Predictive Value of the Pretreatment Extent of Disease System in Hepatoblastoma: Results From the International Society of Pediatric Oncolog
http://www.100md.com
《临床肿瘤学》
the Pediatric Surgical Center of Amsterdam, Emma Children’s Hospital Academic Medical Center, Vrije Universiteit Medical Center
Departments of Radiology and Clinical Epidemiology and Biostatistics, Academic Medical Center, Amsterdam, the Netherlands
Department of Pediatric Surgical Unit, Universitts-Kinderklinik, Inselspital, Bern, Switzerland
Department of Pediatrics, Universita di Padova, Padova, Italy
Northern and Yorkshire Clinical Trial and Research Unit, University of Leeds, Leeds
Department of Surgery, Royal Hospital for Sick Children, Edinburgh, United Kingdom
Department of Pediatric Surgery and Liver Transplantation, Universite Catholique de Louvain, Clinique Saint-Luc, Brussels, Belgium
Department of Pediatric Surgery, Medical University of Gdansk, Gdansk, Poland
ABSTRACT
PATIENTS AND METHODS: Hepatoblastoma (HB) patients younger than 16 years who underwent surgical resection (128 of 154 patients) were analyzed. The centrally reviewed preoperative staging was compared with postoperative pathology (accuracy) in 91 patients (81%), and the local center staging was compared with the central review (interobserver agreement) in 97 patients (86%), using the agreement beyond change method (weighted {kappa}). The predictive values of the three staging systems were compared in 110 patients (97%) using survival curves and Cox proportional hazard ratio estimates.
RESULTS: Preoperative PRETEXT staging compared with pathology was correct in 51%, overstaged in 37%, and understaged in 12% of patients (weighted {kappa} = 0.44; 95% CI, 0.26 to 0.62). The weighted {kappa} value of the interobserver agreement was 0.76 (95% CI, 0.64 to 0.88). The Children's Cancer Study Group/Pediatric Oncology Group–based staging system showed no predictive value for survival (P = .516), but the tumor-node-metastasis–based system and PRETEXT system showed good predictive values (P = .0021 and P = .0006, respectively). PRETEXT seemed to be superior in the statistical fit.
CONCLUSION: PRETEXT has moderate accuracy with a tendency to overstage patients, shows good interobserver agreement (reproducibility), shows superior predictive value for survival, offers the opportunity to monitor the effect of preoperative therapy, and can also be applied in patients who have not had operations. For comparability reasons, we recommend that all HB patients included in trials also be staged according to PRETEXT.
INTRODUCTION
In the SIOPEL-1 prospective trial, a preoperative surgical staging system, the pretreatment extent of disease (PRETEXT) system, which was based on the anatomy of the liver, was developed and adopted.10,11 The main difference from other well-known liver tumor staging systems, such as the tumor-node-metastasis system of the International Union Against Cancer and the system used by the Children's Cancer Study Group (CCSG) and the Pediatric Oncology Group (POG),2,4,12 is that the PRETEXT system was especially developed to compare the efficacy of various chemotherapeutic regimens in HB and to stage the tumor before surgical treatment, whereas the other two systems stage the tumor postoperatively. PRETEXT was used as a relatively objective but noninvasive method to assess tumor extent at diagnosis and subsequent chemotherapy response and to determine the optimal time and type of resection. Its ultimate goal was to ascertain preoperatively whether it would be possible to perform a radical resection.
In 1997, von Schweinitz et al13 investigated the predictive impact of the different staging systems (mentioned in the previous paragraph) in 72 patients treated in the German Pediatric Liver Tumor Study HB89 and proposed using the tumor-node-metastasis system to compare treatment results in HB. The aims of the SIOPEL-1 study group in this article were to evaluate the accuracy of the PRETEXT staging system against surgery (gold standard), to study the interobserver agreement of PRETEXT, and to compare the predictive values of the different staging systems among patients who underwent delayed surgical resection of their tumor and subsequently followed the SIOPEL-1 protocol.7
PATIENTS AND METHODS
Patients were staged according to the PRETEXT system at diagnosis, during neoadjuvant chemotherapy, and before surgery. The original radiologic films were centrally reviewed by one radiologist (C.R.S.). For the comparison study between PRETEXT and pathology, the postchemotherapy PRETEXT taken before surgery was used.
Patients
Between January 1990 and February 1994, patients younger than 16 years old with HB were registered onto the SIOPEL-1 study. See Brown et al5 and Pritchard et al6 for a detailed description of study design, data collection, and definitions of event-free survival and overall survival. In short, all patients were treated preoperatively with PLADO after a biopsy had been taken according to the intent-to-treat principle. In case of unequivocal clinical findings, a biopsy was recommended but was mandatory in patients aged less than 6 months and more than 3 years because of the increased prevalence of other tumor types in these age groups. After four courses of PLADO, tumor resectability was assessed by imaging, and definitive surgery was performed if considered feasible. Tumor resection was then followed by two more courses of PLADO. Orthotopic liver transplantation was considered in patients with HB in all four liver sectors but completely confined to the liver, despite a positive response to adequate first-line chemotherapy. Results of orthotopic liver transplantation in SIOPEL-1 will be reported elsewhere. A total of 154 patients from 91 centers in 33 different countries entered onto the study, 128 of whom underwent resection of their primary liver tumor according to protocol guidelines.7 Of these 128 patients, 15 patients had no central review of their preoperative PRETEXT. Thus, this comparative study focuses on the subset of the remaining 113 patients who all had a centrally reviewed preoperative PRETEXT and who all underwent surgery.
