Lung cancer remains the leading cause of cancer death in the United States, killing more patients than breast, colon, and prostate cancer combined.1 The 5-year survival rate for patients who present with advanced disease is <2%.2 A 1995 meta-analysis of randomized studies that compared best supportive care with chemotherapy, particularly platinum-based regimens, demonstrated symptom palliation and prolongation of survival with the addition of chemotherapy.3 However, this improvement in survival is modest at best, prompting numerous clinical trials over the past decade aimed at identifying more active chemotherapy regimens, combinations, or schedules.
A number of new agents have been identified over the past 10 to 20 years, including paclitaxel, docetaxel, gemcitabine, vinorelbine, and irinotecan. These so-called “third-generation” cytotoxic agents are active in nonsmall cell lung cancer (NSCLC), particularly when combined with a platinum agent. Although no regimen has emerged as significantly more efficacious than another in randomized trials, the general consensus in the community is that they are more active than older regimens.4–6 Phase III trials directly comparing older regimens (cisplatin and etoposide or teniposide) with regimens that contain a newer drug (paclitaxel) have demonstrated this increased efficacy.7, 8
Another hypothesis for the improvement in survival, however, is that the natural history of NSCLC may be changing. To explore this hypothesis further, we analyzed all Eastern Cooperative Oncology Group (ECOG) trials in patients with advanced NSCLC that were conducted between 1981 and 2000. We investigated whether changes in demographic factors or treatment regimens were correlated with differences in outcome between patients who were diagnosed from 1981 to 1990 and from 1991 to 2000.
This analyses allowed us to investigate the role of multiple factors, including changes in natural history, eligibility criteria, and treatments, that may have played a role in improving the survival of patients with NSCLC over time. A multifactorial explanation for this phenomenon is hypothesized.
MATERIALS AND METHODS
Data from Phase II and III ECOG first-line studies in patients with advanced NSCLC that were conducted from 1981 to 2000 were included in this retrospective analysis (Table 1).4, 7, 9–15 During that period, 3553 patients were enrolled in 9 trials. In total, 3398 patients were included in the current analysis. Patients who were known to be ineligible at the time of this analysis, based on the most recent updated eligibility information, were excluded. Patients were divided into 2 groups based on their date of diagnosis (before 1990 and after 1990).
Table 1. Eastern Cooperative Oncology Group First-Line Trials in Patients with Advanced Nonsmall Cell Lung Cancer, 1980-2000
|E1581 (1981): Ruckdeschel et al., 19869||III||0||0-2||No||CAMP||136||17||5.8||18‡|
| || || || || ||MVP||137||31||5.1||NR|
| || || || || ||Vindesine/cisplatin||140||20||6.0||NR|
| || || || || ||Etoposide/cisplatin||138||25||6.1||NR|
|E1583 (1983): Bonomi et al., 198910||III||0||0-2||No||MVP||183||20||5.2||NR|
| || || || || ||Vinblastine/cisplatin||184||13||5.8||NR|
| || || || || ||MVP/CAMP||183||13||5.7||NR|
| || || || || ||Carboplatin||96||9||7.3||NR|
| || || || || ||Iproplatin||97||6||6.0||NR|
|E1585 (1985): Bonomi et al., 199411||II||0||0-1||No||Etoposide/cisplatin||44||11||6.8||NR|
| || || || || ||Avicin||43||0||4.9||NR|
|E1587 (1987): Chang et al., 199812||II||0||0-1||No||Etoposide/cisplatin||43||16||8.7||32|
| || || || || ||Echinomycin||46||5||5.6||20|
| || || || || ||Trimetrexate||47||5||6.4||25|
|E1589 (1989): Chang et al., 199313||II||0||0-1||No||Paclitaxel||25||21||5.6||42|
| || || || || ||Merbarone||47||6||4.6||22|
| || || || || ||Piroxantrone||47||2||6.7||33|
|E2588 (1988): Chang et al., 199514||II||0||0-1||No||Gallium nitrate||17||0||5.6||NR|
| || || || || ||Amonafide||20||0||5.2||NR|
| || || || || ||Teniposide||18||0||3.1||NR|
|E5592 (1992): Bonomi et al., 20007||III||19§||0-1||No||Cisplatin/etoposide||200||12||7.6||32|
| || || || || ||Cisplatin/paclitaxel/G-CSF||201||32||10.1||40|
| || || || || ||Cisplatin/paclitaxel||198||27||9.5||37|
|E1594 (1994): Schiller et al., 20024||III||13∥||0-1 (2)¶||Yes (13%)||Cisplatin/paclitaxel||303||21||7.8||31|
| || || || || ||Cisplatin/gemcitabine||301||22||8.1||36|
| || || || || ||Cisplatin/docetaxel||304||17||7.4||31|
| || || || || ||Carboplatin/paclitaxel||299||17||8.1||34|
|E2598 (1989): Langer et al., 200415||II||9∥||0-1||Yes (17%)||Carboplatin/paclitaxel/trastuzumab||56||18||9.2||42|
Chi-square tests of association and 2-sample t tests were used to compare the proportions and distributions of patient characteristics between pre-1990 patients and post-1990 patients. A 2-sample t test was used to determine whether the mean length of the interval between the date of diagnosis and the date of study entry differed between the 2 groups.
