Chapter  143:  Management of Small Cell Lung Cancer

Discussion and Future Research

The strongest evidence available for this report is a patient-level meta-analysis showing that PCI improves survival of SCLC patients who achieved CR following primary therapy. No other question yielded evidence so robust. Our conclusions typically relied on a single trial showing treatment effects that were modest at best, and sometimes equivocal. This was apparent in our review of evidence for the sequence, timing, dosing and fractionation of TRTx. For example, the case for concurrent over sequential delivery rests largely on a single multicenter trial (Takada, Fukuoka, Kawahara, et al., 2002). Support for early concurrent therapy comes from the multicenter trial by Murray-Coy-Field (Murray, Coy, Pater, et al., 1993/Coy, Hodson, Murray, et al., 1994/Feld, Payne, Hodson, et al., 1988); however, two other multicenter trials, (Perry, Eaton, Propert, et al., 1987/Ahles, Silberfarb, Rundle, et al., 1994/Perry, Herndon, Eaton, et al., 1988; James, Spiro, O'Donnell, et al., 2003 [abstract]) found no advantage. However, the meta-analysis of 11 studies did not find significant reductions in 2- and 3-year mortality for early TRTx. For some questions (i.e., management of mixed histology disease; surgery for early limited SCLC) comparative trials were nonexistent.

Results reported by Jeremic, Shibamoto, Nikolic, et al., (1999) on TRTx for extensive-stage disease, need replication in a multicenter setting.

PET may be more sensitive in detecting disease outside the brain than conventional staging modalities. Future studies should fully report the frequency of correct and incorrect staging changes when PET is added to conventional tests and should link diagnostic performance to outcomes such as improvement in survival or reduced morbidity. Studies should be conducted according to standards described by the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) and reported according to the Standards for Reporting of Diagnostic Accuracy (STARD) statement.

Complicating the evaluation of SCLC treatment are overall poor outcomes and small effect sizes, necessitating large numbers of patients in trials. Furthermore, interventions are multimodal with a multiplicity of variables that might contribute to the effectiveness.

Trials that are poorly designed, conducted, or reported waste limited resources. To advance clinical knowledge and practice, the field should adhere to standards of research quality and set an agenda for research priorities. Given modest gains in survival, quality of life assessment should be integral to clinical trials and should adhere to recommended research methods, including handling of missing data.

Staging and Classification

SCLC is also known as “oat cell” carcinoma or small cell undifferentiated carcinoma (American Cancer Society, 2004). SCLC can be subtyped according to cellular classification as 1) small cell carcinoma; 2) mixed small cell/large cell carcinoma; or 3) combined small cell carcinoma (i.e., small cell lung cancer combined with neoplastic squamous and/or glandular components) (Physician Data Query, 2005).

Although the TNM classification scheme used for non-SCLC is applicable to SCLC staging (Cameron and Schwartz, 2005), most clinicians use a simplified two-stage scheme developed by the Veterans Administration Lung Cancer Study Group (Simon and Wagner, 2003; Physician Data Query, 2005). Limited-stage SCLC (approximately 30 percent of patients at diagnosis) includes those with tumor confined to the hemithorax of origin, the mediastinum, or the supraclavicular lymph nodes (Simon and Wagner, 2003; Physicians Data Query, 2005). In extensive-stage SCLC, tumor has spread outside these limits; patients with distant metastases are always considered to have extensive disease (Physician Data Query, 2005). At the time of diagnosis, 60–65 percent of SCLC patients have extensive disease (Osterlind, 2001). Estimates of median survival with current therapies are 16–24 months for those with limited-stage disease, and 6–12 months for those with extensive-stage disease (Physician Data Query, 2005).