Accuracy and Interobserver Agreement of PRETEXT
To evaluate the accuracy of the PRETEXT system, results from the PRETEXT system taken after chemotherapy and before surgery were compared with results from the pathology report of the operative specimen using agreement beyond change (weighted {kappa}). A weighted {kappa} is a {kappa} calculated with different weights that were given to the disagreements according to the magnitude of the discrepancy. For this purpose, the postoperative staging (gold standard) derived from the pathology report was retrospectively performed by doctors who were unaware of the PRETEXT results (J.M.S. and D.C.A.). The weighted {kappa} was also calculated to evaluate the interobserver agreement by comparing PRETEXT staging results obtained from the local center with those from the central review.
Other Staging Systems
The CCSG/POG staging system and the conventional tumor-node-metastasis system for (adult) liver carcinomas were retrospectively applied to the patients after the pathology report was available. The staging was performed in a blinded fashion, with the PRETEXT staging (at diagnosis) of that tumor being unknown (J.M.S. and D.C.A.). The CCSG/POG system distinguishes the following four disease stages: stage I, complete surgical resection; stage II, microscopic residual disease; stage III, macroscopic residual disease; and stage IV, metastatic spread.2,4 In the tumor-node-metastasis system, the T status comprises tumor size (≤ or > 2 cm), vascular invasion, lobe involvement, multifocality of tumor nodes, and extrahepatic growth; the N status records involvement of lymph nodes; and the M status records distant metastases.12,15 The different tumor-node-metastasis system stages are listed in Table 1. We are aware that, in contrast to the PRETEXT staging system, the CCSG/POG staging system and tumor-node-metastasis system are postoperative staging systems that are validated on the surgical results before any other therapeutic intervention and that now they are being applied to patients who have been pretreated with chemotherapy.
Comparison and Survival Analysis
The predictive values of the three different staging systems were compared using the Akaike information criterion (AIC) obtained from each of the Cox proportional hazards models. The AIC (–2ln [maximum likelihood] + 2 [number of fitted parameters]) is a descriptive statistic only and not a formal hypothesis test. It provides a useful measure for comparing different models.16 Subsequent overall survival curves of the different staging systems were obtained with the Kaplan-Meier method and compared within each system with the log-rank test.17,18 Overall survival was defined as the time interval between the date of diagnosis and the date of death (from any cause) or the date of last follow-up. The level of significance was considered P < .05. Statistical procedures were performed with the SAS statistical package version 8.02 (SAS Institute, Cary, NC).
RESULTS
Accuracy of PRETEXT: Staging Before Surgery Versus Pathologic Specimen
In 91 patients (81%), exact tumor location in the liver could be traced from the pathology report (ie, the gold standard; Fig 2) and could, thus, be compared with the preoperative PRETEXT staging system after central review. In 22 patients, the pathology report was not available. Fifty-one percent of the patients (46 of 91 patients) were staged correctly (ie, tumor found in the sectors predicted by the PRETEXT staging system). In 37% of the patients (34 of 91 patients), the PRETEXT staging was too high (overstaged), compared with the exact tumor localization, whereas in 12% of patients (11 of 91 patients), staging was too low (understaged). A positive resection margin was found in four of these 11 patients, and the other seven children underwent a complete surgical resection. Of the four patients with positive resection margins, none developed a local recurrence, which demonstrated the tumor negative status of the unresected liver segments in all patients.
The cross tabulation of the preoperative (centrally reviewed) and postoperative PRETEXT staging according to the pathology report (ie, the gold standard) is shown in Table 3. The weighted {kappa} value was 0.44 (95% CI, 0.26 to 0.62).
Interobserver Agreement: Original Versus Centrally Staged Preoperative PRETEXT
In 97 patients (86%), original PRETEXT preoperative staging could be compared with the centrally obtained staging. In 16 patients, one or both PRETEXT stagings were missing (Fig 2). There was an interobserver agreement in 79% of the patients (77 of 97 patients; Table 4). The weighted {kappa}, which was calculated by comparing the original and central PRETEXT staging preoperatively of the 97 patients, was 0.76 (95% CI, 0.64 to 0.88); on the basis of this 95% CI, we have a 95% certainty that the {kappa} lies between 0.64 and 0.88 (ie, good agreement). For the 77 patients (68%) in whom the pathologic data were also available, the weighted {kappa} was 0.71 (95% CI, 0.56 to 0.86).
Prognosis According to the Different Staging Systems
Survival analysis according to the different staging systems could be performed in 110 patients (97%). Follow-up data were missing for three patients. Tumor-node-metastasis system–based staging could only be performed in 98 patients (87%) because of missing data. The results according to the different staging systems are listed in Table 5. The 5-year overall survival rates according to the different preoperative PRETEXT groups after central review were 100% for group I, 95% for group II, 93% for group III, and 40% for group IV (Fig 3A). This system revealed a decreasing trend in overall survival related to the different subgroups that seemed to be highly significant (P = .0006).
