Clinical Model to Predict Survival in Chemonaive Patients With Advanced Non–Small-Cell Lung Cancer Treated With Third-Generation Chemotherap
http://www.100md.com
《临床肿瘤学》
the University of Wisconsin Medical School, Madison, WI
Dana-Farber Cancer Institute and Harvard School of Public Health, Boston, MA
Rush Presbyterian St Luke's Medical Center, Chicago, IL
Vanderbilt University, Nashville, TN. Trials in this study were conducted by the Eastern Cooperative Oncology Group
ABSTRACT
PATIENTS AND METHODS: Using data from two randomized, phase III Eastern Cooperative Oncology Group (ECOG) trials (E5592/E1594), we performed univariate and multivariate stepwise Cox regression analyses to identify survival prognostic factors. We used 75% of randomly sampled data to build a prediction model for survival, and the remaining 25% of data to validate the model.
RESULTS: From 1993 to 1999, 1,436 patients with stage IV or IIIB NSCLC with effusion were treated with platinum-based doublets (involving either paclitaxel, docetaxel, or gemcitabine). The response rate and median survival time were 20% and 8.2 months, respectively. One- and 2-year survivals were 33% and 11%, respectively. In multivariate analysis, six independent poor prognostic factors were identified: skin metastasis (hazard ratio [HR], 1.88), lower performance status (ECOG 1 or 2; HR, 1.46), loss of appetite (HR, 1.62), liver metastasis (HR, 1.32), ≥ four metastatic sites (HR, 1.20), and no prior surgery (HR, 1.16). A nomogram using six pretreatment prognostic factors was built to predict 1- and 2-year survival.
CONCLUSION: Six pretreatment factors can be used to predict survival in chemotherapy-naive NSCLC patients treated with standard chemotherapy. Using our prognostic nomogram, 1- and 2-year survival probability of NSCLC patients can be estimated before treatment. This prognostic model may help clinicians and patients in clinical decision making, as well as investigators in research planning.
INTRODUCTION
Several studies have attempted to identify clinical, laboratory, and molecular markers that may help clinicians and researchers distinguish subgroups of non–small-cell lung cancer (NSCLC) patients. Nevertheless, relatively few prognostic factors, such as extent of disease and performance status (PS), have been widely accepted as useful prognostic markers. Furthermore, there is no simple and reliable way to estimate the survival of individual patients undergoing chemotherapy. Therefore, we performed this retrospective analysis to identify clinical factors that influence the outcome of advanced NSCLC patients treated with standard first-line chemotherapy, and to build a model that can be used in daily practice to predict long-term survival in this patient population. Such a practical model may help oncologists and their patients make treatment decisions, and may assist investigators in designing clinical trials.
PATIENTS AND METHODS
Before analysis, we decided to exclude the etoposide arm in E5592 from our patient database because of its inferior survival. Also excluded were 75 patients in E5592 who had stage IIIB disease without effusion. The purpose of the exclusion is to have a homogenous population of patients representing those currently treated in daily practice in the community. All patients in our analysis had metastatic or IIIB disease with malignant effusion and were treated with newer chemotherapy doublets involving cisplatin or carboplatin in combination with a third-generation agent (paclitaxel, docetaxel, or gemcitabine).
Statistical Methods
To identify potential prognostic factors, we analyzed 27 pretreatment clinical variables recorded in the ECOG database, including patient demographics (age, sex, race), history of prior radiation or surgery (therapeutic and/or diagnostic) for lung cancer, significant underlying conditions (chronic lung, heart diseases), disease stage (IIIB v IV), PS, systemic symptoms (loss of appetite, weight loss, fever), tumor characteristics (histology, measurability), and metastatic features (number of metastatic sites, metastatic organs), as presented in Table 2. Univariate Cox regression analysis was used first to assess the association between each variable and survival, followed by multivariate stepwise Cox regression analysis for variable selection (with entry cutoff level of 0.2 and stay cutoff 0.1). The stepwise variable selection was performed on 75% randomly sampled data from which a survival prediction model was built. This model was validated using the remaining 25% of the data. Based on the prediction model with identified prognostic factors, a nomogram was drawn for prediction of 1- and 2-year survival. The analysis was performed using the SAS program (SAS Institute, Cary, NC).
RESULTS
Pretreatment Prognostic Factors
Of 27 pretreatment clinical factors, 15 were associated with poor prognosis in univariate Cox analysis (P ≤ .05), including being male, lower PS (ECOG 1 or 2), loss of appetite, loss of weight, having fever, having no prior lung surgery, history of receiving radiation therapy, and metastasis-related factors (having ≥ 4 metastatic sites, having metastasis in the mediastinum, contralateral lung, bone, liver, brain, subcutaneous tissue, or other organs; Table 3). Of all the factors analyzed, subcutaneous metastasis was associated with the highest hazard ratio (HR) of 2.06. Twelve pretreatment factors were found not to be prognostically significant, including age (≥ 70 v < 70), stage (IIIB with pleural effusion versus IV), tumor histology (squamous v adenocarcinoma v large-cell type), and comorbid diseases.
Of 15 negative variables identified from univariate analysis, only six remained independent poor prognostic factors after stepwise Cox regression in multivariate analysis (all with P < .05) as presented in Table 4. These included subcutaneous metastasis, lower PS (ECOG 1 or 2), loss of appetite, liver metastasis, ≥ four metastatic sites, and no prior surgery. Again, subcutaneous involvement was associated with the highest HR (1.88).