Diagnostic procedures commonly used to establish the presence of distant metastases include bone marrow aspiration, brain scans using computed tomography (CT) or magnetic resonance imaging, chest and abdomen scans using CT, and radionuclide bone scans (Physician Data Query, 2005; Murren, Turrisi, and Pass, 2005). Whether positron emission tomography (PET) metabolic scanning using 18-fluorodeoxyglucose (18-FDG) provides any additional information to current staging techniques is uncertain (Murren, Turrisi, and Pass, 2005; Simon and Wagner, 2003).

Treatment Strategies

Treatments for SCLC are selected by stage and other features of disease extent (Physician Data Query, 2005). Few patients with extensive SCLC currently attain long-term survival. Their survival at 2 years after diagnosis is approximately 5 percent and at 5 years is less than 1 percent (Murren, Turrisi, and Pass, 2005).

Over time, there has been better success in the management of patients with limited disease. The proportion of long-term survivors among these patients has doubled from the 1970s to the 1990s (Janne, Freidlin, Saxman, et al., 2002; Murren, Turrisi, and Pass, 2005). While this may be due in part to stage migration, it is probably more associated with the change in practice of using platinum-based, rather than cyclophosphamide-based, combination chemotherapy regimens (Murren, Turrisi, and Pass, 2005). Attempts to improve on those results, either by adding a third drug or by substituting newer drugs have not yielded more long-term survivors thus far. It appears that further improvement requires both more and more complete responses to primary therapy (i.e., chemotherapy and radiation). Absent that, other interventions seem to largely alter the pattern of relapse, but not overall survival.

Chemotherapy. Chemotherapy is used for most patients, either as adjuvant therapy for the few patients eligible for surgery, or as primary therapy for patients with inoperable tumors. Preferred regimens have evolved over time (Murren, Turrisi, and Pass, 2005). Current guidelines recommend platinum-etoposide combinations in patients with limited-stage disease and platinum-based regimens in patients with extensive-stage disease (Simon and Wagner, 2003; Osterlind, 2001). According to the 2003 ACCP guidelines, there is no evidence on the benefit of maintenance chemotherapy in any patient achieving a partial or complete remission, and maintenance therapy is not recommended outside of a clinical trial (Simon and Wagner, 2003).

Surgery. Surgery is usually limited to patients with smaller tumors (T1 or T2) and no evidence of nodal involvement or spread outside the hemithorax of origin (Physician Data Query, 2005). Whether surgery added to chemotherapy for patients with limited-stage disease improves survival is currently uncertain.

Thoracic Radiotherapy. Meta-analyses published in the 1990s demonstrated the benefit of adding thoracic radiotherapy (TRTx) to chemotherapy in patients with limited-stage disease (Warde and Payne, 1992; Pignon, Arrigada, Ihde, et al., 1992). Addition of TRTx to chemotherapy increased 2- to 3-year overall survival by an absolute 5.4 percent over chemotherapy alone (Warde and Payne, 1992; Pignon, Arrigada, Ihde, et al., 1992; Carney, 1999). Addition of TRTx to chemotherapy in patients with limited-stage SCLC is now the recommended course of therapy (Simon and Wagner, 2003). However, uncertainties remain with respect to optimal timing, sequencing, and radiation regimens (i.e., dosages and fractionation schemes) (Turrisi, 1994; Osterlind, 2001). Table 1 summarizes factors that might influence how chemotherapy and radiation may interact when used for primary treatment of limited stage SCLC.

Meta-analyses using different study inclusion criteria have addressed the timing of TRTx given with chemotherapy for limited-stage SCLC. Cancer Care Ontario (2003) included 5-year survival data for 4 studies involving 777 patients, finding no difference between early and late TRTx. Huncharek and McGarry compared the impact of early (i.e., given with the first or second course of systemic therapy) versus delayed (i.e., with the final courses) TRTx in patents with limited disease (Huncharek and McGarry, 2004). The analysis pooled data from 8 randomized, controlled trials enrolling over 1,500 patients and found that early, concurrent TRTx (i.e., administered during the same time period as chemotherapy) improved 1, 2, and 3-year overall survival relative to delayed TRTx, and that TRTx with etoposide/cisplatin regimens performed better compared with non-etoposide/cisplatin regimens. This meta-analysis was flawed by double-counting data from one study (i.e., Goto, Nishiwaki, Takada, et al. 1999 and Takada, Fukuoka, Kawahara, et al., 2002).