The 5-year overall survival according to the CCSG/POG-based staging system is presented in Figure 3B. Patients with metastases (stage IV), who had complete surgical resection of their primary tumor (a select group of patients who underwent the exact SIOPEL-1 protocol and who were, therefore, included in this analysis), had the same survival rate (95%) as those patients with complete resection without metastases (stage I; 94%), microscopic residual disease (stage II; 88%), or macroscopic residual disease (stage III; 83%). These differences were not significant (P = .516). Note that there was a difference between the absolute figures of CCSG/POG-based staging and true CCSG/POG staging; in the original group of 154 patients, 31 patients who entered onto the trial with lung metastases (CCSG/POG stage IV) showed a 5-year event-free survival rate of 57% and 5-year overall survival rate of 28% (OS), respectively.19 Finally, the 98 patients who were staged according to the tumor-node-metastasis–based staging system (Fig 3C) showed a 5-year overall survival rate of 95% for stage II patients (stage I did not occur), 57% for stage III patients, and 93% for stage IV patients. Patients with a stage IV tumor who underwent the exact SIOPEL-1 protocol and, therefore, included in this analysis were a select group of patients. The tumor-node-metastasis–based staging system seemed to be highly significant in relation to overall survival as well (P = .0021).
Cox Proportional Hazard Ratios
For each of the three staging systems, a Cox proportional model was obtained, which entered the staging levels as independent variables considering the highest level as reference category. AIC was used for comparison of the three models. The best statistical fit was obtained with the PRETEXT staging system, which revealed the lowest AIC score (67.4), followed by the tumor-node-metastasis–based staging system (67.9) and the CCSG/POG-based staging system (75.3). The higher AIC score of the CCSG/POG-based staging system indicates the weakest statistical fit.
DISCUSSION
To assess the accuracy of the PRETEXT system, the preoperative PRETEXT staging was compared with the pathology report of the postoperative resection specimen (ie, the gold standard). Therefore this could only be applied to patients who underwent surgery, which is a selected subgroup of all HB patients. Our data showed that only 46 (51%) of 91 tumors were correctly staged, with a tendency to overstage the tumor (37%). For example, tumors were staged as group IV (ingrowth), whereas, in fact, they should have been staged as group III (compression). This phenomenon may be explained by the difficulty, if not impossibility, of distinguishing parenchymal compression of a tumor-free liver sector from tumor ingrowth into that sector. The weighted {kappa} value of 0.44 supports this assumption because it means that the accuracy of PRETEXT is moderate. Hopefully, future improvement of imaging quality and obligatory central review may improve this discrepancy, maybe even by using other imaging techniques, like magnetic resonance imaging. However, this assumption has to be studied prospectively.
However, the interobserver agreement of staging tumors according to the PRETEXT system is good as shown by the weighted {kappa} value of 0.76. This means that the system is reproducible, and one might assume that the system can easily be applied by different clinicians and that a relative uniformity of tumor staging exists. Although, one has to keep in mind that 63% of all patients (97 of 154 patients) who were eligible on the SIOPEL-1 study had their local PRETEXT staging compared with central staging, and 50% of all patients (77 of 154 patients) who were eligible on the SIOPEL-1 study were available to compare pathology with pretreatment staging (see flow chart in Fig 2).
Similarly, only 64% (98 of 154 patients) to 71% of the patients (110 of 154 patients) were available for comparing the three different staging systems in use for HB (Fig 2). Still, our data show that the predictive value in relation to survival of the PRETEXT system is at least as good as the well-known tumor-node-metastasis–based system. Both systems had a highly significant predictive value in relation to survival in the SIOPEL-1 study. In contrast, in this select group of patients, the CCSG/POG-based system seemed to be not significantly related to survival, probably because most patients had stage I disease. This finding was also confirmed by the statistical fit of the three staging systems in the Cox proportional hazards models, which showed a superiority for the PRETEXT system.
In conclusion, the results of the present data show that the accuracy of the PRETEXT system is moderate when the pre- and postoperative stages are being compared, probably as a result of the difficulty to distinguish parenchymal compression from true parenchymal ingrowth of the tumor; there was a tendency to overstage the patients; and the PRETEXT system demonstrated a good interobserver agreement, which means that this staging system is reproducible. The predictive value for survival of PRETEXT and of the tumor-node-metastasis–based system was highly significant in contrast to the predictive value of the CCSG/POG-based system. However, the PRETEXT system has an advantage because it offers the opportunity to monitor the effect of the neoadjuvant therapy used before surgery. Further research is necessary to evaluate the predictive value of PRETEXT in patients who do not receive surgical resection to evaluate the predictive value of this PRETEXT system and its use in monitoring the effects of preoperative chemotherapy, not only in patients who receive surgical resection, but in all patients. We recommend that all patients with HB included in the trials from the different study groups be staged both by their own preferred staging system as well as according to the PRETEXT system. This offers the opportunity to monitor preoperative treatment and to compare the results from the various trials in a more accurate way.
