The addition of chemotherapy to radiotherapy might improve the outcome for these patients [4]. Chemotherapy may reduce local tumor burden, thereby improving local control. More important, chemotherapy may also eliminate or delay the emergence of metastatic disease. In an attempt to verify these hypotheses, several randomized studies of radiotherapy alone compared with chemotherapy and radiotherapy in combination have been done [5-18]. Unfortunately, the results of these studies are conflicting; consequently, it is unknown whether the addition of chemotherapy to radiotherapy improves survival in patients with locally advanced, unresectable, non-small-cell lung cancer.

This lack of knowledge concerns not only whether chemotherapy provides additional benefit but also whether any benefit is substantial enough to offset the side effects of therapy. To better estimate the magnitude of the treatment effect, we did a meta-analysis of randomized trials of combination therapy and radiotherapy alone in patients with locally advanced, unresectable, non-small-cell lung cancer.

Methods

Top Methods Results Discussion Author & Article Info References

Overview

Our primary intent was to ascertain whether the addition of chemotherapy to radiotherapy improved survival in patients with locally advanced, unresectable (stage III), non-small-cell lung cancer. However, because randomized trials in this group of patients have used different drugs in different sequences with radiotherapy, other relevant questions about the value of chemotherapy also emerge. The questions addressed in this analysis are listed in Table 1.

Literature Search

Randomized, clinical trials that compared chemotherapy plus radiotherapy with radiotherapy alone for patients with locally advanced, unresectable, non-small-cell lung cancer were identified in a MEDLINE search of English-language articles published between January 1987 and June 1995 and in a manual review of the references of potentially relevant articles. Eligible articles included an actuarial survival curve or a life Table and reported survival for at least 2 years. Articles were excluded from analysis if the trials were reported only in abstract form or were preliminary publications for which more mature data were subsequently published.

Data Extraction

From the reported survival curves in the eligible articles, we took the number of patients randomly assigned to each treatment group and their probability of survival at 6, 12, 18, 24, 30, and 36 months. Explicit information on censoring was obtained if available. If this information was not available, we assumed that no censoring had occurred. This assumption does not introduce substantial bias in the point estimate of the treatment effect if the pattern of censoring was similar in both treatment groups. However, it may result in unduly narrow CIs caused by underestimation of the variance. From the data extracted from eligible articles, we determined the number of deaths and the number of patients at risk at 6-month intervals.

Statistical Analysis

The goal was to calculate a pooled relative risk for death at 1, 2, and 3 years in the group that received combined therapy compared with the risks in the group that received radiotherapy. We used the two-step method reported by Grilli and colleagues [19]. First, a summary relative risk for death at 1, 2, and 3 years was calculated according to the Mantel-Haenszel technique [20] on the basis of the estimated number of deaths at 6-month intervals and the number of patients at risk during each interval. Thus, the 1-year summary relative risk for death was derived from the number of deaths and patients at risk at 6 and 12 months; the 2-year summary relative risk from the number of deaths and patients at risk at 6, 12, 18, and 24 months; and the 3-year summary relative risk from the number of deaths and patients at risk at 6, 12, 18, 24, 30, and 36 months. The variance of the logarithm of the summary relative risk was estimated by the method of Greenland and Robbins [21], and the 95% CIs shown in Figure 1 were calculated.

View larger version (20K): [in this window] [in a new window]   Figure 1. Chemotherapy plus radiotherapy compared with radiotherapy alone for the treatment of locally advanced, unresectable, non-small-cell lung cancer. The 1-, 2-, and 3-year relative risks for death and 95% CIs are shown for each study (numbers in parentheses are reference numbers). The results of the meta-analysis pooled across all studies and only those in which cisplatin was part of the chemotherapeutic regimen are also shown. Two studies allowed two comparisons: The study by Schaake-Koning and colleagues was a three-group comparison of cisplatin administered weekly (w) or daily (d) with radiotherapy compared with radiotherapy alone, and the trial by Jeremic and colleagues was a three-group comparison of two different schedules of chemotherapy (II and III) plus radiotherapy compared with radiotherapy alone. The trials by van Houtte and colleagues, Mattson and colleagues, and Soresi and colleagues did not present 3-year survival data and thus are not included in the pooled estimated of the effect of combined therapy at 3 years.

