zyx zyx Prognostic Factors in Advanced Nonseminomatous Testicular Cancer A Multivariate Logistic Regression Analysis JEAN P. DROZ, MD," ANDREW KRAMAR, MSc,t MARWANE GHOSN, MD,* GILLES PIOT, MD,* ANNIE REY, AS,? CHRISTINE THEODORE, MD,* PIERRE WIBAULT, MD,* BERNARD H. COURT, MD,§ JEAN L. PERRIN, MD,§ JEAN P. TRAVAGLI, M D , ~DOMINIQUE BELLET, MD,/I J. M. CAILLAUD, MD,II JOSE L. PICO, MD,' AND MARCEL HAYAT, MD* zyxwvut zyxwvutsrqp In order to define prognostic factors for advanced stage of nonseminomatous germ cell tumors (NSGCr) of the testis, the authors reviewed 84 patients treated from 1978 through 1985. The survival rate was 51% at 3 years. Patients with elevated seric levels of human chorionic gonadotropin (HCG) and/or alpha-fetoprotein (AFP), or the presence of an abdominal mass had significantly worse survival. Only HCG and AFP levels retained their significance when multivariate Cox analysis was performed. The probability that a patient achieves a complete remission (CR) was assessed by a function of certain patient characteristics using a multivariate logistic regression analysis. The significant variables were a function of HCG and AFP values. Since both variables are related to the CR rate and survival the authors define the obtention of a CR as a unique outcome of interest. The probability of a CR greater than 70%adequately separates the patients into two prognostic subgroups. This model currently is being used to enrole NSGCT patients in a prospective modulated clinical trial according to these prognostic factors. Cancer 62564-568,1988. T Of advanced Stage of nonseminomatous germ cell tumors (NSGCT) of the testis has been greatly improved by modern chemotherapy regimens including cisplatin.1-4 Nevertheless the complete remission rate is still low in some subgroups of patients.*p3 Some prognostic factors have been described by different authors: size and number of lung metastasis,596 palpable abdominal mass,' visceral meta~tasis,~ performance status,' choriocarcinoma component," presence of human chorionic gonadotropin (HCG),'." HCG level,73 O3 2* alpha-fetoprotein (Am) level, 2, lactate dehydrogenase (LDH) level,7 rate of markers' decrease during ~hemotherapy,'~ prior treatment7," and time between diagnosis and treatment.I2 These factors often are correlated and their combined statistical imHE RESULT OF THE TREATMENT portance needs to be investigated with multivariate analyses. zyxwvu Patients and Methods Eighty-four patients with advanced stage of NSGCT of the testis, treated between January 1978 and February 1985, at Institut Gustave-Roussy (IGR) were reviewed. The histologic material of all the patients was reviewed by the same pathologist and classified according to the World Health Organisation (WHO) classification.'' The advanced stage was defined as the presence of a palpable abdominal mass (greater than 10 cm at computerized tomography [CT] scan or ultrasonography), and/or the presence of distant metastases. The biological markers, AFP, HCG, P-HCG, were determined in all patients before therapy by a radioimmunoassay. Alpha-fetoprotein was expressed in ng/l (normal, <30 ng/l) and HCG in mIU/ml(6.6 mIU/ml = 1 ng/l; normal, <10 mIU/ ml). Lactate dehydrogenase was performed in a very small proportion of patients. Patients were treated by three successive protocols (Table 1). Between January and December 1978, 18 patients were treated by IGR protocol T78. The IGR protocol T79 then was administered to 41 patients be- zyxwvutsr ' 3" ' From the Departments of *Medicine, tstatistics, $Radiotherapy, §Urology, (1 Biology, and llPathology,Institut Gustave-Roussy,France. Address for reprints: Jean-Pierre Droz, MD, Department of Medicine, Institut Gustave-Roussy, Rue Camille Desmoulins, 94800-Villejuif-France. The authors thank Dr. Goupil, Dr. Kerbrat, and Dr. Fargeot for providing data from the Urologic Intergroup from the French Federation of Cancer Centers, and Mrs. M. Belleteste for secretarial help. Accepted for publication January 9, 1988. 