Log-rank tests were used to compare progression-free survival (PFS), overall survival (OS), and the time to death from the date of progression (among patients who had progressive disease) between the 2 groups. Log-rank tests also were used to compare OS and time to death from the date of progression between patients who had pulmonary metastatic disease only and patients who had other distant metastases at study registration. OS and PFS were calculated from the date of study entry.
A Cox proportional hazards model was examined with respect to OS and PFS (both measured from the date of study entry) as well as the time to death from the date of progression. Patients who remained alive at the time of this analysis were censored at the last time they were known to be alive. In addition to the year of diagnosis (pre-1990 vs. post-1990, inclusive), known prognostic factors were included in the model, specifically performance status (PS) (0 or 1 vs. 2), treatment regimen (noncisplatin-based vs. cisplatin-based), weight loss, pulmonary metastases only versus nonpulmonary metastases, age (younger than 70 years vs. 70 years and older), gender, and histology (adenocarcinoma vs. nonadenocarcinoma). For the analysis of the time to death from the date of progression, only patients who experienced disease progression were included in the analysis.
Cox proportional hazards models that incorporated the varying dates of study entry from the date of diagnosis also were examined with respect to OS and PFS. These models used the date of diagnosis as the baseline for measuring OS and PFS. With this method of calculating OS and PFS, events are not observable (or are censored) in the interval between the date of diagnosis and the date of study entry (these data are considered left-truncated). Because of this unobservable portion of the survival curves, it is not possible to calculate accurately the medians of the survival distributions from the date of diagnosis. However, these models correctly measure the effect of the covariates on the unspecified baseline hazard function. The hazard ratios are calculated over the range for which both cohorts of interest are observed. These models help correct the lead-time or ascertainment bias that may occur by simply measuring OS and PFS from the date of study entry.
Patient and Study Characteristics
Six advanced NSCLC trials were identified that took place between 1981 and 1990, and 3 trials were identified that took place after 1990 (Table 1). The post-1990 trials tended to be larger and were more likely to allow patients with Stage IIIB disease (0 of 6 trials vs. 3 of 3 trials pre-1990 and post-1990, respectively) and with brain metastases (0 of 6 trials vs. 2 of 3 trials, respectively). However, more trials pre-1990 allowed patients who had a PS of 2 (on the ECOG scale; 2 of 6 trials vs. 0 of 3 trials). In addition, more patients received treatment with cisplatin after 1990 than before 1990 (Table 2).