Second, a directly pooled estimate of the effect of chemotherapy at 1, 2, and 3 years was calculated by taking a weighted average of the study-specific summary relative risks. The inverse of the logarithms of the study-specific variances was used as weights [20] so that trials with the most stable estimates of the effect of chemotherapy (generally those with larger sample sizes) were more influential. To construct approximate 95% CIs, an estimate of the variance of the pooled effect was calculated as the similarly weighted average of the study-specific variances [20]. For all comparisons, combined therapy was compared with radiotherapy. Thus, a relative risk less than 1 indicates a beneficial effect of chemotherapy, and a relative risk greater than 1 indicates a harmful effect of chemotherapy. A chi-square test of homogeneity for each pooled effect was calculated as described elsewhere [20].

A summary actuarial curve of survival was calculated by pooling the relative risk for death from every study for each 6-month interval between 0 and 36 months. To accomplish this, we adopted a generalized linear statistical model that was similar to the one used in the meta-analysis of chemotherapy for metastatic lung cancer by Grilli and colleagues [19]. The information taken from the studies can be viewed as discrete survival data with censoring and failures occurring during one of six periods: 0 to 6 months, 7 to 12 months, 13 to 18 months, 19 to 24 months, 25 to 30 months, or 31 to 36 months. This generalized linear model was used to estimate the hazard in each time period and was calculated in STATA (College Station, Texas) using the complementary log-log link and a binomial error distribution with the number of trials equal to the number at risk at the beginning of the interval. The resulting hazard is interpreted as the probability of death at some time in that interval, with the condition of being alive at the start of the interval. The dependent variable was the number of deaths in each period, and the covariates included categorical variables for the time frame, treatment, and study. To test whether the effect of chemotherapy was proportional in each 6-month interval, an interaction term between the time frame and treatment was included in the model. The relative risk for death during treatment in each interval shown in Figure 2 was then approximated by the ratio of the estimated hazards. Finally, the hazards were used to estimate the interval-specific probability of surviving time interval t (p_t). The cumulative probability for surviving time interval t (S_t) was then calculated as the product of the interval-specific survival probabilities: S_t = p_t·p_t-1·...·p₂·p₁ (Figure 3).

View larger version (11K): [in this window] [in a new window]   Figure 2. Pooled relative risk for death in locally advanced, unresectable, non-small-cell lung cancer and 95% CI for each 6-month period. The relative risk represents the risk for death during that period conditional on being alive at the start of that period. The number of patients across all studies who were estimated to be at risk at the start of the period is shown on the right side of the figure.

View larger version (10K): [in this window] [in a new window]   Figure 3. Cumulative survival probabilities in patients with locally advanced, unresectable, non-small-cell lung cancer treated with chemotherapy plus radiotherapy (solid line) and radiotherapy alone (dashed line).

To aid description and interpretation of the overall result, we also estimated a measure of the absolute benefit of the addition of chemotherapy to radiotherapy. First, the mean gain in life expectancy seen with chemotherapy at the end of 3 years was taken as the area between the cumulative survival curves (as calculated above). Second, the weighted average of median survival was calculated by assigning each study a weight equal to its total sample size.

Results

Top Methods Results Discussion Author & Article Info References

Literature Search

Seventeen randomized trials published between January 1987 and June 1995 that compared radiotherapy alone with combined chemotherapy and radiotherapy in patients with locally advanced, unresectable, non-small-cell lung cancer were identified for possible review. Three were considered ineligible for further review: Two were reported in abstract form only [22, 23], and one included weekly chemotherapy in the group receiving radiotherapy [24]. Fourteen trials [5-18] were thus eligible for this meta-analysis.

These 14 trials are listed in Table 2 with treatment specification, trial size, and median length of survival. Chemotherapy was administered concurrently with radiotherapy in 6 trials [7, 9, 10, 13, 15, 17]; all but 1 of these [17] included cisplatin as part of the chemotherapy. In the remaining 8 trials, patients received chemotherapy before [5, 8, 11, 12, 14], after [16], or before and after [6, 18] radiotherapy. Cisplatin was used in 5 of these 8 trials.