A meta-analysis by the Cochrane Collaboration (Pijls-Johannesma, De Ruysscher, Lambin, et al. 2004), included 7 studies, 6 of which overlapped with those in the Huncharek and McGarry meta-analysis, and found that the 2–3 year survival difference as a function of timing was less certain. The Cochrane meta-analysis identified patient selection issues and differences in systemic regimens as potential confounders. Fried, Morris, Poole, et al. (2004) included 7 studies with 1,500 patients and found that 2-year survival was significantly improved by early TRTx, but the pooled result was not significant at 3 years. Two-year subgroup analysis showed that using hyperfractionation and platinum chemotherapy were associated with significant advantages favoring early TRTx, but significant results were not obtained in studies using conventional fractionation and non-platinum chemotherapy.

The role of radiation therapy in extensive disease is less established than in patients with limited-stage disease (Murren, Turrisi, and Pass, 2005). Several large studies reported in the 1980s by the Southwest Oncology Group (SWOG) and that did not randomize patients to TRTx versus no TRTx, suggested that, although thoracic radiation reduced initial relapse at the primary tumor site, there was no effect on overall survival (Murren, Turrisi, and Pass, 2005; Livingston, Mira, Chen, et al., 1984; Livingston, Schulman, Mira, et al., 1986).

Prophylactic Cranial Irradiation. The frequency of brain metastasis in SCLC patients led to the hypothesis that subclinical metastases are commonly present in the brain at diagnosis. Thus, clinicians often add prophylactic cranial irradiation (PCI), particularly for patients achieving a complete remission (CR) after primary therapy. Without PCI, patients who achieve an extracranial CR have a 50–80 percent actuarial risk of developing CNS metastases within 2–3 years (Simon and Wagner, 2003; Murren, Turrisi, and Pass, 2005; Carney, 1999). In addition, among patients who achieve a CR with chemotherapy, approximately 15 percent have brain metastases as the initial or only manifestation of recurrence (Carney, 1999). A patient-level meta-analysis of almost 1,000 patients in complete remission from 7 randomized, controlled trials showed the addition of PCI can reduce the risk of CNS metastases by over half and significantly improves survival (Auperin, Arriagada, Pignon, et al. 1999; Prophylactic Cranial Irradiation Overview Collaborative Group, 2000; Carney, 1999).

Definitive recommendations regarding optimal timing of PCI and radiation dosage issues (e.g., optimizing dose to balance efficacy and toxicity, fractionation) still require additional study (Prophylactic Cranial Irradiation Overview Collaborative Group, 2000; Boher and Wenz, 2002). According to one of the PCI meta-analyses, “Establishing the optimal dose and timing of treatment so as to reduce further the incidence of brain metastases with minimal and acceptable toxicity should be the aim of future clinical trials” (Prophylactic Cranial Irradiation Overview Collaborative Group, 2000).

Data on the adverse effects of PCI, both acute (e.g., skin burns, headaches) and late-developing (e.g., neurocognitive impairment, overt cerebral necrosis) are also not well characterized from analyses of controlled trials (Boher and Wenz, 2002). Although many retrospective studies describe an association between PCI and neurotoxicity, evidence from prospective, controlled trials does not appear to support that association (Bohrer and Wenz, 2002).

