aner vs benign lung diseases. The areas under the ROC urve for Cyfra 21-1, SCC, NSE, and CEA were 0.88, 0.81, 0.8, and 0.67, respetively. ll lll 3ll 4ll 5ll 6ll FP Clllll-speifiit\1 in%) 78 ae entrations was examined by measuring a ontrol serum ontrol in 13 different assays (interassay). The mean ± standard deviation and oeffiient of variation was 3.67 ng/ ml ± 0.41 and 4.6%, respetively. Cyfra 21-1 and Histology The median (interquartile) serum Cyfra 21-1 in all NSCLC, squamous ell arinoma, NSCLC other than squamous ell, and SCLC were 4.3 (2.05 to 16), 6.0 (2.55 to 19), 2.6 (1.25 to 9.75), and 1.7 (0.98 to 2.4) ng/ ml, respetively. The median (interquartile) serum Cyfra 21-1 in nonmalignant ontrol subjets was 1.0 (0.87 to 1.4). The Cyfra 21-1 level was found to vary signifiantly with respet t the ~ different histologi groups (p=l.5xl0-15). ROC Curves for Serum Cyfra 21-1 and Other Tumor Markers A ROC urve with various utoff levels of Cyfra 21-1 was first onstruted to study the ability of this marker to predit NSCLC (Fig 1). For all NSCLC, the best ompromise between TP (80.5%) and FP (19.5%) was given by the threshold of 1.5 ng/ ml. At the threshold of 3.3 ng/ ml, TP rate was 59% and FP rate was only 5.6%. For squamous ell arinoma, the best ompromise between TP (83.1%) and FP (16.9%) was given by the threshold of 1.5 ng/ ml (Fig 2). At the threshold of 3.3 ng/ ml, speifiity was greater (94.4%) at the expense of a loss of sensitivity (68.1%). For other NSCLC, the best ompromise was also given by the threshold of 1.5 ng/ ml with a sensitivity of 70.2% and a FP rate of 14%. At the threshold of 3.3 ng/ ml, sensitivity was 40.9% and speifiity was 94.4%. CEA and NSE had a limited ability to predit either NSCLC (Fig 1) or squamous ell arinoma (Fig 2) or other NSCLC. For squamous ell arinoma, SCC with a threshold of 1.5 ng/ ml was a less sensitive marker than Cyfra 21-1 (sensitivity=51 %; speifiity=87.5%). There was a moderate but signifiant orrelation between se and Cyfra 21-1 (Spearmans oeffiient=0.35, Clinial Investigations

5 " 35 Cl II I II IliA nib IV FIGURE 3. Distribution of Cyfra 21-1 aording to the stage of all NSCLC presented as multiple box and whisker plots showing median value, upper and lower quartile, 5th perentile, and 95th perentile. p=0.007). In SCLC, Cyfra 21-1 was not as good as NSE in prediting SCLC (area under ROC urve 0.68 for Cyfra 21-1 vs 0.94 for NSE). At the threshold of 3.3 ng/ ml, sensitivity of Cyfra 21-1 was 19% and speifiity was 94.4%. Cyfra 21-1 and TNM Stage The Cyfra 21-1level differed signifiantly aording to stage both in all patients with NSCLC and in those with squamous subtype (p=9xl0-4 and 0.006, respetively). Among all NSCLC, the median (interquartile range) serum Cyfra 21-1 of stage I-11, Ilia, Illb, and IV was 1.1 (0.75 to 2), 5.8 (2.3 to 12), 5.7 (1.8 to 16), and 5.35 (2.7 to 25), respetively (Fig 3). Among squamous ell arinomas, the median (interquartile range) serum Cyfra 21-1 of stage I-II, Ilia, Illb, and IV was 1.05 (0.75 to 1.80), 4.15 (1.10 to 10.4), 5.70 (1.8 to 11.0), and 12.4 (4 to 34.5), respetively. All 29 patients with a Cyfra 21-l serum onentration above 32 ng/ ml had stage IIIB or l V. Only one patient with stage I-II disease had a Cyfra 21-l serum onentration above 18 ng/ ml. Prognosti Role of Cyfra 21-1 in NSCLC Univariate analysis of survival of all patients with NSCLC showed a signifiant differene aording to performane status (<70 vs >70), stage, Cyfra 21-1 level (<3.3 or >3.3 ng/ ml), CEA level (<5 vs >5 ng/ ml), and LDH (below or more than the upper normal limit). There was no signifiant differene in survival aording to age (<60 years vs >60 years), sex, and histologi subtype. All variables with signifiant prognosti impat in univariate analysis were inluded in Coxs model. In addition, sex and age were fored into the model. Proportionality of hazards hypothesis was verified for eah of the variables. Performane status, stage, and Cyfra 21-l serum onentration were the only independent prognosti fators (Table 2). DISCUSSION Our study onfirms that Cyfra 21-l is a both sensitive and speifi tumor marker for NSCLC and espeially for squamous ell arinoma. It appears more sensitive and more speifi than other tumor markers suh as CEA, NSE, and slightly better than CHEST / 108 / 1 I JULY,