Appendix
Authors' Disclosures of Potential Conflicts of Interest
NOTES
Supported by the Swiss Cancer League, the Bernese Cancer League, and the Liver Tumour Parents Group, United Kingdom.
Both D.C.A. and J.M.S. contributed equally to this article.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Mann JR, Kasthuri N, Raafat F, et al: Malignant hepatic tumours in children: Incidence, clinical features and aetiology. Paediatr Perinat Epidemiol 4:276-289, 1990
2. Ortega JA, Krailo MD, Haas JE, et al: Effective treatment of unresectable or metastatic hepatoblastoma with cisplatin and continuous infusion doxorubicin chemotherapy: A report from the Children's Cancer Study Group. J Clin Oncol 9:2167-2176, 1991
3. von Schweinitz D, Byrd DJ, Hecker H, et al: Efficiency and toxicity of ifosfamide, cisplatin and doxorubicin in the treatment of childhood hepatoblastoma: Study Committee of the Cooperative Paediatric Liver Tumour Study HB89 of the German Society for Paediatric Oncology and Haematology. Eur J Cancer 33:1243-1249, 1997
4. Douglass EC, Reynolds M, Finegold M, et al: Cisplatin, vincristine, and fluorouracil therapy for hepatoblastoma: A Pediatric Oncology Group study. J Clin Oncol 11:96-99, 1993
5. Brown J, Perilongo G, Shafford E, et al: Pretreatment prognostic factors for children with hepatoblastoma: Results from the International Society of Paediatric Oncology (SIOP) study SIOPEL 1. Eur J Cancer 36:1418-1425, 2000
6. Pritchard J, Brown J, Shafford E, et al: Cisplatin, doxorubicin and delayed surgery for childhood hepatoblastoma a successful approach: Results of the first prospective study of the International Society of Pediatric Oncology. J Clin Oncol 18:3819-3828, 2000
7. Schnater JM, Aronson DC, Plaschkes J, et al: Surgical view of the treatment of patients with hepatoblastoma: Results from the first prospective trial of the International Society of Paediatric Oncology Liver Tumor Study group (SIOPEL-1). Cancer 94:1111-1120, 2002
8. Plaschkes J, Vos A, Czauderna P, et al: Into the year 2000 with the SIOPEL III: Cisplatin (CDDP) alone or in combination for standard and high risk hepatoblastoma, the SIOP XXX meeting. Med Pediatr Oncol 31:240, 1998 (abstr 0-181)
9. Von Schweinitz D: Identification of risk groups in hepatoblastoma: Another step in optimising therapy. Eur J Cancer 36:1343-1346, 2000
10. MacKinlay GA, Pritchard J: A common language for childhood liver tumours. Pediatr Surg Int 7:325-326, 1992
11. Vos A: Primary liver tumours in children. Eur J Surg Oncol 21:101-105, 1995
12. Hermanek P, Sobin J: UICC: TNM Classification Of Malignant Tumours (ed 4). Berlin, Germany, Springer, 1992
13. von Schweinitz D, Hecker H, Schmidt-von-Arndt G, et al: Prognostic factors and staging systems in childhood hepatoblastoma. Int J Cancer 74:593-599, 1997
14. Couinaud C: Lobes et segments hepatiques. Presse Med 62:709-712, 1954
15. Sobin LH, Wittekind C: UICC: TNM Classification of Malignant Tumours (ed 5). New York, NY, J Wiley & Sons, 1997
16. Akaike H: A new look at the statistical model identification. IEEE Trans Autom Contr 19:716-723, 1974
17. Kaplan EL, Meier P: Non-parametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
18. Kalbfleisch JD, Prentice RL: The Statistical Analysis of Failure Time Data. New York, NY, J Wiley, 1980
19. Perilongo G, Brown J, Shafford E, et al: Hepatoblastoma presenting with lung metastases. Cancer 89:1845-1853, 2000
20. Evans AE, Land VJ, Newton WA, et al: Combination chemotherapy (vincristine, Adriamycin, cyclophosphamide, and 5-fluorouracil) in the treatment of children with malignant hepatoma. Cancer 50:821-826, 1982
21. Sasaki F, Ohmuna N, Iwafuchi M: Outcome of hepatoblastoma in children treated with the JPLT protocol. Med Pediatr Oncol 32:308, 1998 (abstr P-186)
22. Ortega JA, Douglass EC, Feusner JH, et al: Randomized comparison of cisplatin/vincristine/fluorouracil and cisplatin/continuous infusion doxorubicin for treatment of pediatric hepatoblastoma: A report from the Children's Cancer Group and the Pediatric Oncology Group. J Clin Oncol 18:2665-2675, 2000
23. von Schweinitz D, Burger D, Mildenberger H: Is laparatomy the first step in treatment of childhood liver tumors The experience from the German Cooperative Pediatric Liver Tumor Study HB-89. Eur J Pediatr Surg 4:82-86, 1994
24. Morita K, Okabe I, Uchino J, et al: The proposed Japanese TNM classification of primary liver carcinoma in infants and children. Jpn J Clin Oncol 13:361-369, 1983(Danil C. Aronson, J. Marc)
Departments of Radiology and Clinical Epidemiology and Biostatistics, Academic Medical Center, Amsterdam, the Netherlands
Department of Pediatric Surgical Unit, Universitts-Kinderklinik, Inselspital, Bern, Switzerland
Department of Pediatrics, Universita di Padova, Padova, Italy
Northern and Yorkshire Clinical Trial and Research Unit, University of Leeds, Leeds
Department of Surgery, Royal Hospital for Sick Children, Edinburgh, United Kingdom
Department of Pediatric Surgery and Liver Transplantation, Universite Catholique de Louvain, Clinique Saint-Luc, Brussels, Belgium
Department of Pediatric Surgery, Medical University of Gdansk, Gdansk, Poland
ABSTRACT
PATIENTS AND METHODS: Hepatoblastoma (HB) patients younger than 16 years who underwent surgical resection (128 of 154 patients) were analyzed. The centrally reviewed preoperative staging was compared with postoperative pathology (accuracy) in 91 patients (81%), and the local center staging was compared with the central review (interobserver agreement) in 97 patients (86%), using the agreement beyond change method (weighted {kappa}). The predictive values of the three staging systems were compared in 110 patients (97%) using survival curves and Cox proportional hazard ratio estimates.