Nomogram
A nomogram to predict survival of chemotherapy-naive patients with advanced NSCLC undergoing standard chemotherapy, shown in Figure 1, was built based on the six independent pretreatment prognostic markers. In this Cox model (Table 4), each negative marker was given a score implying survival prognosis: subcutaneous metastasis = 66; decrease in PS (ECOG 1 or 2) = 43; loss of appetite = 38; liver metastasis = 35; having ≥ 4 metastatic sites = 19; having no previous lung surgery = 15. A higher score implies a poorer prognosis. A particular patient's survival can be estimated by determining the negative prognostic factors in that patient, adding up all the scores corresponding to those factors, locating the total score on the total point scale, and finally, drawing a straight line down to determine the estimated 1- and 2-year survivals on the survival scales. Figure 2 illustrates an example of a patient with PS 1 (score = 43), appetite loss (score = 38), liver metastasis (score = 35), and no previous lung surgery (score = 15). The total score of 131 (43 +38 + 35 + 15) corresponds to an estimated 1- and 2-year survival of 12% and 1.1%, respectively.
Validation of the Nomogram
Our nomogram was built from 75% of the data from the database. The remaining 25% of the data was used to validate the nomogram. Patients in this validation set were divided into quartile groups (low-risk, low-intermediate–risk, high-intermediate–risk, and high-risk) according to their prognostic scores (Table 5 and Fig 3). As shown in Figure 3, the survival predicted by our prognostic model for each quartile group (x-axis) was compared with the observed (actual) survival (y-axis). The dotted line illustrates the ideal scenario in which the predicted survival perfectly matches the observed survival. Overall, the 1- and 2-year predicted survival lines follow the "ideal line" and remain within the 95% CI of the observed survival, suggesting that there is a relatively good agreement between predicted and actual survivals.
DISCUSSION
Three cooperative groups, two from the United States and one from Europe, have analyzed their data to identify important prognostic factors (Table 6). The first analysis, published in the mid 1980s by Finkelstein et al, was based on data from two ECOG randomized phase III trials (EST 2575 and EST 1581).8 Eight hundred ninety-three metastatic NSCLC patients were treated on seven chemotherapy arms involving older drugs (eg, cyclophosphamide, bleomycin, doxorubicin) in combination with or without cisplatin. Because few patients (4%) survived beyond two years, the main analytic end point was to identify pretreatment clinical factors to distinguish patients who survived more than 1 year ("long-term survivors") from those who did not. After logistic regression multivariate analysis of 36 on-study factors, eight were associated with favorable outcome (in order of importance): PS 0 (P < .0001), no bone metastasis (P < .0001), female sex (P = .0005), no weight loss (P = .001), no subcutaneous metastasis (P = .006), non–large-cell histology (P = .011), no prior symptom of shoulder or arm pain (P = .029), and no liver metastasis (P = .046). Further analysis performed among 1-year survivors identified three factors predicting long-term survival: being female (P = .005), no subcutaneous metastasis (P = .016), and PS 0 (P = .04).
In the second report, Southwest Oncology Group (SWOG) investigators analyzed data from 2,531 advanced NSCLC patients enrolled on 14 chemotherapy trials (five phase III and nine phase II studies) from 1974 to 1988.9 Most patients were chemotherapy-naive, except those in two phase II trials for recurrent disease. A significant number (40%) of participants had PS ≥ 2. Chemotherapy regimens varied from cisplatin or nonplatin monotherapy (eg, chlorozotocin, fludarabine, menogaril) to older platin or nonplatin combinations (eg, adriamycin/cisplatin, adriamycin/ifosfamide, mechlorethamine/adriamycin/lomustine). In a multivariate analysis, four pretreatment clinical factors were found to predict better survival: PS 0 to 1 (P < .00005), cisplatin-based therapy (P < .00005), being female (P < .00005), and advanced age ≥ 70 years (P = .02). In a subset analysis based on 362 patients with better PS (0 to 1) and more complete data on metastasis and laboratory results, hemoglobin ≥ 11 (P = .001), normal lactic dehydrogenase (P = .002), normal calcium (P = .007), single metastasis (P = .02), and cisplatin-based therapy (P = .05) were favorable predictive factors.
The European Lung Cancer Working Party (ELCWP) analyzed data from 1,052 unresectable NSCLC patients from three phase III and four phase II trials conducted between 1980 and 1991.10 The patient population was heterogenous, including those with limited (stage I to III) or advanced disease, and those with or without prior chemotherapy. Some patients received chemoradiotherapy. The chemotherapy involved cisplatin with or without second-generation agents (etoposide being the most common). Of 23 pretreatment clinical and laboratory variables analyzed, eight negative factors were identified in multivariate analysis: metastatic disease (P < .0001), Karnofsky PS ≤ 70 (P = .0003), leukocytosis (P = .0003), skin metastasis (P = .008), increased calcium level (P = .007), abnormal neutrophil count (P = .02), age greater than 60 years (P = .02), and being male (P = .03). Of note, skin metastasis had the highest relative risk (2.80).
Compared with the three analyses described above, our analysis was based on more recent clinical studies involving newer, third-generation regimens. As in the case of the previous ECOG study reported by Finkelstein, the database for our analysis was built entirely from two large randomized phase III trials, providing a more homogenous patient population. All patients in our analysis had stage IV or IIIB with malignant effusion and had received no prior chemotherapy. Almost all had a PS of 0 or 1. All were treated with modern regimens consisting of cisplatin or carboplatin in combination with a third-generation agent, which are presently considered the standard of care for advanced disease. The survival outcome was similar to results from other large contemporary phase III studies. However, as with the other three analyses, this is a retrospective study with inherent limitations. For example, laboratory data were not included in our analysis because only absolute values were entered in the database, and there were concerns about differences in reference values between institutions.
Of 27 pretreatment clinical factors analyzed in univariate and multivariate analyses, six factors stood out as independent survival prognostic markers. Interestingly, subcutaneous metastasis was associated with the worse prognosis, with the highest HR at 1.88. The prognostic importance of lung cancer spreading to subcutaneous tissue or skin was also found in the previous ECOG and ELCWP analyses, as well as in another study from Great Britain.11 Skin or subcutaneous metastasis from lung cancer is uncommon and is rarely included in prognostic analyses. In a retrospective study involving 200 patients with skin metastasis from solid tumors, lung (18%) was among the most common sites of tumor origin, together with breast (32%) and melanoma (16%).12 However, with a median survival of 2.9 months, survival for patients with lung cancer was much shorter than those with breast cancer (13.8 months) or melanoma (13.5 months). The ECOG study by Finkelstein et al8 showed that subcutaneous metastasis is one of three important clinical features discriminating long-term survivors among those who lived beyond 1 year; the other two being PS and sex.