Second-Line Therapy. Most patients respond to primary therapy, but relapse after remissions of varying duration (Murren, Turrisi, and Pass, 2005). Second-line therapy is offered to most patients if the first remission has lasted 3–6 months; relapse after 3 months or more is also known as “sensitive relapse” (Murren, Turrisi, and Pass, 2005). Evidence of benefit is lacking from second-line therapy for refractory SCLC (i.e., no remission after primary therapy). Response to second-line therapy appears to be related to the chemotherapy agents given in both the induction and second-line regimens (Murren, Turrisi, and Pass, 2005). It is also unknown whether third or subsequent lines of therapy for relapsed or progressive SCLC improve outcomes compared with best supportive care.

Types of Studies

All questions, except Question 6, addressed therapeutic interventions. We sought randomized, controlled trials that compared the interventions of interest. No minimum number of patients per study arm was required for randomized, controlled trials. Because there were few randomized, controlled trials available to address Questions 8 and 9, we sought additional studies. For Question 8 (surgery), we also sought nonrandomized comparative trials, both prospective and retrospective in design. For Question 9 (second- or subsequent-line therapy), we also sought phase II multicenter trials reporting on at least 25 patients.

Question 6 (PET for staging) addresses a diagnostic intervention. Although we sought randomized, controlled trials comparing the outcomes of SCLC patients staged with and without use of PET, no such studies were identified. We then sought prospective, single-arm trials that reported on at least 25 patients undergoing imaging to stage SCLC; correlated 18-fluorodeoxyyglucose (FDG) PET findings with findings from other imaging modalities and an appropriate reference standard; and permitted computation of sensitivity and specificity.

Our search and selection criteria included English-language studies, as well as foreign-language studies that had an English-language abstract.

Studies were excluded if no outcome of interest to this review was reported. Studies were also excluded if the patient population of interest was fewer than 80 percent of included patients, or, alternatively, results for the patient population of interest were not separately reported. When multiple reports were available for the same study, it was counted as a single trial and outcome data from the report with the longest follow-up were used.

Types of Participants

• Key Questions 1–3 (First-line chemotherapy with thoracic radiotherapy [TRTx]) — patients with a histopathologically confirmed diagnosis of SCLC staged as limited disease.

• Key Question 4 (thoracic radiation therapy) — Patients with a histopathologically confirmed diagnosis of SCLC staged as extensive disease undergoing first-line therapy.

• Key Question 5 (prophylactic cranial irradiation) — Patients with a histopathologically confirmed diagnosis of SCLC that has completely responded to primary therapy (regardless of stage).

• Key Question 6 (PET staging) — Patients with a histopathologically confirmed diagnosis of SCLC.

• Key Question 7 (management mixed disease) — Patients with a histopathologically confirmed diagnosis of mixed small cell/non-small cell lung cancer.

• Key Question 8 (surgery) — Patients with a histopathologically confirmed diagnosis of SCLC staged as limited disease with small tumors and no nodal involvement

• Key Question 9 (second- or subsequent-line therapy) — Patients with a histopathologically confirmed diagnosis of SCLC that either relapsed or progressed after a response that lasted at least 3 months following primary therapy for: (a) limited-stage or (b) extensive-stage disease; or (c) patients with refractory disease (defined as no response or progression within 3 months of primary therapy).

Types of Interventions

• Key Question 1 — Comparison of chemotherapy combined with sequential TRTx, chemotherapy combined with concurrent TRTx and chemotherapy combined with alternating TRTx.

• Key Question 2 — Chemotherapy combined with concurrent TRTx initiated early cycles (i.e., 1 or 2) versus chemotherapy combined with concurrent TRTx initiated in late cycles (i.e., 3 or later).

• Key Question 3 — Chemotherapy combined with standard-interval TRTx versus chemotherapy combined with accelerated TRTx: OR chemotherapy combined with split-course TRTx chemotherapy combined with standard-interval TRTx; OR chemotherapy combined with single daily fractions of TRTx; OR chemotherapy combined with hyperfractionated TRTx.

• Key Question 4 — Chemotherapy combined with TRTx versus chemotherapy alone.