RESULTS: Preoperative PRETEXT staging compared with pathology was correct in 51%, overstaged in 37%, and understaged in 12% of patients (weighted {kappa} = 0.44; 95% CI, 0.26 to 0.62). The weighted {kappa} value of the interobserver agreement was 0.76 (95% CI, 0.64 to 0.88). The Children's Cancer Study Group/Pediatric Oncology Group–based staging system showed no predictive value for survival (P = .516), but the tumor-node-metastasis–based system and PRETEXT system showed good predictive values (P = .0021 and P = .0006, respectively). PRETEXT seemed to be superior in the statistical fit.
CONCLUSION: PRETEXT has moderate accuracy with a tendency to overstage patients, shows good interobserver agreement (reproducibility), shows superior predictive value for survival, offers the opportunity to monitor the effect of preoperative therapy, and can also be applied in patients who have not had operations. For comparability reasons, we recommend that all HB patients included in trials also be staged according to PRETEXT.
INTRODUCTION
In the SIOPEL-1 prospective trial, a preoperative surgical staging system, the pretreatment extent of disease (PRETEXT) system, which was based on the anatomy of the liver, was developed and adopted.10,11 The main difference from other well-known liver tumor staging systems, such as the tumor-node-metastasis system of the International Union Against Cancer and the system used by the Children's Cancer Study Group (CCSG) and the Pediatric Oncology Group (POG),2,4,12 is that the PRETEXT system was especially developed to compare the efficacy of various chemotherapeutic regimens in HB and to stage the tumor before surgical treatment, whereas the other two systems stage the tumor postoperatively. PRETEXT was used as a relatively objective but noninvasive method to assess tumor extent at diagnosis and subsequent chemotherapy response and to determine the optimal time and type of resection. Its ultimate goal was to ascertain preoperatively whether it would be possible to perform a radical resection.
In 1997, von Schweinitz et al13 investigated the predictive impact of the different staging systems (mentioned in the previous paragraph) in 72 patients treated in the German Pediatric Liver Tumor Study HB89 and proposed using the tumor-node-metastasis system to compare treatment results in HB. The aims of the SIOPEL-1 study group in this article were to evaluate the accuracy of the PRETEXT staging system against surgery (gold standard), to study the interobserver agreement of PRETEXT, and to compare the predictive values of the different staging systems among patients who underwent delayed surgical resection of their tumor and subsequently followed the SIOPEL-1 protocol.7
PATIENTS AND METHODS
Patients were staged according to the PRETEXT system at diagnosis, during neoadjuvant chemotherapy, and before surgery. The original radiologic films were centrally reviewed by one radiologist (C.R.S.). For the comparison study between PRETEXT and pathology, the postchemotherapy PRETEXT taken before surgery was used.
Patients
Between January 1990 and February 1994, patients younger than 16 years old with HB were registered onto the SIOPEL-1 study. See Brown et al5 and Pritchard et al6 for a detailed description of study design, data collection, and definitions of event-free survival and overall survival. In short, all patients were treated preoperatively with PLADO after a biopsy had been taken according to the intent-to-treat principle. In case of unequivocal clinical findings, a biopsy was recommended but was mandatory in patients aged less than 6 months and more than 3 years because of the increased prevalence of other tumor types in these age groups. After four courses of PLADO, tumor resectability was assessed by imaging, and definitive surgery was performed if considered feasible. Tumor resection was then followed by two more courses of PLADO. Orthotopic liver transplantation was considered in patients with HB in all four liver sectors but completely confined to the liver, despite a positive response to adequate first-line chemotherapy. Results of orthotopic liver transplantation in SIOPEL-1 will be reported elsewhere. A total of 154 patients from 91 centers in 33 different countries entered onto the study, 128 of whom underwent resection of their primary liver tumor according to protocol guidelines.7 Of these 128 patients, 15 patients had no central review of their preoperative PRETEXT. Thus, this comparative study focuses on the subset of the remaining 113 patients who all had a centrally reviewed preoperative PRETEXT and who all underwent surgery.