6 Table 2-Coxs Regression Analysis: Estimated Relative Risks (n=105) Variable Performane status Cyfra 21-1 Stage Relative Risk (95% Confidene Interval) 5.25 ( ) 2.8 (1.5-5) 1.5 (1.1-2) p V alue SCC in squamous ell arinoma. The inability of Cyfra 21-1 to detet SCLC was also onfirmed. The Cyfra 21-1 level in NSCLC orrelates with the stage and, moreover, it appears as an independent prognosti fator along with performane status and stage. Cytokeratins are not randomly distributed in different epithelia, but they appear to be harateristi for ertain types of epithelial differentiation. Their expression is not lost by malignant transformation of epithelial ells. Therefore, subtypes of lung aner demonstrate harateristi ytokeratin profiles that an be evaluated immunohistohemially to assist with tumor lassifiation. 22 Although the normal role of the ytokeratin as a omponent of ytoskeleton is unknown, nerosis and ell lysis are likely to be the events preeding the release of the ytokeratin 19 fragment into the blood. Cell type heterogeneity may explain some differenes between the published studies with respet to sensitivity of Cyfra 21-1 in the various histologi subtypes of lung aner. Table 3 shows that the magnitude of these differenes is very small for NSCLC and rather large for SCLC. Elevated Cyfra levels in SCLC ould be either due to tumor heterogeneity with subpopulations of squamous ell differentiation or due to ytokeratin expression by the SCLC tumor ells. 25 There is a strong relationship between serologi Cyfra 21-1 levels and staging of NSCLC. Elevated Cyfra 21-1levels were usually indiative of extensive disease, findings that are very onsistent with those of the previous published studies A primary objetive of onology health-are workers is to determine those patients with NSCLC who would benefit from surgial resetion, ie, patients with stage I and II and some with stage IliA. Aurate determination of the stage is strongly related first to the extent of the staging proedures and seond to their sensitivity and speifiity. In this setting, biologi variables that would be able to predit extensive disease are potentially very useful. In this study, serum onentrations >32 ng/ ml were assoiated with stage IIIB or IV disease. Careful staging to rule out advaned disease is indiated in patients having Cyfra 21-1levels of this magnitude. Even though Cyfra 21-1 level was orrelated with stage of disease in NSCLC, it nevertheless was an independent prognosti fator in our study. Pujol et al 13 found that Cyfra 21-1 was an independent prognosti fator in their entire population of lung aners (NSCLC and SCLC) along with performane status and disease stage (relative risk [95% onfidene interval], respetively, 1.64 [1.1 to 2.5]), 2.81 [1.8 to 4.3], and 2.47 [1.4 to 4.3]). The importane of Cyfra 21-1 as a prognosti fator in their study was inferior to that found by us, as they inluded patients with SCLC in their analysis. The haraterization of prognosti fators in a population with NSCLC is of primary importane in assessing the effiay of a new treatment, whether it be hemotherapy for advaned disease or adjuvant therapy for reseted early-stage disease. For instane, it is known that performane status will impat as muh as hemotherapy on the prognosis of stage IV NSCLC. Also, reent studies have emphasized the prognosti signifiane of K-ras mutations and ploidy Table 3-Sensitivity and Speifiity of Cyfra 21-1 in Various Histologi Subtypes of Lung Caner* Cut-off Level of Cyfra 21-l, Soure ng/ ml Histologi Subtype Se Sp Pujol et aii NSCLC SQC SCLC Ebert et al NSCLC NO NO SQC SCLC Stieber et al U/ ml 1 NSCLC SQC SCLC Present Study 3.3 NSCLC SQC SCLC AUROCC *SQC=squamous ell arinoma subtype; NO=not done; S e=sensitivity; Sp=speifiity; AUROCC=area under ROC urve.!conentrations of the prototype version are 2.5-fold higher ompared with urrently available test. 168 Clinial Investigations