Other metastasis-related prognostic factors in our multivariate analysis were liver involvement and number of metastatic sites (≥ four), with HRs of 1.32 and 1.20, respectively. While hepatic metastases was also an important factor in the prior ECOG study, the number of metastatic lesions (≥ two) predicted a poor outcome in PS 0 to 1 patients in the SWOG analysis. Brain is among the most common metastatic sites in NSCLC patients. Those with brain metastasis have been generally thought to have poor outcome, and traditionally have been excluded from chemotherapy trials. In our study, brain metastasis was associated with an HR of 1.21 in univariate analysis. However, its statistical significance disappeared in multivariate analysis. Because NSCLC patients with brain metastasis are a heterogenous group, with prognosis possibly depending on the extent of intra- and extracranial disease, treatment for these patients should be individualized.
Poor performance status has been widely accepted as one of the most important negative prognostic factors in all cancer patients. The importance of this marker was validated in all four analyses, in which it was among the most significant factors in predicting survival of lung cancer patients. In the SWOG report, patients with PS 0 to 1 did significantly better than those with PS ≥ 2. In the two ECOG analyses, which included primarily patients with PS 0 to 1, those with excellent status (PS 0) lived longer than those with lower PS (1 to 2). In fact, PS 0 was the most important factor in the first ECOG analysis. In the Veterans Administration Lung Group analysis published in 1980, Stanley analyzed survival by Karnofsky score and found that a higher score (by 10% to 15% increments) correlated with increase in median survival.4 Thus, it seems that a better pretreatment PS is associated with a longer survival (PS 0 > PS 1 > PS 2 or higher), suggesting that participants in clinical trials should be stratified further based on their PS (PS 0 v PS 1 v PS 2).
Unlike PS, the prognostic role of weight loss has been somewhat controversial. Some studies, including the previous ECOG analysis, found significant weight loss being among the poor prognostic factors,4,8,13 but others, including the SWOG and ELCWP analyses, did not.9,10,14,15 In our study, we looked at both weight loss and reduced appetite as separate variables. Although the significance of both weight loss and reduced appetite was demonstrated in univariate analysis, only the latter remained an independent negative marker in multivariate analysis.
An interesting finding from our study is that a history of prior surgery for lung cancer was a positive prognostic factor. However, this variable in our database included both diagnostic and therapeutic surgery without speculation of procedures that were performed. It seems plausible that an individual who underwent resection of a primary lung lesion and then recurred several years later with multiple metastases might have a disease with a completely different biology than someone who presented "de novo" with widespread metastases who therefore never had a chance to undergo a surgical procedure. Nevertheless, without complete data for further analysis, we do not have a definitive explanation for this observation.
Of several clinical characteristics analyzed in our research, age, sex, and tumor histology were not associated with survival outcome. In the past, there had been concern of increase in mortality in elderly patients treated with chemotherapy. Although a few studies such as the ELCWP report found that advanced age was a negative factor, most, including the two ECOG analyses, have concluded that older patients have the same outcome as their younger counterparts.16,17 Of note, age ≥ 70 years was actually considered a favorable factor in the SWOG analysis.9 These observations suggest that age should not be the reason to exclude older but fit patients from therapy, particularly with the availability of newer chemotherapy drugs and better supportive measures. Several studies have found that gender is an important prognostic factor, with women faring better than men.6,8-10,14,15,18 However, the influence of gender in survival was not seen in our patient population. The possible role of histology subtypes of NSCLC in determining clinical outcome, such as the favorable impact of squamous cell type in early stage6 or the negative effect of large cell histology in advanced disease,8 has been raised. However, in our study, we found no difference in survival between tumor histology subtypes. The 1997 American Society of Clinical Oncology guideline, updated in 2003, also suggested that NSCLC histology is not an important prognostic factor in patients with advanced, unresectable disease.7,19
In the prior ECOG analysis, Finkelstein created a discriminant function model to predict the probability of surviving longer than 1 year.8 This model requires a discriminant table and different steps to calculate
where S is the calculated prognostic score. A P value greater than .18 predicts that a particular patient will survive longer than 1 year, while P < .18 gives reverse prediction. SWOG and ELCWP investigators used recursive partitioning and amalgamation method to divide patients into subgroups based on prognostic factors identified in their studies. For example, in the SWOG study, three prognostic subgroups were identified: (1) PS 0 to 1, hemoglobin ≥ 11, and age ≥ 47 years (1- and 2-year survivals of 27% and 8%, respectively), (2) PS 2 to 4 and lactate dehydrogenase higher than normal (1- and 2-year survivals of 6% and 1%, respectively), and (3) others, including PS 2 to 4 and normal lactic dehydrogenase (1- and 2-year survival of 16% and 3%).9 We used the Cox regression method to build a prognostic nomogram based on six negative predictive factors. Nomogram models have been used in prostate cancer to predict survival and recurrence probability.20-22 Using our model, 1- and 2-year survival of NSCLC chemotherapy-naive patients undergoing first-line chemotherapy can be easily estimated. The use of this nomogram is limited to patients who are chemotherapy-naive, have good PS, and receive standard third-generation doublets.