Accuracy and Interobserver Agreement of PRETEXT
To evaluate the accuracy of the PRETEXT system, results from the PRETEXT system taken after chemotherapy and before surgery were compared with results from the pathology report of the operative specimen using agreement beyond change (weighted {kappa}). A weighted {kappa} is a {kappa} calculated with different weights that were given to the disagreements according to the magnitude of the discrepancy. For this purpose, the postoperative staging (gold standard) derived from the pathology report was retrospectively performed by doctors who were unaware of the PRETEXT results (J.M.S. and D.C.A.). The weighted {kappa} was also calculated to evaluate the interobserver agreement by comparing PRETEXT staging results obtained from the local center with those from the central review.
Other Staging Systems
The CCSG/POG staging system and the conventional tumor-node-metastasis system for (adult) liver carcinomas were retrospectively applied to the patients after the pathology report was available. The staging was performed in a blinded fashion, with the PRETEXT staging (at diagnosis) of that tumor being unknown (J.M.S. and D.C.A.). The CCSG/POG system distinguishes the following four disease stages: stage I, complete surgical resection; stage II, microscopic residual disease; stage III, macroscopic residual disease; and stage IV, metastatic spread.2,4 In the tumor-node-metastasis system, the T status comprises tumor size (≤ or > 2 cm), vascular invasion, lobe involvement, multifocality of tumor nodes, and extrahepatic growth; the N status records involvement of lymph nodes; and the M status records distant metastases.12,15 The different tumor-node-metastasis system stages are listed in Table 1. We are aware that, in contrast to the PRETEXT staging system, the CCSG/POG staging system and tumor-node-metastasis system are postoperative staging systems that are validated on the surgical results before any other therapeutic intervention and that now they are being applied to patients who have been pretreated with chemotherapy.
Comparison and Survival Analysis
The predictive values of the three different staging systems were compared using the Akaike information criterion (AIC) obtained from each of the Cox proportional hazards models. The AIC (–2ln [maximum likelihood] + 2 [number of fitted parameters]) is a descriptive statistic only and not a formal hypothesis test. It provides a useful measure for comparing different models.16 Subsequent overall survival curves of the different staging systems were obtained with the Kaplan-Meier method and compared within each system with the log-rank test.17,18 Overall survival was defined as the time interval between the date of diagnosis and the date of death (from any cause) or the date of last follow-up. The level of significance was considered P < .05. Statistical procedures were performed with the SAS statistical package version 8.02 (SAS Institute, Cary, NC).
RESULTS
Accuracy of PRETEXT: Staging Before Surgery Versus Pathologic Specimen
In 91 patients (81%), exact tumor location in the liver could be traced from the pathology report (ie, the gold standard; Fig 2) and could, thus, be compared with the preoperative PRETEXT staging system after central review. In 22 patients, the pathology report was not available. Fifty-one percent of the patients (46 of 91 patients) were staged correctly (ie, tumor found in the sectors predicted by the PRETEXT staging system). In 37% of the patients (34 of 91 patients), the PRETEXT staging was too high (overstaged), compared with the exact tumor localization, whereas in 12% of patients (11 of 91 patients), staging was too low (understaged). A positive resection margin was found in four of these 11 patients, and the other seven children underwent a complete surgical resection. Of the four patients with positive resection margins, none developed a local recurrence, which demonstrated the tumor negative status of the unresected liver segments in all patients.
The cross tabulation of the preoperative (centrally reviewed) and postoperative PRETEXT staging according to the pathology report (ie, the gold standard) is shown in Table 3. The weighted {kappa} value was 0.44 (95% CI, 0.26 to 0.62).
Interobserver Agreement: Original Versus Centrally Staged Preoperative PRETEXT
In 97 patients (86%), original PRETEXT preoperative staging could be compared with the centrally obtained staging. In 16 patients, one or both PRETEXT stagings were missing (Fig 2). There was an interobserver agreement in 79% of the patients (77 of 97 patients; Table 4). The weighted {kappa}, which was calculated by comparing the original and central PRETEXT staging preoperatively of the 97 patients, was 0.76 (95% CI, 0.64 to 0.88); on the basis of this 95% CI, we have a 95% certainty that the {kappa} lies between 0.64 and 0.88 (ie, good agreement). For the 77 patients (68%) in whom the pathologic data were also available, the weighted {kappa} was 0.71 (95% CI, 0.56 to 0.86).
Prognosis According to the Different Staging Systems
Survival analysis according to the different staging systems could be performed in 110 patients (97%). Follow-up data were missing for three patients. Tumor-node-metastasis system–based staging could only be performed in 98 patients (87%) because of missing data. The results according to the different staging systems are listed in Table 5. The 5-year overall survival rates according to the different preoperative PRETEXT groups after central review were 100% for group I, 95% for group II, 93% for group III, and 40% for group IV (Fig 3A). This system revealed a decreasing trend in overall survival related to the different subgroups that seemed to be highly significant (P = .0006).