564 No. 3 zyxw zyxw zyxwvu PROGNOSTIC FACTORS IN TESTIS CANCER tween January 1979 and December 1981.16The last 25 patients were treated by IGR protocol T82 between January 1982 and February 1985. All patients were previously untreated before entering these three successive protocols. Droz et al. 565 TABLE1. Chemotherapy Regimens T78 T79 VAB-3 regimen (12 patients) PVB (6 patients) Doxorubicin 25 mg/m2 dl Vinblastine 7 mg/m2 d2 d 2, 3, 4, 5 in CI Bleomycin 7 mg/m2/d Dactinomycin 0.3 mg/m2/d d 3,4 Cyclophosphamide 200 mg/m2/d d 3,4 90 mg/m2 d4 Cisplatin Vinblastine I mg/m2 dl I0 mg/m2/d d 1, 2, 3, 4 in CI Bleomycin Dactinomycin 0.3 mg/m2/d d 2, 3 Cyclosphamide 200 mg/m2 d2,3 Cisplatin 100 mg/m2 d2 Analysis and Statistical Methods The initial characteristics of the patients, the histologic type of the tumor, the treatment regimens, and the follow-up data were registered on a specific form. Data were analyzed using a general database management sy~tem.’~ Complete remission (CR) was defined as the disappearance of any clinical or radiologic evidence of metastasis and normalisation of HCG and AFT markers. In case of clinical and/or radiologic partial response and normalization of HCG and AFP markers, patients were submitted to a surgical excision of residual disease. If the histologic pattern of the removed lesions was necrosis, fibrosis, or mature teratoma, the patient was classified as a complete response. All records were reviewed by one investigator. The criteria of CR were the same throughout the period of the study. The obtainment of a CR was defined as a unique goal for identifying prognostic factors. The importance of each known prognostic factor measured at diagnosis on the achievement of a complete remission was analysed using a x2 test. A logistic regression analysis” then was performed in order to identify prognostic factors for a predictive CR rate. Patients were then classified according to whether or not their probability of a CR was greater than 70%. The origin of survival was the date of the primary chemotherapy and the endpoint was the date of last follow-up or death of the patient. Survival curves were estimated by the Kaplan-Meier method. l9 The log-rank test” was used to select variables to be included in the Cox regression analysis.’l All variables which were significant at the 5% level in the univariate analysis were included in the multivariate analysis. Model selection for identifying the variables that have an important effect on survival was based on a forward stepwise procedure. The variables were entered on the basis of significance calculated from a large sample partial likelihood ratio test. . T82 VAB-3: vinblastine, bleomycin, actinomycin, doxorubicin, cisplatin, cyclophosphamide; PVB: cisplatin, vinblastine, bleomycin; CI: continuous infusion. no significant difference as to survival between the three groups. Three patients had some biological variables not available for study. Complete Remission The probability (P) of a patient with advanced-stage NSGCT achieving a CR was assessed by a function of certain patient characteristics using a multivariate logistic regression analysis. The characteristics included were those found significant in an univariate analysis (Table 2). The significant variables were HCG and AFP values and the presence of an abdominal mass, as well as the presence of visceral metastasis. The probability (Pi)for a patient i to be in CR is as follows: zyxwvuts zyxwvu TABLE2. Complete Remission Rate According to Patient Characteristics Prognostic variables Histopathologic characteristics Choriocarcinoma component No choriocarcinoma component Markers HCG (mIU/ml) <10 10-6000 >6000 AFP (ng/ml) 130 30- 1000 > 1000 Lung metastasis No 5 2 cm and <5/lung 2 2 cm and/or >5/lung Abdominal mass No Yes Visceral metastasis No Yes No. CR Percent Pvalue 46 38 28 23 60 60 NS 22 38 21 17 26 7 77 68 33 <0.01 42 22 17 28 17 ~0.01 5 66 77 29 31 21 32 19 16 16 61 76 50 47 37 33 18 70 48 o.04 76 8 49 64 25 o.03 zyx zyxwvut Results The clinical findings and the histologic features were not different between the three treatment protocols. The overall CR rate was 61%,and it was not significantly different between the three groups (IGR T78 = 12/18; IGR T79 = 26/41; IGR T82 = 13/25). The overall survival rate at 3 years was 5 1%. A log-rank analysis showed 2 NS CR: complete remission; HCG: human chorionic gonadotropin; AFP: alpha-fetoprotein;N S not significant. 566 zyxwvu zyxwvu zyxw z zyxwvut zyxwvutsrqponmlkj CANCERAugust 1 1988 .................................................. 