In summary, we have identified six negative survival prognostic factors in advanced NSCLC patients undergoing first-line standard chemotherapy: subcutaneous metastasis, decreased PS (ECOG 1 or 2), loss of appetite, liver metastasis, ≥ four metastatic sites, and no previous lung surgery. Based on a large patient database, we have built a unique nomogram using pretreatment clinical factors to predict survival probability of chemotherapy-naive patients treated with third-generation platinum-based doublets. This prognostic model could help clinicians and patients in clinical decision making and treatment tailoring based on the estimated prognosis. In addition, the nomogram and survival markers could be applied in clinical research to stratify patients, avoid biases, as well as plan appropriate studies targeting different subgroups of patients. Using our model instead of historic control data when analyzing survival result of phase II studies, promising first-line regimens could be selected for phase III trials. Although the nomogram was already validated with internal data as described previously, we plan to further validate it using data from other studies.
Authors' Disclosures of Potential Conflicts of Interest
Acknowledgment
We thank the investigators and staff of the Eastern Cooperative Oncology Group for their participation in the trials listed in this study.
NOTES
Supported in part by grant K12-CA 87718 (T.H.).
Presented at the 39th American Society of Clinical Oncology Annual Meeting, Chicago, IL, May 31-June 3, 2003.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Schiller JH, Harrington D, Belani CP, et al: Comparison of four chemotherapy regimens for advanced non-small-cell lung cancer. N Engl J Med 346:92-98, 2002
2. Kelly K, Crowley J, Bunn PA, et al: Randomized phase III trial of paclitaxel plus carboplatin versus vinorelbine plus cisplatin in the treatment of patients with advanced non-small-cell lung cancer: A Southwest Oncology Group trial. J Clin Oncol 19:3210-3218, 2001
3. Bonomi P, Kim K, Fairclough D, et al: Comparison of survival and quality of life in advanced non-small cell lung cancer patients treated with two dose levels of paclitaxel combined with cisplatin versus etoposide with cisplatin: Results of an Eastern Cooperative Oncology Group trial. J Clin Oncol 18:623-631, 2000
4. Stanley KE: Prognostic factors for survival in patients with inoperable lung cancer. J Natl Cancer Inst 65:25-32, 1980
5. Brundage MD, Davies D, Mackillop WJ: Prognostic factors in non-small cell lung cancer: A decade of progress. Chest 122:1037-1057, 2002
6. Feld R, Abratt R, Graziano S, et al: Pretreatment minimal staging and prognostic factors for non-small cell lung cancer. Lung Cancer 17:S3-10, 1997
7. Pfister DG, Johnson DH, Azzoli CG, et al: American Society of Clinical Oncology treatment of unresectable non-small-cell lung cancer guideline: Update 2003. J Clin Oncol 22:330-353, 2004
8. Finkelstein DM, Ettinger DS, Ruckdeschel JC: Long-term survivors in metastatic non-small-cell lung cancer: An Eastern Cooperative Oncology Group Study. J Clin Oncol 4:702-709, 1986
9. Albain KS, Crowley JJ, LeBlanc M, et al: Survival determinants in extensive-stage non-small-cell lung cancer: The Southwest Oncology Group experience. J Clin Oncol 9:1618-1626, 1991
10. Paesmans M, Sculier JP, Libert P, et al: Prognostic factors for survival in advanced non-small-cell lung cancer: Univariate and multivariate analyses including recursive partitioning and amalgamation algorithms in 1,052 patients—The European Lung Cancer Working Party. J Clin Oncol 13:1221-1230, 1995
11. Hickish TF, Smith IE, O'Brien ME, et al: Clinical benefit from palliative chemotherapy in non-small-cell lung cancer extends to the elderly and those with poor prognostic factors. Br J Cancer 78:28-33, 1998
12. Schoenlaub P, Sarraux A, Grosshans E, et al: [Survival after cutaneous metastasis: A study of 200 cases]. Ann Dermatol Venereol 128:1310-1315, 2001
13. Fukuoka M, Masuda N, Furuse K, et al: A randomized trial in inoperable non-small-cell lung cancer: Vindesine and cisplatin versus mitomycin, vindesine, and cisplatin versus etoposide and cisplatin alternating with vindesine and mitomycin. J Clin Oncol 9:606-613, 1991
14. O'Connell JP, Kris MG, Gralla RJ, et al: Frequency and prognostic importance of pretreatment clinical characteristics in patients with advanced non-small-cell lung cancer treated with combination chemotherapy. J Clin Oncol 4:1604-1614, 1986
15. Shinkai T, Eguchi K, Sasaki Y, et al: A prognostic-factor risk index in advanced non-small-cell lung cancer treated with cisplatin-containing combination chemotherapy. Cancer Chemother Pharmacol 30:1-6, 1992
16. Langer CJ, Manola J, Bernardo P, et al: Cisplatin-based therapy for elderly patients with advanced non-small-cell lung cancer: Implications of Eastern Cooperative Oncology Group 5592, a randomized trial. J Natl Cancer Inst 94:173-181, 2002
17. Lilenbaum R, Herndon J, List M: Single-agent (SA) versus combination chemotherapy (CC) in advanced non-small cell lung cancer (NSCLC): A CALGB randomized trial of efficacy, quality of life (QOL), and cost-effectiveness. Proc Am Soc Clin Onc 21:1a, 2002 (abstr 2)
18. Radzikowska E, Glaz P, Roszkowski K: Lung cancer in women: Age, smoking, histology, performance status, stage, initial treatment and survival—Population-based study of 20 561 cases. Ann Oncol 13:1087-1093, 2002
19. ASCO: Clinical practice guidelines for the treatment of unresectable non- small-cell lung cancer. Adopted on May 16, 1997 by the American Society of Clinical Oncology. J Clin Oncol 15:2996-3018, 1997
20. Smaletz O, Scher HI, Small EJ, et al: Nomogram for overall survival of patients with progressive metastatic prostate cancer after castration. J Clin Oncol 20:3972-3982, 2002
21. Kattan MW, Wheeler TM, Scardino PT: Postoperative nomogram for disease recurrence after radical prostatectomy for prostate cancer. J Clin Oncol 17:1499-1507, 1999
22. Kattan MW, Zelefsky MJ, Kupelian PA, et al: Pretreatment nomogram for predicting the outcome of three-dimensional conformal radiotherapy in prostate cancer. J Clin Oncol 18:3352-3359, 2000(Tien Hoang, Ronghui Xu, J)
Dana-Farber Cancer Institute and Harvard School of Public Health, Boston, MA
Rush Presbyterian St Luke's Medical Center, Chicago, IL
Vanderbilt University, Nashville, TN. Trials in this study were conducted by the Eastern Cooperative Oncology Group
ABSTRACT
PATIENTS AND METHODS: Using data from two randomized, phase III Eastern Cooperative Oncology Group (ECOG) trials (E5592/E1594), we performed univariate and multivariate stepwise Cox regression analyses to identify survival prognostic factors. We used 75% of randomly sampled data to build a prediction model for survival, and the remaining 25% of data to validate the model.