Table 2. Patient Characteristics
|No. of patients*||1520||1878|| |
|No. of trials||6||3|| |
|Patients on cisplatin regimens, %||79.7||94.8||<.0001|
|Mean age, y||60.0||60.9||.0048|
|Performance status 0-1, %||81.8||96.5||<.0001|
|Stage IIIB, %||0||15||<.0001|
|Weight loss >10%, %||15.1||12.3||.0197|
|Adenocarcinoma and adenosquamous carcinoma histology||47.2||57.3||<.0001|
|Sites of metastases|| || || |
| ≤1 Metastatic site||45.9||29.4||<.0001|
| Bone disease||37.5||32.3||.0016|
| Pulmonary metastasis||59.7||79.4||<.0001|
| Liver metastasis||17.0||20.4||.0138|
| Brain metastasis||0.2||8.2||<.0001|
| Pulmonary metastasis ONLY||22.2||25.9||<.0134|
| Bone metastasis ONLY||8.6||2.8||<.0001|
| Liver metastasis ONLY||3.0||1.4||.0022|
For the analyses presented in this report, patients were divided into 2 groups based on their dates of diagnosis (pre-1990 vs. post-1990, inclusive). Because of the large number of patients, all of the variables examined were statistically significant (Table 2). The different eligibility criteria for the various trials resulted in a greater proportion of patients with a PS of 0 or 1 after post-1990 (96.5%; pre-1990, 81.8%;) and a greater proportion of patients with brain metastasis post-1990 (8.2%; pre-1990, 0.2%). There were more patients who had weight loss >10% pre-1990 (15.1%; post-1990, 12.3%), and a greater percentage of women entered post-1990 (36.8%; pre-1990, 29.9%). Fewer patients with adenocarcinoma were entered pre-1990 (45.1%) compared with post-1990 (57.7%), whereas more patients with large cell carcinoma (pre-1990, 16.7%; post-1990, 7.7%) and squamous cell carcinoma (pre-1990, 32.8%; post-1990, 20.1%; P<.0001) were entered before 1990. Only patients with Stage IV metastatic disease were enrolled on the 6 pre-1990 trials, but patients with Stage IIIB disease were included in the later studies (19% on E5592, 13% on E1594, and 9% on E2598). Overall, a greater proportion of patients with a PS of 2 and/or brain metastases were entered pre-1990 (18.4%; post-1990, 11.4%; P<.0001).
Differences were observed in on-study characteristics (Table 2). The percentage of patients with no metastasis or with only 1 metastatic site was 45.9% in the decade preceding 1990, compared with 29.4% post-1990. The percentage of patients with bony metastases dropped slightly (37.5% pre-1990 vs. 32.3% post-1990), whereas the percentage of patients with pulmonary and liver metastases increased (59.7% vs. 79.4% and 17.0% vs. 20.4%, respectively, pre-1990 vs. post-1990) between the 2 decades.
The mean interval between the date of diagnosis and date of study entry in the decade between 1991 and 2000 was roughly 50% of the same interval in the preceding decade (6.9 months pre-1990 vs. 3.9 months post-1990) (Table 3). The median PFS was 2.6 months pre-1990 compared with 3.5 months post-1990. Patients who were diagnosed before 1990 had a median time to death from the date of progression of 2.7 months compared with 4.3 months for patients who were diagnosed after 1990. The median OS from the date of study entry was 5.8 months pre-1990 compared with 8.2 months post-1990. OS and PFS were measured from the date of study entry. All analyses were statistically significant (P<.0001).
Table 3. Patient Outcomes
|Mean interval between date of diagnosis and date of study entry||6.9||3.9||<.0001*|
|Median PFS (from study entry)||2.6||3.5||<.0001†|
|Median time to death from the date of disease progression||2.7||4.3||<.0001†|
|Median overall survival from study entry‡||5.8||8.2||<.0001†|
Analysis of Effects on Survival
A Cox proportional hazards model was examined with respect to OS measured from the date of study entry (Table 4). Factors that were associated with better survival included being diagnosed after 1990, a PS of 0 or 1, cisplatin-based therapy, weight loss ≤10%, metastases confined to the lung, and female gender. No significant effects of age or histology on OS were found.