1o 7'"' Vol. 62 the two groups (CR versus non-CR) and the correct predictions were similar. Complete Remission Analysis of an Independant Data Set 0 0.2 0.4 0.6 0.8 PROBABILITY The validity of our model was then tested against a sample of 29 patients treated by the Intercenter Urologic Group of the French Federation of Cancer Centers. The patients were previously untreated but received different classical regimens of cisplatin-containing chemotherapy. The observed responses were compared against responses predicted by our linear function and correctly predicted 72% overall. The correct prediction of not responding eventually was 69% (1 1/16) and the correct prediction of responding eventually was 77% ( 1O/ 13). zyxwvuts FIG.1. Probability of CR in observed non-CR group. zyxwvut Survival Pi = exp(hi)/(1 + exp(hi)) where hi is a linear function of several variables for patient i. The coefficients were estimated by maximizing the likelihood function and the significance of variables were tested using the likelihood ratio test. The following variables were retained by the model after a forward stepwise procedure: initial HCG and initial AFP values. The estimated linear function was as follows: hi = 1.90 - 0.033 VAFPi - 0.02 1 lHCGi + 0.033 HCGi/ 1.000 The correct overall prediction was 75% (61/81); the correct prediction of not responding eventually was 77% (24/3 1) (Fig. 1);and the correct prediction ofresponding eventually was 74% (Fig. 2). The cutoff point at 70%was chosen because it achieved the best separation between ................................ .... 20 > a hit) = hdt) X exp(0.634 H 1 + 1.495 H2 + 0.187 A 1 + 1.045 A2) where HI = 1 if 10 IHCG < 6000,O otherwise; H2 = 1 if HCG > 6000, 0 otherwise; A1 = 1 if 30 s AFP < 1000, 0 otherwise; and A2 = 1 if AFP 2 1000, 0 otherwise. Table 4 shows the relative risk of death for different combinations of HCG and AFP for this model. For example, the relative risk for a patient with HCG = 2000 and AFP = 300 is 2.42 times greater than the risk for the reference category which is defined as HCG < 10 and AFP < 30. 20 Discussion z 3 At the time of analysis, 39 of the 84 patients had died. The median follow-up was 3 years. The survival rate at 3 years was 5 1%. Eight variables were tested in a univariable analysis using the log-rank test (Table 3). Forward stepwise procedure using the Cox proportional hazards model revealed the prognostic significance of serum HCG and serum AFP levels. Interactions between HCG and AFP were tested for using a likelihood test and were not found to be significant. The hazard function of the Cox survival model is given by the following: In a 10 5 -I .................................. 0 0 0.2 0.4 0.8 0.6 PROBIBILITY FIG.2. Probability of CR in observed CR group. The broad range of survival rates for patients with NSGCT is very large and clearly related to the occurrence of a complete remission with the first treatment.'-3,12Our study has used a logistic regression analysis to analyze the impact of prognostic factors on the obtention of a complete remission and a Cox regression model to analyze the impact of prognostic factors on the survival. Both analyses has shown the significance of the level of serum HCG and the level of serum AFP. Since both factors are related to the CR and survival, we de- zyxwvutsrq zyxw zyxwvutsrq zyx No. 3 PROGNOSTIC FACTORS IN TESTISCANCER * DrOZ et d. 567 TABLE 4. Relative Risk of Death TABLE 3. Survival Analysis for Prognosis Factors Using Log-Rank Test AFP Variables HCG (mIU/ml) 510 10-6000 >6000 AFP (ng/ml) 530 30- 1000 >loo0 Abdominal mass ( < l o cm) No Yes Histopathologic characteristics Choriocarcinoma component No choriocarcinoma component Lung mtastasis No 5 2 cm and <5/lung >2 cm and/or >5/ lung Visceral mtastasis No Yes Protocol T78 T79 T82 2-yr survival rates No. of patients HCG P value <10 10-6000 26000 0.72 (0.10) 0.64 (0.08) 0.36 (0.11) 22 38 21 0.004 0.63 (0.08) 0.67 (0.10) 0.37 (0.13) 42 22 17 0.036 0.67 (0.07) 0.50 (0.08) 47 37 0.042 0.54 (0.08) 46 0.66 (0.09) 38 0.60 (0.07) 31 21 0.50 (0.09) 32 0.60 (0.06) 0.57 (0.19) 76 8 0.168 18 41 25 0.812 <30 30- 1000 2 1000 1 1.89 4.46 1.21 2.27 5.38 2.84 5.36 12.68 ~ HCG: human chorionic gonadotropin; AFP: alpha-fetoprotein. 