RESULTS: From 1993 to 1999, 1,436 patients with stage IV or IIIB NSCLC with effusion were treated with platinum-based doublets (involving either paclitaxel, docetaxel, or gemcitabine). The response rate and median survival time were 20% and 8.2 months, respectively. One- and 2-year survivals were 33% and 11%, respectively. In multivariate analysis, six independent poor prognostic factors were identified: skin metastasis (hazard ratio [HR], 1.88), lower performance status (ECOG 1 or 2; HR, 1.46), loss of appetite (HR, 1.62), liver metastasis (HR, 1.32), ≥ four metastatic sites (HR, 1.20), and no prior surgery (HR, 1.16). A nomogram using six pretreatment prognostic factors was built to predict 1- and 2-year survival.
CONCLUSION: Six pretreatment factors can be used to predict survival in chemotherapy-naive NSCLC patients treated with standard chemotherapy. Using our prognostic nomogram, 1- and 2-year survival probability of NSCLC patients can be estimated before treatment. This prognostic model may help clinicians and patients in clinical decision making, as well as investigators in research planning.
INTRODUCTION
Several studies have attempted to identify clinical, laboratory, and molecular markers that may help clinicians and researchers distinguish subgroups of non–small-cell lung cancer (NSCLC) patients. Nevertheless, relatively few prognostic factors, such as extent of disease and performance status (PS), have been widely accepted as useful prognostic markers. Furthermore, there is no simple and reliable way to estimate the survival of individual patients undergoing chemotherapy. Therefore, we performed this retrospective analysis to identify clinical factors that influence the outcome of advanced NSCLC patients treated with standard first-line chemotherapy, and to build a model that can be used in daily practice to predict long-term survival in this patient population. Such a practical model may help oncologists and their patients make treatment decisions, and may assist investigators in designing clinical trials.
PATIENTS AND METHODS
Before analysis, we decided to exclude the etoposide arm in E5592 from our patient database because of its inferior survival. Also excluded were 75 patients in E5592 who had stage IIIB disease without effusion. The purpose of the exclusion is to have a homogenous population of patients representing those currently treated in daily practice in the community. All patients in our analysis had metastatic or IIIB disease with malignant effusion and were treated with newer chemotherapy doublets involving cisplatin or carboplatin in combination with a third-generation agent (paclitaxel, docetaxel, or gemcitabine).
Statistical Methods
To identify potential prognostic factors, we analyzed 27 pretreatment clinical variables recorded in the ECOG database, including patient demographics (age, sex, race), history of prior radiation or surgery (therapeutic and/or diagnostic) for lung cancer, significant underlying conditions (chronic lung, heart diseases), disease stage (IIIB v IV), PS, systemic symptoms (loss of appetite, weight loss, fever), tumor characteristics (histology, measurability), and metastatic features (number of metastatic sites, metastatic organs), as presented in Table 2. Univariate Cox regression analysis was used first to assess the association between each variable and survival, followed by multivariate stepwise Cox regression analysis for variable selection (with entry cutoff level of 0.2 and stay cutoff 0.1). The stepwise variable selection was performed on 75% randomly sampled data from which a survival prediction model was built. This model was validated using the remaining 25% of the data. Based on the prediction model with identified prognostic factors, a nomogram was drawn for prediction of 1- and 2-year survival. The analysis was performed using the SAS program (SAS Institute, Cary, NC).
RESULTS
Pretreatment Prognostic Factors
Of 27 pretreatment clinical factors, 15 were associated with poor prognosis in univariate Cox analysis (P ≤ .05), including being male, lower PS (ECOG 1 or 2), loss of appetite, loss of weight, having fever, having no prior lung surgery, history of receiving radiation therapy, and metastasis-related factors (having ≥ 4 metastatic sites, having metastasis in the mediastinum, contralateral lung, bone, liver, brain, subcutaneous tissue, or other organs; Table 3). Of all the factors analyzed, subcutaneous metastasis was associated with the highest hazard ratio (HR) of 2.06. Twelve pretreatment factors were found not to be prognostically significant, including age (≥ 70 v < 70), stage (IIIB with pleural effusion versus IV), tumor histology (squamous v adenocarcinoma v large-cell type), and comorbid diseases.
Of 15 negative variables identified from univariate analysis, only six remained independent poor prognostic factors after stepwise Cox regression in multivariate analysis (all with P < .05) as presented in Table 4. These included subcutaneous metastasis, lower PS (ECOG 1 or 2), loss of appetite, liver metastasis, ≥ four metastatic sites, and no prior surgery. Again, subcutaneous involvement was associated with the highest HR (1.88).
Nomogram
A nomogram to predict survival of chemotherapy-naive patients with advanced NSCLC undergoing standard chemotherapy, shown in Figure 1, was built based on the six independent pretreatment prognostic markers. In this Cox model (Table 4), each negative marker was given a score implying survival prognosis: subcutaneous metastasis = 66; decrease in PS (ECOG 1 or 2) = 43; loss of appetite = 38; liver metastasis = 35; having ≥ 4 metastatic sites = 19; having no previous lung surgery = 15. A higher score implies a poorer prognosis. A particular patient's survival can be estimated by determining the negative prognostic factors in that patient, adding up all the scores corresponding to those factors, locating the total score on the total point scale, and finally, drawing a straight line down to determine the estimated 1- and 2-year survivals on the survival scales. Figure 2 illustrates an example of a patient with PS 1 (score = 43), appetite loss (score = 38), liver metastasis (score = 35), and no previous lung surgery (score = 15). The total score of 131 (43 +38 + 35 + 15) corresponds to an estimated 1- and 2-year survival of 12% and 1.1%, respectively.