Table 4. Cox Proportional Hazards Model: Overall Survival
|Date of diagnosis: post-1990 vs. pre-1990||<.0001||0.765||<.0001||0.762|
|Performance status 2 vs. 0-1||<.0001||1.580||<.0001||1.538|
|Cisplatin-based chemotherapy vs. noncisplatin regimens||.0261||0.882||.0244||0.881|
|Age ≥70 y vs. <70 y||.6513||0.978||.9120||1.005|
|Weight loss >10% (6 mo) vs. weight loss ≤10%||<.0001||1.262||<.0001||1.233|
|Pulmonary metastases ONLY vs. any nonpulmonary metastases||<.0001||0.703||<.0001||0.723|
|Nonadenocarcinoma histology vs. adenocarcinoma or adenosquamous carcinoma||.4425||1.028||.8075||1.009|
|Female vs. male||<.0001||0.862||.0001||0.865|
In the Cox proportional hazards model for PFS (measured from the date of study entry), the following factors were associated with a longer PFS: diagnosis after 1990, a PS of 0 or 1, cisplatin-based therapy, age 70 years or older, weight loss ≤10%, the presence of pulmonary metastases only, and female gender. No significant effect of histology was found on PFS (Table 5).
Table 5. Cox Proportional Hazards Model: Progression-Free Survival
|Date of diagnosis: post-1990 vs. pre-1990||.0089||0.905||<.0001||0.751|
|Performance status 2 vs. 0-1||<.0001||1.388||<.0001||1.395|
|Cisplatin-based chemotherapy vs. noncisplatin regimens||<.0001||0.755||<.0001||0.735|
|Age ≥70 y vs. <70 y||.0514||0.910||.2728||0.943|
|Weight loss >10% (6 mo) vs. weight loss ≤10%||.0018||1.180||.1590||1.085|
|Pulmonary metastases ONLY vs. any nonpulmonary metastases||<.0001||0.730||<.0001||0.708|
|Nonadenocarcinoma histology vs. adenocarcinoma or adenosquamous carcinoma||.9361||0.997||.5832||1.022|
|Female vs. male||.0013||0.886||.0266||0.914|
The median interval between the date of diagnosis and the date of study entry was longer by 3 months in the pre-1990 group compared with the post-1990 group (Table 3). Therefore, in the preceding analyses, in which both endpoints were measured from the date of study entry, it may be argued that the differences observed in the hazard rates between the pre-1990 and post-1990 cohorts were because patients in the latter group entered the study at an earlier point in their disease process. That is, if we assume increasing hazard rates throughout the disease process, then the hazard rates from the pre-1990 and post-1990 groups are being compared (incorrectly) at different points in the disease process. Therefore, the hazard ratio provides a misleading comparison of the 2 cohorts. To address this issue, we examined Cox proportional hazards models that measured OS and PFS from the date of diagnosis. For both OS and PFS, we obtained results very similar to those obtained from the model that measured endpoints from the date of study entry (Tables 4, 5). The only difference between the 2 models was that, when the date of diagnosis was used instead of the date of study entry, age and weight loss no longer were significant factors for PFS.
A Cox proportional hazards model was also used to explore the correlations of known prognostic factors with the interval between disease progression and death (Table 6). Factors that were associated with a longer interval between progression and death were post-1990 diagnosis, a PS of 0 or 1, weight loss ≤10%, the presence of pulmonary metastasis only, and female gender. Treatment regimen, age, and histology were not identified as statistically significant factors.
Table 6. Cox Proportional Hazard Regression Model: Time to Death from the Date of Progression*
|Date of diagnosis: post-1990 vs. pre-1990||<.0001||0.729|
|Performance status 2 vs. 0-1||<.0001||1.567|
|Cisplatin-based chemotherapy vs. noncisplatin regimens||.4788||1.045|
|Age ≥70 y vs. <70 y||.4935||1.038|
|Weight loss >10% (6 mo) vs. weight loss ≤10%||<.0001||1.256|
|Pulmonary metastases ONLY vs. any nonpulmonary metastases||<.0001||0.737|
|Nonadenocarcinoma histology vs. adenocarcinoma or adenosquamous carcinoma||.1685||1.055|
|Female vs. male||.0293||0.915|
Given the improvement in survival post-1990 and the increasing incidence of pulmonary metastases, we sought to determine whether these 2 factors were related (Table 7). Patients who had pulmonary metastases only had an improved OS (9.4 months from the date of study entry) and an improved median time to death from the date of disease progression (4.9 months) compared with patients who had metastases in other sites (6.3 months and 3.2 months, respectively). The OS for patients with both pulmonary and nonpulmonary metastases, however, was similar to the OS for patients with nonpulmonary metastases only (6.0 months vs. 6.3 months, respectively). Based on these findings and on the results from the Cox proportional hazards regression model described above with respect to OS, it appears that having pulmonary metastases only is a favorable prognostic factor for survival compared with patients who have any nonpulmonary metastases. However, because the percentage of patients with metastases only in the lungs rose by only 3% (from 22.9% to 25.9%) pre-1990 and post-1990, respectively, it is unlikely that the observed improvement in survival between the 2 decades (median OS: pre-1990, 5.8 months; post-1990, 8.2 months) can be attributed fully to the small increase in the percentage of patients with pulmonary metastases only.