0.210 0.446 good prognosis was chosen to be 0.50. Using this cutoff point they achieved an overall correct prediction of 83%, 56% of non-CR, and 94% of CR. When the cutoff point of 70% was used, the overall correct prediction is 79%, 84% of non-CR, and 77% of CR. When a cutoff point of 50% is used in our study, the results are an overall correct prediction of 75% with 55% of non-CR and 88% of CR. The results of this study are very similar to ours, even though our study was based on only 84 patients. The choice of a particular cutoff point in defining good and poor prognostic patients depends on the treatment strategy to be adopted and the balance between treatment efficacy and toxicity. Bajorin et al. tested three models on their data and came to the conclusion that all of them overestimated poor-risk patientsz2 The model developed at Indiana University" uses a clinicobiological classification issued from the M. D. Anderson A European Organization for Research on the Treatment of Cancer (EORTC) study" also used a logistic regression model and identified four group of patients according to the combination of the following prognostic factors: presence of trophoblastic component, level of HCG, presence of lung metastasis, and size and number of the lung metastasis. Our model is based on the HCG and AFP serum levels. These markers, when elevated or nonelevated, are closely related to the histologic component of the tumor, and their level is correlated to the tumor burden.z3Nevertheless, the assessment of a predicted CR value is closely dependent on the quality of the markers dosage. The same method of dosage is warranted for the reproductibility of this model. We emphasize the interest in defining prognostic subgroups of patients with numeri- zy zyxwvut zyxwvuts HCG human chorionic gonadotropin; AFP: alpha-fetoprotein. fine obtaining of a CR as a unique outcome of interest. The probability of a CR greater than 70% adequately separates the patients in two prognostic subgroups. No advantage is noted when an intermediate subgroup with a predicted CR rate between 0.30 and 0.70 is individualized (Table 5). Our predictive model has been tested on a population of 29 patients and the CR predictive rate also was valid in that population. The prognostic factors were first described in 1975.5 These factors were obtained by an univariate analysis of different variables. Since 1983 four multivariateanalysis were r e p ~ r t e d . ~ "The ~ - ' ~Cox regression model performed by the Medical Research Council Working Party on Testicular Tumors'* revealed three prognosticgroups identified by combination of the following variables: tumor volume, serum AFP, and serum HCG. The first multivariate logistic regression analysis was published in 1983.7It gives a continuous spectrum of predicted CR according to three prognostic factors: logarithm of serum HCG level and serum LDH level and the number of involved sites. The cutoff point between poor and TABLE 5. Observed Versus Predicted Response With Three Subgroups ~~ ~ Probability of CR Observed response Complete remission Yes No CR: complete remission. 530% 3 1%-69% 270% 3 14 10 10 37 7 568 zyxw zyxwv zyxwv CANCERAugust 1 1988 cal measurements instead of clinical assessment, which could be difficult to define according to the workup of patients. Currently, no model could be considered as universal, as has recently been pointed out.24It is thus highly recommended to take the greatest care in the interpretation of all published results concerning the treatment of poor-prognosis patients. The model presented in this study currently is being used in our institution to enroll previously untreated patients in a prospective clinical trial for the treatment of NSGCT according to prognostic factors. Good prognostic patients are treated with standard chemotherapy whereas poor prognostic patients are treated with high-dose chemotherapy followed by autologous bone marrow transplantation after two cycles of conventional intensive chemotherapy.*’ This trial validates the model and measures the efficacy and the toxic cost of intensified treatment in poor-prognosis patients. Vol. 62 cancer: Response to treatment and prognostic correlations. A Southwest Oncology Group study. Eur J Cancer 1980 16:1359- 1 366. 