Validation of the Nomogram
Our nomogram was built from 75% of the data from the database. The remaining 25% of the data was used to validate the nomogram. Patients in this validation set were divided into quartile groups (low-risk, low-intermediate–risk, high-intermediate–risk, and high-risk) according to their prognostic scores (Table 5 and Fig 3). As shown in Figure 3, the survival predicted by our prognostic model for each quartile group (x-axis) was compared with the observed (actual) survival (y-axis). The dotted line illustrates the ideal scenario in which the predicted survival perfectly matches the observed survival. Overall, the 1- and 2-year predicted survival lines follow the "ideal line" and remain within the 95% CI of the observed survival, suggesting that there is a relatively good agreement between predicted and actual survivals.
DISCUSSION
Three cooperative groups, two from the United States and one from Europe, have analyzed their data to identify important prognostic factors (Table 6). The first analysis, published in the mid 1980s by Finkelstein et al, was based on data from two ECOG randomized phase III trials (EST 2575 and EST 1581).8 Eight hundred ninety-three metastatic NSCLC patients were treated on seven chemotherapy arms involving older drugs (eg, cyclophosphamide, bleomycin, doxorubicin) in combination with or without cisplatin. Because few patients (4%) survived beyond two years, the main analytic end point was to identify pretreatment clinical factors to distinguish patients who survived more than 1 year ("long-term survivors") from those who did not. After logistic regression multivariate analysis of 36 on-study factors, eight were associated with favorable outcome (in order of importance): PS 0 (P < .0001), no bone metastasis (P < .0001), female sex (P = .0005), no weight loss (P = .001), no subcutaneous metastasis (P = .006), non–large-cell histology (P = .011), no prior symptom of shoulder or arm pain (P = .029), and no liver metastasis (P = .046). Further analysis performed among 1-year survivors identified three factors predicting long-term survival: being female (P = .005), no subcutaneous metastasis (P = .016), and PS 0 (P = .04).
In the second report, Southwest Oncology Group (SWOG) investigators analyzed data from 2,531 advanced NSCLC patients enrolled on 14 chemotherapy trials (five phase III and nine phase II studies) from 1974 to 1988.9 Most patients were chemotherapy-naive, except those in two phase II trials for recurrent disease. A significant number (40%) of participants had PS ≥ 2. Chemotherapy regimens varied from cisplatin or nonplatin monotherapy (eg, chlorozotocin, fludarabine, menogaril) to older platin or nonplatin combinations (eg, adriamycin/cisplatin, adriamycin/ifosfamide, mechlorethamine/adriamycin/lomustine). In a multivariate analysis, four pretreatment clinical factors were found to predict better survival: PS 0 to 1 (P < .00005), cisplatin-based therapy (P < .00005), being female (P < .00005), and advanced age ≥ 70 years (P = .02). In a subset analysis based on 362 patients with better PS (0 to 1) and more complete data on metastasis and laboratory results, hemoglobin ≥ 11 (P = .001), normal lactic dehydrogenase (P = .002), normal calcium (P = .007), single metastasis (P = .02), and cisplatin-based therapy (P = .05) were favorable predictive factors.
The European Lung Cancer Working Party (ELCWP) analyzed data from 1,052 unresectable NSCLC patients from three phase III and four phase II trials conducted between 1980 and 1991.10 The patient population was heterogenous, including those with limited (stage I to III) or advanced disease, and those with or without prior chemotherapy. Some patients received chemoradiotherapy. The chemotherapy involved cisplatin with or without second-generation agents (etoposide being the most common). Of 23 pretreatment clinical and laboratory variables analyzed, eight negative factors were identified in multivariate analysis: metastatic disease (P < .0001), Karnofsky PS ≤ 70 (P = .0003), leukocytosis (P = .0003), skin metastasis (P = .008), increased calcium level (P = .007), abnormal neutrophil count (P = .02), age greater than 60 years (P = .02), and being male (P = .03). Of note, skin metastasis had the highest relative risk (2.80).
Compared with the three analyses described above, our analysis was based on more recent clinical studies involving newer, third-generation regimens. As in the case of the previous ECOG study reported by Finkelstein, the database for our analysis was built entirely from two large randomized phase III trials, providing a more homogenous patient population. All patients in our analysis had stage IV or IIIB with malignant effusion and had received no prior chemotherapy. Almost all had a PS of 0 or 1. All were treated with modern regimens consisting of cisplatin or carboplatin in combination with a third-generation agent, which are presently considered the standard of care for advanced disease. The survival outcome was similar to results from other large contemporary phase III studies. However, as with the other three analyses, this is a retrospective study with inherent limitations. For example, laboratory data were not included in our analysis because only absolute values were entered in the database, and there were concerns about differences in reference values between institutions.
Of 27 pretreatment clinical factors analyzed in univariate and multivariate analyses, six factors stood out as independent survival prognostic markers. Interestingly, subcutaneous metastasis was associated with the worse prognosis, with the highest HR at 1.88. The prognostic importance of lung cancer spreading to subcutaneous tissue or skin was also found in the previous ECOG and ELCWP analyses, as well as in another study from Great Britain.11 Skin or subcutaneous metastasis from lung cancer is uncommon and is rarely included in prognostic analyses. In a retrospective study involving 200 patients with skin metastasis from solid tumors, lung (18%) was among the most common sites of tumor origin, together with breast (32%) and melanoma (16%).12 However, with a median survival of 2.9 months, survival for patients with lung cancer was much shorter than those with breast cancer (13.8 months) or melanoma (13.5 months). The ECOG study by Finkelstein et al8 showed that subcutaneous metastasis is one of three important clinical features discriminating long-term survivors among those who lived beyond 1 year; the other two being PS and sex.