Comparative Analyses

The answers to the question "What is the relative efficacy of chemotherapy plus radiotherapy compared with radiotherapy alone?" are presented graphically in Figure 1. All 14 trials were eligible for this analysis and thus allowed 16 comparisons; 2 trials [9, 13] had three treatment groups, allowing 2 comparisons in each. A trend toward a reduced risk for death was seen with combined therapy (relative risk < 1) in 12 trials (3 were significant) and with radiotherapy (relative risk > 1) in 4 trials (1 was significant at 1 year only). However, when the results of all these trials were pooled, there was a clear indication that the addition of chemotherapy to radiotherapy reduced the risk for death at 1 year (relative risk, 0.88 [95% CI, 0.80 to 0.96]; test for homogeneity, P = 0.07), 2 years (relative risk, 0.87 [CI, 0.81 to 0.94]; test for homogeneity, P > 0.2), and 3 years (relative risk, 0.83 [CI, 0.77 to 0.90]; test for homogeneity, P < 0.2). The heterogeneity at 1 year was the result of two trials [7, 13]. When these trials were excluded from the analysis, heterogeneity was reduced (test for homogeneity, P > 0.2), and the overall estimate of the effect of chemotherapy at 1 year was somewhat less (relative risk, 0.91 [CI, 0.83 to 0.99]). We could not identify any systematic difference in these two studies that might account for the heterogeneity that was seen.

The relative risk for death and the corresponding 95% CIs for each 6-month interval are shown in (Figure 2). For example, for patients alive 12 months after treatment was started, the addition of chemotherapy to radiotherapy reduced the relative risk for death by 15% (relative risk, 0.85 [CI, 0.73 to 0.99]) in the next 6 months. The benefit of chemotherapy appeared to increase with time, reaching a maximum of a 46% reduction in the risk for death between the 31st and 36th month after the start of treatment (relative risk, 0.54 [CI, 0.36 to 0.81]). However, relatively few patients survived long enough to appreciate this benefit.

Table 3 shows the results of the meta-analyses pertaining to the other questions about the relative efficacy of chemotherapy in patients with locally advanced, non-small-cell lung cancer. When considered separately, trials of concurrent and sequential chemotherapy yielded similar treatment effects. The addition of chemotherapy to radiotherapy was associated with a 10% to 20% decrease in the risk for death. When the analysis was restricted to only trials in which cisplatin was used as part of the chemotherapy regimen, the results were similar.

The cumulative survival probabilities for radiotherapy and combined therapy, calculated from the estimates of our model, are shown in Figure 3. The addition of chemotherapy to radiotherapy was associated with a mean gain in life expectancy of approximately 2 months by the end of 3 years. We also calculated the weighted average of median survival for each of the comparisons; these results are shown in Table 4. Overall, the addition of chemotherapy to radiotherapy improved median survival from 10.3 months to 12.0 months. The magnitude of the benefit measured in these terms was similar across all comparisons.

Discussion

Top Methods Results Discussion Author & Article Info References

Current treatment of locally advanced, unresectable, non-small-cell lung cancer is less than optimal, and survival remains poor [2]. The use of thoracic irradiation, considered the gold standard for therapy in the United States and elsewhere, has done little to improve survival of patients with this disease. In an attempt to improve outcome for these patients, chemotherapy has been added to radiotherapy for various reasons [4], including to decrease the size of the tumor before radiotherapy, to act synergistically with radiotherapy, and to control distant metastasis. Several randomized trials have compared chemotherapy plus radiotherapy with radiotherapy alone, and the results have been conflicting. This study represents an effort to consolidate the data about the value of chemotherapy in locally advanced, unresectable, non-small-cell lung cancer.

The results of our analysis suggest that compared with radiotherapy alone, the combination of chemotherapy and radiotherapy improves overall survival in patients with locally advanced, unresectable, non-small-cell lung cancer. The relative risk for death was reduced by 12% at 1 year, 13% at 2 years, and 17% at 3 years. This translates into a mean potential gain in survival of about 2 months in the 3 years after therapy is initiated. The clinical relevance of this result, however, must be balanced against the toxicities commonly associated with chemotherapy (such as nausea, vomiting, hair loss, and the increased risk for infection that results from myelosuppression).