9. Von Eyben FE, Jacobsen GK, Pedersen H et al. Multivariate analysis of risk factors in patients with metastatic testicular germ cell tumors treated with vinblastine and bleomycin. Invasion Metastasis 1982; 2:125-135. 10. Stoter G, Sylvester R, Sleijfer DT et a/. Multivariate analysis of prognostic factors in patients with disseminated nonseminomatous testicular cancer: Results from an European Organisation for Research on Treatment of Cancer Multi-Institutional Phase 111 study. Cancer Res 1987; 47:2714-2718. 1 1. Birch R, Williams S, Cone A et a/. Prognostic factors for favorable outcome in disseminated germ cell tumors. J Clin Oncol 1986; 4~400-407. 12. Medical Research Council Working Party on Testicular Tumors. Prognostic factors in advanced nonseminomatous germ cell testicular tumors: Results ofa multicentric study. Lance? 1985; 1:8-11. 13. Javadpour N, Young JD. Prognostic factors in nonseminomatous testicular cancer. J Urol 1986; 135:497-499. 14. Lange PH, Vogelzang NJ, Goldman A, Kennedy BJ, Fraley EE. Marker half-life analysis as a prognostictool in testicular cancer. J Uro/ 1982; 128:708-711. 15. Mostofi FK, Sobin LH. Histological Typing of Germ Cell Tumors, no. 16. Geneva: World Health Organization, 1977. 16. Droz JP, Kramar A, Piot G et al. Treatment of advanced germ cell tumors of the testis: Gustave Roussy Institute experience (1979198 1). Forty-one cases (Submitted for publication). 17. Wartelle M, Kramar A, Jan P, Kruger D. PIGAS: An interactive statistical database management system. In: Proceedings of the International Workshop on Statistical Database Management, 1983; 124- 132. 18. Anderson S , Auquier A, Hauck WW, Oakes D, Vandaele W, Weisburg HI. Logit analysis. In: Statistical Methods for Comparative Studies. New York John Wiley and Sons, 1980; 161-273. 19. Kaplan EL, Meir P. Non parametric estimation from incomplete observations. J A m Stat Assoc 1958; 53:457-48 1 . 20. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Nut/ Cancer Inst 1959; 22:7 19748. 21. Cox DR. Regressions models and life tables (with discussion). J R Stat SOCSer 1972; B34: 187-220. 22. Bajorin D, Katz A, Bosl GJ, Myers WPL. Comparison of eligibility criteria of studies assigning germ cell tumor (GCT) patients (pts) to good risk (GR) and poor risk (PR) categories (Abstr). Proc Am Soc Clin Oncol 1986; 5: 104. 23. Droz JP, Sancho-Garnier H, Caillaud JM, Amiel JL, Brule G. Prognostic factors in advanced stages of non-seminomatous germ cell testicular tumors (Abstr). Proceedings of the 2nd European Conference on Clinical Oncology and Cancer Nursing, 1983; 153. 24. Vogelzang NJ. Prognostic factors in metastatic testicular cancer. Int JAndrol 1987; 10:225-237. 25. Droz JP, Pic0 JL, Ghosn M eta/.Preliminary results ofa modulated protocol of treatment for advanced stage of nonseminomatous germ cell tumors based on prognostic factors (Abstr). Proceedings of the 14th International Cancer Congress, 1986; Karger, Basel, 205. zyxwvutsrqponm zyxwvutsrq zyx zy REFERENCES I . Einhorn LH, Donohue JP. Cisdiamminedichloroplatinum, vinblastine and bleomycin combination chemotherapy in disseminated testicular cancer. Ann Intern Med 1977; 87:293-298. 2. Pontes JE, Wajsman Z, Beckley S, Williams P, Murphy GP. The treatment of Stage 111 nonseminomatous testicular tumors: Roswell Park Memorial Institute results (1970-1979). Cancer 1983; 51:13271331. 3. Vugrin D, Witmore WF, Golbey RB. Vinblastine, actinomycin D, bleomycin, cyclophosphamide and cis-platinum combination chemotherapy in metastatic testis cancer: A 1-year program. J Urol 1982; 128:1205-1208. 4. Peckham MJ, Barret A, Liew LH et al. The treatment of metastatic germ-cell testicular tumours with bleomycin, etoposide and cisplatin (BEP). Br JCancer 1983; 47:613-619. 5 . Samuels ML, Johnson DE, Holoye PY. Continuous intravenous bleomycin (NSC- 125066) therapy with vinblastin (NSC-49842) in Stage I11 testicular neoplasia. Cancer Chemother Rep 1975; 59563570. 6. Logothetis CJ, Samuels ML, Selig D et al. Improved survival with cyclic chemotherapy for nonseminomatous germ cell tumors of the testis. J Clin Oncol 1985; 3:326-335. 7. Bosl GJ, Geller NL, Cimncione C et a/. Multivariate analysis of prognostic variables in patients with metastatic testicular cancer. Cancer Res 1983; 433403-3407. 8. Samson MK, Fischer R, Stephens RL et a/. Vinblastine, bleomycin and cis-diamminedichloroplatinum in disseminated testicular