Other metastasis-related prognostic factors in our multivariate analysis were liver involvement and number of metastatic sites (≥ four), with HRs of 1.32 and 1.20, respectively. While hepatic metastases was also an important factor in the prior ECOG study, the number of metastatic lesions (≥ two) predicted a poor outcome in PS 0 to 1 patients in the SWOG analysis. Brain is among the most common metastatic sites in NSCLC patients. Those with brain metastasis have been generally thought to have poor outcome, and traditionally have been excluded from chemotherapy trials. In our study, brain metastasis was associated with an HR of 1.21 in univariate analysis. However, its statistical significance disappeared in multivariate analysis. Because NSCLC patients with brain metastasis are a heterogenous group, with prognosis possibly depending on the extent of intra- and extracranial disease, treatment for these patients should be individualized.
Poor performance status has been widely accepted as one of the most important negative prognostic factors in all cancer patients. The importance of this marker was validated in all four analyses, in which it was among the most significant factors in predicting survival of lung cancer patients. In the SWOG report, patients with PS 0 to 1 did significantly better than those with PS ≥ 2. In the two ECOG analyses, which included primarily patients with PS 0 to 1, those with excellent status (PS 0) lived longer than those with lower PS (1 to 2). In fact, PS 0 was the most important factor in the first ECOG analysis. In the Veterans Administration Lung Group analysis published in 1980, Stanley analyzed survival by Karnofsky score and found that a higher score (by 10% to 15% increments) correlated with increase in median survival.4 Thus, it seems that a better pretreatment PS is associated with a longer survival (PS 0 > PS 1 > PS 2 or higher), suggesting that participants in clinical trials should be stratified further based on their PS (PS 0 v PS 1 v PS 2).
Unlike PS, the prognostic role of weight loss has been somewhat controversial. Some studies, including the previous ECOG analysis, found significant weight loss being among the poor prognostic factors,4,8,13 but others, including the SWOG and ELCWP analyses, did not.9,10,14,15 In our study, we looked at both weight loss and reduced appetite as separate variables. Although the significance of both weight loss and reduced appetite was demonstrated in univariate analysis, only the latter remained an independent negative marker in multivariate analysis.
An interesting finding from our study is that a history of prior surgery for lung cancer was a positive prognostic factor. However, this variable in our database included both diagnostic and therapeutic surgery without speculation of procedures that were performed. It seems plausible that an individual who underwent resection of a primary lung lesion and then recurred several years later with multiple metastases might have a disease with a completely different biology than someone who presented "de novo" with widespread metastases who therefore never had a chance to undergo a surgical procedure. Nevertheless, without complete data for further analysis, we do not have a definitive explanation for this observation.
Of several clinical characteristics analyzed in our research, age, sex, and tumor histology were not associated with survival outcome. In the past, there had been concern of increase in mortality in elderly patients treated with chemotherapy. Although a few studies such as the ELCWP report found that advanced age was a negative factor, most, including the two ECOG analyses, have concluded that older patients have the same outcome as their younger counterparts.16,17 Of note, age ≥ 70 years was actually considered a favorable factor in the SWOG analysis.9 These observations suggest that age should not be the reason to exclude older but fit patients from therapy, particularly with the availability of newer chemotherapy drugs and better supportive measures. Several studies have found that gender is an important prognostic factor, with women faring better than men.6,8-10,14,15,18 However, the influence of gender in survival was not seen in our patient population. The possible role of histology subtypes of NSCLC in determining clinical outcome, such as the favorable impact of squamous cell type in early stage6 or the negative effect of large cell histology in advanced disease,8 has been raised. However, in our study, we found no difference in survival between tumor histology subtypes. The 1997 American Society of Clinical Oncology guideline, updated in 2003, also suggested that NSCLC histology is not an important prognostic factor in patients with advanced, unresectable disease.7,19
In the prior ECOG analysis, Finkelstein created a discriminant function model to predict the probability of surviving longer than 1 year.8 This model requires a discriminant table and different steps to calculate
where S is the calculated prognostic score. A P value greater than .18 predicts that a particular patient will survive longer than 1 year, while P < .18 gives reverse prediction. SWOG and ELCWP investigators used recursive partitioning and amalgamation method to divide patients into subgroups based on prognostic factors identified in their studies. For example, in the SWOG study, three prognostic subgroups were identified: (1) PS 0 to 1, hemoglobin ≥ 11, and age ≥ 47 years (1- and 2-year survivals of 27% and 8%, respectively), (2) PS 2 to 4 and lactate dehydrogenase higher than normal (1- and 2-year survivals of 6% and 1%, respectively), and (3) others, including PS 2 to 4 and normal lactic dehydrogenase (1- and 2-year survival of 16% and 3%).9 We used the Cox regression method to build a prognostic nomogram based on six negative predictive factors. Nomogram models have been used in prostate cancer to predict survival and recurrence probability.20-22 Using our model, 1- and 2-year survival of NSCLC chemotherapy-naive patients undergoing first-line chemotherapy can be easily estimated. The use of this nomogram is limited to patients who are chemotherapy-naive, have good PS, and receive standard third-generation doublets.
In summary, we have identified six negative survival prognostic factors in advanced NSCLC patients undergoing first-line standard chemotherapy: subcutaneous metastasis, decreased PS (ECOG 1 or 2), loss of appetite, liver metastasis, ≥ four metastatic sites, and no previous lung surgery. Based on a large patient database, we have built a unique nomogram using pretreatment clinical factors to predict survival probability of chemotherapy-naive patients treated with third-generation platinum-based doublets. This prognostic model could help clinicians and patients in clinical decision making and treatment tailoring based on the estimated prognosis. In addition, the nomogram and survival markers could be applied in clinical research to stratify patients, avoid biases, as well as plan appropriate studies targeting different subgroups of patients. Using our model instead of historic control data when analyzing survival result of phase II studies, promising first-line regimens could be selected for phase III trials. Although the nomogram was already validated with internal data as described previously, we plan to further validate it using data from other studies.