7 in reseted NSCLC, espeially adenoarinomas. 26 Another independent prognosti fator suh as serum Cyfra 21-1 may serve to explain some of the differenes in survival between different populations with advaned NSCLC who were given the same hemotherapy. Hypothetially, in reseted early-stage NSCLC, elevated Cyfra 21-1 level might identify those patients at high risk for reurrene and thus selet a population for trials of adjuvant therapy. We onlude that Cyfra 21-1, a soluble fragment of ytokeratin 19, is more sensitive and speifi than other markers in NSCLC. Its level orrelates with disease stage in NSCLC. Cyfra 21-1 performane status and disease stage are independent prognosti fators in NSCLC. ACKNOWLEDGMENTS: We are grateful to Yolande Ledermann for exellent tehnial assistane. REFERENCES 1 Carr DT. Histopathology of lung aner. Int J Epid 1990; 19(suppi):S8-Sl0 2 Hansen H. Histopathologial typing of lung aner: etiologi and epidemiologi features. In: MVie JS, ed. Clinial and experimental pathology of lung aner, developments in onology 39. Boston: Martinus Nijhoff, 1986; 2 3 Mountain CF. A new international staging system for lung aner. Chest 1986; 89(suppl): Albain KS, Crowley JJ, Livingston RB. Long-term survival and toxiity in small ell lung aner: expanded Southwest Onology Group experiene. Chest 1991; 99: Spiegelman D, Maurer LH, Ware JH, et al. Prognosti fators in small-ell arinoma of the lung: an analysis of 1,521 patients. J Clin Onol1989; 7: Bonomi P, Gale M, Rowland K, et al. Pre-treatment prognosti fators in stage III non-small ell lung aner patients reeiving ombined modality treatment. Int J Radiat Onol Phys 1991; 20: Sun N, Edgington T, Carpentier C, et al. Immunohistohemial loalization of arinoembryoni antigen, CEA-S, and nonspeifi ross-reating antigen in arinomas of lung. Caner 1983; 52: Akoun GM, Sarna HM, Milleron BJ, et al. Serum neuron-speifi enolase: a marker for disease extent and response to therapy for small ell lung aner. Chest 1985; 87: Muller LC, Gasser R, Huber H, et al. Neuro-speifi-enolase in small ell lung aner: longitudinal tumors marker evaluation. Lung Caner 1992; 8: Mino N, lio A, Hamamoto K. Availability of tumor-antigen as a marker of squamous ell arinoma of the lung and other organs. Caner 1988; 62: Moll R, Franke WW, Shiller DL, et al. The atalog of human ytokeratins: patterns of expression in normal epithelia, tumors and ultured ells. Cell1982; 31: Rastel D, Ramaioli A, Cornillie F, et al. Cyfra 21-1, sensitive and speifi new tumour marker for squamous ell lung aner: report of the first European multienter evaluation. Eur J Caner 1994; 30A: Pujol JL, Grenier J, Daures JP, et al. Serum fragment of ytokeratin subunit 19 measured by Cyfra 21-1 immunoradiometri assay as a marker of lung aner. Caner Res 1993; 53: Zelen M. Keynote address on biostatistis and data retrieval. Caner Chemother Rep 1973; 4: Kasper M, Stosier P, Typit H, et al. Histologial evaluation or three new monolonal anti-ytokeratin antibodies. Eur J Caner Clin Onol1987; 23: Sulier JP, Body JJ, Jaobowitz D, et al. Value of CEA determination in biologial fluids and tissues. Eur J Caner Clin Onol1987; 23: Ebert W, Leihtweis B, Biilzebruk H, et al. The role of IMx se assays in the detetion and prognosis of primary squamousell arinoma of the lung. Diagn Onol1992; 2: Cooper EH, Splinter TAW, Brown DA, et al. Evaluation of a radioimmunoassay for neuron-speifi enolase in small ell lung aner. Br J Caner 1985; 52: Hennikens C, Buring J. Epidemiology in mediine. Boston: Little Brown & Co, Robertson E, Zweig H. Use of reeiver operating harateristi (ROC) urves to evaluate the linial performane of analytial systems. Clin Chern 1987; 27: Cox DR. Regression models and life-tables. J R Stat So 1972; 34: Broers J, Ramaekers F, Klein R, et al. Cytokeratins in different types of human lung aner as monitored by hain-speifi monolonal antibodies. Caner Res 1988; 48: Ebert W, Liehtweis B, Shapohler B, et al. Der neue tumormarker yfra ist der priitherapeutishen diagnostik des bronhialkarinoms den markern SCC antigen und CEA iiberlegen. Tumordiagn U Ther 1993; 14: Stieber P, Hasholzner V, Bodenmiiller H, et al. Cyfra 21-1: a new marker in lung aner. Caner 1993; 72: Moll R, Blobel CA, Franke WW. Zytoskelett-proteine in der diagnostik der bronhialkarzonime. In: Drings P, Shamiihl D, Vogt-Moykopf I, eds. Miinhen: Verlag, 1986; Rosell R, Gomez-Codina J, Camps C, et al. A randomized trial omparing preoperative hemotherapy plus surgery with surgery alone in patients with non-small ell lung aner. N Eng! J Med 1994; 330: CHEST /108/1 I JULY,