Table 7. Effect of Pulmonary Metastases on Outcome
|No. of patients||811||1545||871||2416||1682|
|Months between progression and death||4.9||3.2||3.2||3.2||6.0|
|Log-rank P|| ||<.0001||<.001||<.0001||<.0001|
|Log-rank P|| ||<.0001||<.0001||<.0001||<.0001*|
Numerous meta-analyses and retrospective reviews have demonstrated a small but significant improvement in survival for patients with advanced NSCLC over the past 3 decades.3, 16–18 The current analysis of data from 2 decades of ECOG trials adds to the body of literature on the topic and indicates that this improvement may be because of a shift in the natural history of lung cancer. More women are being diagnosed with the disease, patients tend to have less weight loss prior to diagnosis, the distribution of metastases has changed, and the time from disease progression to death has increased, all indicating a change in the disease itself. Clearly, however, to consider the changing natural history of lung cancer the only factor leading to improved survival would ignore improvements in chemotherapy over time and changes in eligibility criteria. A more realistic view is that the survival improvement is multifactorial.
Factors that differed between the decades and that may indicate a change in the natural history of lung cancer, leading to better outcomes in the post-1990 studies, were an increase in the number of women and fewer patients with significant weight loss. It is known from Surveillance, Epidemiology, and End Results and other data that women with NSCLC generally have better survival than men.2, 19 Indeed, it has been proposed that “gender migration” may account for some of the better outcomes that have been described in recent clinical trials.20 According to this hypothesis, the increased incidence of lung cancer in women, who inherently appear to have better survival than men, may explain why outcomes for all lung cancer patients are improving. The greater percentage of women with lung cancer who have a long median survival increases the median survival for the entire group of patients with lung cancer. We are only now beginning to understand gender differences in NSCLC biology and susceptibility.21 It can be argued that the increasing number of women with NSCLC reflects the increased smoking rates among women, which lagged behind those of men. This epidemiologic factor may reflect societal changes more than “natural history” in the strictest sense but does not detract from the impact of increased numbers of women with lung cancer on the improved survival for lung cancer patients as a group over time.
Weight loss >10% also was seen in a slightly greater percentage of patients who were enrolled pre-1990 (15.1%; post-1990, 12.3%). Weight loss as a symptom of lung cancer is associated independently with poorer survival and increased toxicity from therapy.22–24
Patients who were treated after 1990 were more likely to have >1 metastatic site (more advanced disease usually is considered a poorer prognostic factor) and a shorter interval between the date of diagnosis and the date of study entry. These latter observations may be explained by stage migration, in which the increasing use of sensitive computed tomography scans has resulted in picking up smaller amounts of disease earlier. Improved imaging also may account for the increasing incidence of pulmonary and hepatic metastases.
It is noteworthy that patients with pulmonary only metastases, a subset with significantly improved survival, were more frequent post-1990. Although the increased incidence of pulmonary only disease is unlikely to be responsible completely for the improvement in survival post-1990, because the incidence of pulmonary only metastases increased by only 3.7%, it may explain in part the improved survival (Table 7). Improved imaging techniques alone cannot explain this trend entirely, because those techniques would be just as likely to find distant metastases.
More patients with adenocarcinoma were entered on studies after 1990 than before 1990. This finding is in keeping with a worldwide increase in the incidence of adenocarcinoma of unclear etiology.25–27 In the current study, adenocarcinoma was not identified as an independent prognostic factor for OS or long-term survival. This is consistent with other studies, which also failed to identify histology as a significant prognostic factor.28, 29 Thus, despite the striking change in the natural history of lung cancer reflected by an increase in adenocarcinoma, this change does not explain the improved survival for patients with NSCLC over time.