Authors' Disclosures of Potential Conflicts of Interest
Acknowledgment
We thank the investigators and staff of the Eastern Cooperative Oncology Group for their participation in the trials listed in this study.
NOTES
Supported in part by grant K12-CA 87718 (T.H.).
Presented at the 39th American Society of Clinical Oncology Annual Meeting, Chicago, IL, May 31-June 3, 2003.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Schiller JH, Harrington D, Belani CP, et al: Comparison of four chemotherapy regimens for advanced non-small-cell lung cancer. N Engl J Med 346:92-98, 2002
2. Kelly K, Crowley J, Bunn PA, et al: Randomized phase III trial of paclitaxel plus carboplatin versus vinorelbine plus cisplatin in the treatment of patients with advanced non-small-cell lung cancer: A Southwest Oncology Group trial. J Clin Oncol 19:3210-3218, 2001
3. Bonomi P, Kim K, Fairclough D, et al: Comparison of survival and quality of life in advanced non-small cell lung cancer patients treated with two dose levels of paclitaxel combined with cisplatin versus etoposide with cisplatin: Results of an Eastern Cooperative Oncology Group trial. J Clin Oncol 18:623-631, 2000
4. Stanley KE: Prognostic factors for survival in patients with inoperable lung cancer. J Natl Cancer Inst 65:25-32, 1980
5. Brundage MD, Davies D, Mackillop WJ: Prognostic factors in non-small cell lung cancer: A decade of progress. Chest 122:1037-1057, 2002
6. Feld R, Abratt R, Graziano S, et al: Pretreatment minimal staging and prognostic factors for non-small cell lung cancer. Lung Cancer 17:S3-10, 1997
7. Pfister DG, Johnson DH, Azzoli CG, et al: American Society of Clinical Oncology treatment of unresectable non-small-cell lung cancer guideline: Update 2003. J Clin Oncol 22:330-353, 2004
8. Finkelstein DM, Ettinger DS, Ruckdeschel JC: Long-term survivors in metastatic non-small-cell lung cancer: An Eastern Cooperative Oncology Group Study. J Clin Oncol 4:702-709, 1986
9. Albain KS, Crowley JJ, LeBlanc M, et al: Survival determinants in extensive-stage non-small-cell lung cancer: The Southwest Oncology Group experience. J Clin Oncol 9:1618-1626, 1991
10. Paesmans M, Sculier JP, Libert P, et al: Prognostic factors for survival in advanced non-small-cell lung cancer: Univariate and multivariate analyses including recursive partitioning and amalgamation algorithms in 1,052 patients—The European Lung Cancer Working Party. J Clin Oncol 13:1221-1230, 1995
11. Hickish TF, Smith IE, O'Brien ME, et al: Clinical benefit from palliative chemotherapy in non-small-cell lung cancer extends to the elderly and those with poor prognostic factors. Br J Cancer 78:28-33, 1998
12. Schoenlaub P, Sarraux A, Grosshans E, et al: [Survival after cutaneous metastasis: A study of 200 cases]. Ann Dermatol Venereol 128:1310-1315, 2001
13. Fukuoka M, Masuda N, Furuse K, et al: A randomized trial in inoperable non-small-cell lung cancer: Vindesine and cisplatin versus mitomycin, vindesine, and cisplatin versus etoposide and cisplatin alternating with vindesine and mitomycin. J Clin Oncol 9:606-613, 1991
14. O'Connell JP, Kris MG, Gralla RJ, et al: Frequency and prognostic importance of pretreatment clinical characteristics in patients with advanced non-small-cell lung cancer treated with combination chemotherapy. J Clin Oncol 4:1604-1614, 1986
15. Shinkai T, Eguchi K, Sasaki Y, et al: A prognostic-factor risk index in advanced non-small-cell lung cancer treated with cisplatin-containing combination chemotherapy. Cancer Chemother Pharmacol 30:1-6, 1992
16. Langer CJ, Manola J, Bernardo P, et al: Cisplatin-based therapy for elderly patients with advanced non-small-cell lung cancer: Implications of Eastern Cooperative Oncology Group 5592, a randomized trial. J Natl Cancer Inst 94:173-181, 2002
17. Lilenbaum R, Herndon J, List M: Single-agent (SA) versus combination chemotherapy (CC) in advanced non-small cell lung cancer (NSCLC): A CALGB randomized trial of efficacy, quality of life (QOL), and cost-effectiveness. Proc Am Soc Clin Onc 21:1a, 2002 (abstr 2)
18. Radzikowska E, Glaz P, Roszkowski K: Lung cancer in women: Age, smoking, histology, performance status, stage, initial treatment and survival—Population-based study of 20 561 cases. Ann Oncol 13:1087-1093, 2002
19. ASCO: Clinical practice guidelines for the treatment of unresectable non- small-cell lung cancer. Adopted on May 16, 1997 by the American Society of Clinical Oncology. J Clin Oncol 15:2996-3018, 1997
20. Smaletz O, Scher HI, Small EJ, et al: Nomogram for overall survival of patients with progressive metastatic prostate cancer after castration. J Clin Oncol 20:3972-3982, 2002
21. Kattan MW, Wheeler TM, Scardino PT: Postoperative nomogram for disease recurrence after radical prostatectomy for prostate cancer. J Clin Oncol 17:1499-1507, 1999
22. Kattan MW, Zelefsky MJ, Kupelian PA, et al: Pretreatment nomogram for predicting the outcome of three-dimensional conformal radiotherapy in prostate cancer. J Clin Oncol 18:3352-3359, 2000(Tien Hoang, Ronghui Xu, J)