The 5-year overall survival according to the CCSG/POG-based staging system is presented in Figure 3B. Patients with metastases (stage IV), who had complete surgical resection of their primary tumor (a select group of patients who underwent the exact SIOPEL-1 protocol and who were, therefore, included in this analysis), had the same survival rate (95%) as those patients with complete resection without metastases (stage I; 94%), microscopic residual disease (stage II; 88%), or macroscopic residual disease (stage III; 83%). These differences were not significant (P = .516). Note that there was a difference between the absolute figures of CCSG/POG-based staging and true CCSG/POG staging; in the original group of 154 patients, 31 patients who entered onto the trial with lung metastases (CCSG/POG stage IV) showed a 5-year event-free survival rate of 57% and 5-year overall survival rate of 28% (OS), respectively.19 Finally, the 98 patients who were staged according to the tumor-node-metastasis–based staging system (Fig 3C) showed a 5-year overall survival rate of 95% for stage II patients (stage I did not occur), 57% for stage III patients, and 93% for stage IV patients. Patients with a stage IV tumor who underwent the exact SIOPEL-1 protocol and, therefore, included in this analysis were a select group of patients. The tumor-node-metastasis–based staging system seemed to be highly significant in relation to overall survival as well (P = .0021).
Cox Proportional Hazard Ratios
For each of the three staging systems, a Cox proportional model was obtained, which entered the staging levels as independent variables considering the highest level as reference category. AIC was used for comparison of the three models. The best statistical fit was obtained with the PRETEXT staging system, which revealed the lowest AIC score (67.4), followed by the tumor-node-metastasis–based staging system (67.9) and the CCSG/POG-based staging system (75.3). The higher AIC score of the CCSG/POG-based staging system indicates the weakest statistical fit.
DISCUSSION
To assess the accuracy of the PRETEXT system, the preoperative PRETEXT staging was compared with the pathology report of the postoperative resection specimen (ie, the gold standard). Therefore this could only be applied to patients who underwent surgery, which is a selected subgroup of all HB patients. Our data showed that only 46 (51%) of 91 tumors were correctly staged, with a tendency to overstage the tumor (37%). For example, tumors were staged as group IV (ingrowth), whereas, in fact, they should have been staged as group III (compression). This phenomenon may be explained by the difficulty, if not impossibility, of distinguishing parenchymal compression of a tumor-free liver sector from tumor ingrowth into that sector. The weighted {kappa} value of 0.44 supports this assumption because it means that the accuracy of PRETEXT is moderate. Hopefully, future improvement of imaging quality and obligatory central review may improve this discrepancy, maybe even by using other imaging techniques, like magnetic resonance imaging. However, this assumption has to be studied prospectively.
However, the interobserver agreement of staging tumors according to the PRETEXT system is good as shown by the weighted {kappa} value of 0.76. This means that the system is reproducible, and one might assume that the system can easily be applied by different clinicians and that a relative uniformity of tumor staging exists. Although, one has to keep in mind that 63% of all patients (97 of 154 patients) who were eligible on the SIOPEL-1 study had their local PRETEXT staging compared with central staging, and 50% of all patients (77 of 154 patients) who were eligible on the SIOPEL-1 study were available to compare pathology with pretreatment staging (see flow chart in Fig 2).
Similarly, only 64% (98 of 154 patients) to 71% of the patients (110 of 154 patients) were available for comparing the three different staging systems in use for HB (Fig 2). Still, our data show that the predictive value in relation to survival of the PRETEXT system is at least as good as the well-known tumor-node-metastasis–based system. Both systems had a highly significant predictive value in relation to survival in the SIOPEL-1 study. In contrast, in this select group of patients, the CCSG/POG-based system seemed to be not significantly related to survival, probably because most patients had stage I disease. This finding was also confirmed by the statistical fit of the three staging systems in the Cox proportional hazards models, which showed a superiority for the PRETEXT system.
In conclusion, the results of the present data show that the accuracy of the PRETEXT system is moderate when the pre- and postoperative stages are being compared, probably as a result of the difficulty to distinguish parenchymal compression from true parenchymal ingrowth of the tumor; there was a tendency to overstage the patients; and the PRETEXT system demonstrated a good interobserver agreement, which means that this staging system is reproducible. The predictive value for survival of PRETEXT and of the tumor-node-metastasis–based system was highly significant in contrast to the predictive value of the CCSG/POG-based system. However, the PRETEXT system has an advantage because it offers the opportunity to monitor the effect of the neoadjuvant therapy used before surgery. Further research is necessary to evaluate the predictive value of PRETEXT in patients who do not receive surgical resection to evaluate the predictive value of this PRETEXT system and its use in monitoring the effects of preoperative chemotherapy, not only in patients who receive surgical resection, but in all patients. We recommend that all patients with HB included in the trials from the different study groups be staged both by their own preferred staging system as well as according to the PRETEXT system. This offers the opportunity to monitor preoperative treatment and to compare the results from the various trials in a more accurate way.