Limitations

Although meta-analysis is a powerful statistical tool for pooling data from several sources to increase the power to determine associations, it has limitations. First, publication bias may lead to an overestimation of the effect of chemotherapy because reports of trials with negative results may be less likely to be published [25]. However, 10 of the 14 trials included in this analysis reported negative results—either a CI that included 1 or a P value less than 0.05. Furthermore, our literature search identified only two abstracts that were not subsequently published as full articles. Publication bias cannot be entirely excluded and could account for some of the effect that was seen.

Second, the data from the studies included in this analysis are somewhat old. If the techniques of radiotherapy and chemotherapy improved while the studies were done, the results presented here may be conservative. However, it is on the basis of these studies, whether considered individually or collectively, that current judgments about the treatment of unresectable, non-small-cell lung cancer are made.

Third, we did not have access to patient-level data. We relied on data extracted directly from the published survival curves to calculate the pooled estimate of the effect of chemotherapy. Because explicit information on censoring was not always available in the published reports, it is likely that we have underestimated the number of censored observations. Although the point estimate of the effect size is not affected, the precision of that estimate is affected, resulting in an unduly narrow CI. However, our results are fairly robust. Assuming a constant pattern of censoring (which may or may not be valid), an estimated 30% of the patients would have to have been censored in each 6-month period in trials for which this information was unavailable before the upper bound of the confidence interval crossed the boundary of no effect (that is, a relative risk of 1.0). The lack of patient-level data may introduce other sources of bias, the magnitude and direction of which are difficult to discern [26].

In addition, some care is required before these results can be generalized to all patients with locally advanced, unresectable, non-small-cell lung cancer. Patients with poor performance status or substantial weight loss were typically excluded from the trials that we analyzed. Our results therefore cannot be generalized to such patients in whom the benefit of adding chemotherapy is uncertain but is likely to be less and possibly harmful.

Combining studies that use different treatment protocols and different end points also carries some risk. We were primarily interested in the more general question of whether chemotherapy added to radiotherapy in locally advanced, unresectable, non-small-cell lung cancer effectively improves survival. The studies under consideration, however, used various drug combinations and treatment schedules. We addressed this in several subgroup analyses Table 3 and found that the effect of chemotherapy was similar for different therapeutic approaches. Nevertheless, this type of analysis does not permit us to assess which combination of drugs and which treatment schedule are most effective.

Finally, patients are concerned about outcomes other than survival when comparing treatments. They are likely to be concerned about outcomes that pertain to quality of life, such as time in the hospital, changes in symptoms, and treatment-related toxicities. To date, information that would allow a comparison of the quality of life of persons receiving different forms of therapy is not available. Society as a whole is also interested in an outcome that cannot be addressed by this analysis: the relative cost of these therapies.

Clinical Implications

Despite these potential limitations, this meta-analysis suggests that chemotherapy offers a beneficial contribution to the management of locally advanced, unresectable, non-small-cell lung cancer. However, because most patients with this form of cancer still die of their disease, improved methods of treatment are needed. Because the development of distant metastasis remains an important cause of treatment failure, the search for more effective systemic therapies is an important area for clinical investigation in this disease.

Although the benefit of adding chemotherapy to radiotherapy for any individual patient may vary considerably and may be substantial for some patients, the absolute benefit is small on average. This amount of benefit must be weighed against the toxicity associated with chemotherapy. Patients must therefore choose between a decrease in quality of life and future survival. The magnitude of the survival benefit is therefore critical. For example, if patients and their physicians decide that a 5% reduction in the relative risk for death at 2 years is sufficient to accept the toxicity of chemotherapy, the data presented here suggest that the addition of chemotherapy is worthwhile. On the other hand, if the patient decides that a 20% reduction in the 2-year relative risk for death is required to accept the toxicity of chemotherapy, these data suggest that the addition of chemotherapy is not worthwhile. For a relative risk reduction between 5% and 20%, the addition of chemotherapy is a close call. Therefore, recommendations to patients about the use of chemotherapy should depend on information about quality of life and patient preferences.

Dr. Anthony: Section of Hematology/Oncology, Dartmouth-Hitchcock Medical Center, 1 Medical Center Drive, Lebanon, NH 03756.