One major difference between the 2 decades was that the median time to death from the date of disease progression increased by nearly 62% (2.7 months pre-1990 vs. 4.2 months post-1990). Presumably, this was not because of differences in first-line chemotherapy but either because of changes in the patient population being treated or because of events after patients went off study. A shift toward a more indolent disease course for patients NSCLC may explain such a difference, although data are lacking to support this conclusion. A more likely explanation is that improvements in second-line regimens and/or supportive care therapies have lengthened the time between disease progression and death. Recent studies, in fact, have demonstrated that second-line and even third-line regimens can improve survival and reduce symptoms.30–33 However, because no second-line treatment data were collected on any of these studies, it is impossible to determine whether better second-line or supportive therapies were responsible for the changes noted.
Despite the compelling evidence for a change in the natural history of lung cancer presented above, the NSCLC survival improvement over time has been attributed largely to the introduction of platinum-based regimens, with a 23% reduction in the risk of death and an absolute benefit of 10% in 1-year survival compared with best supportive care according to a 1995 meta-analysis.3 Eighty percent of patients who were treated on ECOG trials from 1981 to 1990 received a platinum agent, compared with 95% of patients who were treated from 1991 to 2000. Patients who were treated prior to 1990 also were more likely to receive a single chemotherapeutic agent instead of a combination regimen. At least 1 large randomized trial demonstrated only marginal benefit from the addition of a second agent to 1 of the third-generation chemotherapeutics (docetaxel)34; however, several other recent randomized studies, along with the American Society of Clinical Oncology guidelines, support the superiority of doublet regimens over single agents.35–38 Authors of a report on a recent meta-analysis that included 65 trials (13,601 patients) concluded that both tumor response and 1-year survival improved significantly with the addition of a second chemotherapeutic agent.39 Although the differences in chemotherapy may explain in part the improved outcomes post-1990, we explored whether other factors may account for this difference.
The trials in the 2 decades differed in terms of on-study characteristics, which resulted in an imbalance in the distribution of patients with poor prognostic factors, such as poor PS, brain metastases, and Stage IV disease.22, 23, 29 Later studies were more likely to exclude patients with a poorer PS: Before 1990, 267 patients (18.2%) had a PS of 2 compared with 65 patients (3.5%) after 1990. However, fewer patients had had brain metastases (also a poor prognostic factor) pre-1990 (0.2%) than post-1990 (8.2%). When the groups were analyzed together, a greater proportion of patients with a PS of 2 and/or with brain metastases were entered pre-1990 (18.4%) than post-1990 (11.4%; P<.0001). In addition, only patients with Stage IV disease were enrolled in the 6 trials pre-1990, whereas the later studies included patients with Stage IIIB disease (i.e., 15% of patients enrolled on the later 3 trials). Thus, the difference in outcome may be explained in part by the imbalance in the distribution of these patients with poorer prognostic factors.
These data reconfirm the prognostic importance of such traditional factors as PS, gender, weight loss, and cisplatin-based regimens.19, 22–24, 29 They also reconfirm the lack of prognostic significance for age among the “fit elderly.” Other studies also demonstrated that response rates and survival in fit elderly patients with NSCLC who received platinum-based treatment appear to be similar to those in younger patients.23, 35–37, 40–42
In conclusion, the clinical characteristics of patients who participate in advanced disease NSCLC ECOG studies have changed over the past 20 years, with the result that patients with more favorable prognostic characteristics are entered on these studies. Part of this is because of changes in eligibility criteria, but a change in the natural history of the disease also is indicated. The longer time to progression in the years post-1990 suggests that improved survival in part also may be because of better chemotherapy. The longer interval between disease progression and death suggests that there have been advances in second-line therapies and supportive care or, less likely, that patients with more indolent disease were entered on ECOG trials after 1990. Although the reasons behind these changes were not identified definitively in our retrospective analysis, the current data are consistent with the hypothesis that the improved survival observed over the past decade is multifactorial.
The authors thank David Harrington, PhD, for his valuable statistical advice.