Appendix
Authors' Disclosures of Potential Conflicts of Interest
NOTES
Supported by the Swiss Cancer League, the Bernese Cancer League, and the Liver Tumour Parents Group, United Kingdom.
Both D.C.A. and J.M.S. contributed equally to this article.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Mann JR, Kasthuri N, Raafat F, et al: Malignant hepatic tumours in children: Incidence, clinical features and aetiology. Paediatr Perinat Epidemiol 4:276-289, 1990
2. Ortega JA, Krailo MD, Haas JE, et al: Effective treatment of unresectable or metastatic hepatoblastoma with cisplatin and continuous infusion doxorubicin chemotherapy: A report from the Children's Cancer Study Group. J Clin Oncol 9:2167-2176, 1991
3. von Schweinitz D, Byrd DJ, Hecker H, et al: Efficiency and toxicity of ifosfamide, cisplatin and doxorubicin in the treatment of childhood hepatoblastoma: Study Committee of the Cooperative Paediatric Liver Tumour Study HB89 of the German Society for Paediatric Oncology and Haematology. Eur J Cancer 33:1243-1249, 1997
4. Douglass EC, Reynolds M, Finegold M, et al: Cisplatin, vincristine, and fluorouracil therapy for hepatoblastoma: A Pediatric Oncology Group study. J Clin Oncol 11:96-99, 1993
5. Brown J, Perilongo G, Shafford E, et al: Pretreatment prognostic factors for children with hepatoblastoma: Results from the International Society of Paediatric Oncology (SIOP) study SIOPEL 1. Eur J Cancer 36:1418-1425, 2000
6. Pritchard J, Brown J, Shafford E, et al: Cisplatin, doxorubicin and delayed surgery for childhood hepatoblastoma a successful approach: Results of the first prospective study of the International Society of Pediatric Oncology. J Clin Oncol 18:3819-3828, 2000
7. Schnater JM, Aronson DC, Plaschkes J, et al: Surgical view of the treatment of patients with hepatoblastoma: Results from the first prospective trial of the International Society of Paediatric Oncology Liver Tumor Study group (SIOPEL-1). Cancer 94:1111-1120, 2002
8. Plaschkes J, Vos A, Czauderna P, et al: Into the year 2000 with the SIOPEL III: Cisplatin (CDDP) alone or in combination for standard and high risk hepatoblastoma, the SIOP XXX meeting. Med Pediatr Oncol 31:240, 1998 (abstr 0-181)
9. Von Schweinitz D: Identification of risk groups in hepatoblastoma: Another step in optimising therapy. Eur J Cancer 36:1343-1346, 2000
10. MacKinlay GA, Pritchard J: A common language for childhood liver tumours. Pediatr Surg Int 7:325-326, 1992
11. Vos A: Primary liver tumours in children. Eur J Surg Oncol 21:101-105, 1995
12. Hermanek P, Sobin J: UICC: TNM Classification Of Malignant Tumours (ed 4). Berlin, Germany, Springer, 1992
13. von Schweinitz D, Hecker H, Schmidt-von-Arndt G, et al: Prognostic factors and staging systems in childhood hepatoblastoma. Int J Cancer 74:593-599, 1997
14. Couinaud C: Lobes et segments hepatiques. Presse Med 62:709-712, 1954
15. Sobin LH, Wittekind C: UICC: TNM Classification of Malignant Tumours (ed 5). New York, NY, J Wiley & Sons, 1997
16. Akaike H: A new look at the statistical model identification. IEEE Trans Autom Contr 19:716-723, 1974
17. Kaplan EL, Meier P: Non-parametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
18. Kalbfleisch JD, Prentice RL: The Statistical Analysis of Failure Time Data. New York, NY, J Wiley, 1980
19. Perilongo G, Brown J, Shafford E, et al: Hepatoblastoma presenting with lung metastases. Cancer 89:1845-1853, 2000
20. Evans AE, Land VJ, Newton WA, et al: Combination chemotherapy (vincristine, Adriamycin, cyclophosphamide, and 5-fluorouracil) in the treatment of children with malignant hepatoma. Cancer 50:821-826, 1982
21. Sasaki F, Ohmuna N, Iwafuchi M: Outcome of hepatoblastoma in children treated with the JPLT protocol. Med Pediatr Oncol 32:308, 1998 (abstr P-186)
22. Ortega JA, Douglass EC, Feusner JH, et al: Randomized comparison of cisplatin/vincristine/fluorouracil and cisplatin/continuous infusion doxorubicin for treatment of pediatric hepatoblastoma: A report from the Children's Cancer Group and the Pediatric Oncology Group. J Clin Oncol 18:2665-2675, 2000
23. von Schweinitz D, Burger D, Mildenberger H: Is laparatomy the first step in treatment of childhood liver tumors The experience from the German Cooperative Pediatric Liver Tumor Study HB-89. Eur J Pediatr Surg 4:82-86, 1994
24. Morita K, Okabe I, Uchino J, et al: The proposed Japanese TNM classification of primary liver carcinoma in infants and children. Jpn J Clin Oncol 13:361-369, 1983(Danil C. Aronson, J. Marc)