• Key Question 5 — Prophylactic cranial irradiation (PCI) versus no prophylactic radiation after primary therapy is completed and response is assessed.

• Key Question 6 — Positron-emission tomography (PET) vs. no PET, added to other staging modalities, including computed tomography (CT) and magnetic resonance imaging (MRI).

• Key Question 7 — Chemotherapy with or without TRTx delivered in any sequence or schedule used for limited-stage SCLC

• Key Question 8 — Surgical excision of SCLC tumors, preceded by neoadjuvant chemotherapy or followed by adjuvant chemotherapy, and either with or without TRTx and PCI, versus no surgical excision

• Key Question 9 — Chemotherapy using drugs approved by the U.S. Food and Drug Administration for at least one indication to treat a malignant disease (various regimens).

Types of Outcomes

Primary (health) outcomes of interest include:

• duration of survival, disease-free survival, and/or progression-free survival

• quality of life

• brain metastasis-free survival and subsequent treatment(s) for brain metastasis

• palliation of measurable symptoms

• treatment-related adverse events

• perioperative adverse events

Secondary (intermediate) outcomes include:

• objective response rates (complete and partial responses; separately and summed)

• response durations

• pathologically complete resection rates

• recurrence rates

For key question 6 (PET staging) additional outcomes of interest are:

• diagnostic accuracy

• outcomes other than diagnostic accuracy, such as staging accuracy, change in stage and impact on management decisions

Search Strategy

Electronic databases. The following databases were searched for citations. The full search strategy is displayed in Appendix A. ^* The search was not limited to English-language references, but foreign-language references without abstracts were disregarded.

• MEDLINE® (through 12/21/04)

• EMBASE (through 03/04/05)

• Cochrane Controlled Trials Register (through 03/11/05)

Additional Sources of Evidence. The Technical Expert Panel and individuals and organizations providing peer review were asked to inform the project team of any studies relevant to the key questions that were not included in the draft list of selected studies.

Search Screen

Search results were stored in a ProCite® database. Using the study selection criteria for screening titles and abstracts, a single reviewer marked each citation as either: (1) eligible for review as full-text articles; (2) ineligible for full-text review; or (3) uncertain. Citations marked as uncertain were reviewed by a second reviewer and resolved by consensus opinion, with a third reviewer to be consulted if necessary. Using the final study selection criteria, review of full-text articles was conducted in the same fashion to determine inclusion in the systematic review. A total of 630 references were retrieved at a full-text level; 89 were included in this review (Figure 1). Records of the reason for exclusion for each paper retrieved in full-text, but excluded from the review, were kept in the ProCite® database (see Appendix D, Excluded Studies).

Data Elements

The data elements below were abstracted, or recorded as not reported, from therapeutic intervention studies.

• critical features of the study design (for example, patient inclusion/exclusion criteria, number of subjects, use of blinding);

• potential patient characteristic confounders:

  • age

  • gender

  • race

  • extent of disease and stage

  • performance status

  • comorbidities

• treatment protocols (for example, treatment intensity, frequency, duration, other prior and concurrent treatment factors);

• patient monitoring procedures (for example, follow-up duration and frequency, outcome assessment methods); and

• the specified key outcomes and data analysis method (when statistical test results were lacking for adverse events data, reviewers performed tests with the STATMAN statistical program).

The data elements below were abstracted, or recorded as not reported, from diagnostic accuracy studies of imaging modalities used in staging SCLC:

• patient selection criteria

• details about the reference standard (validity and degree of detail in description)

• decision rules for determining which patients received the reference standard

• whether the index test and reference standard were interpreted blind to each other

• whether verification bias (index test results influenced decisions to perform reference standard) was avoided

• details about the index test (degree of detail about performing of test, interpretation)

• study design (prospective, retrospective)

• reporting of diagnostic accuracy results (completeness, appropriate calculation of accuracy measures, use of confidence intervals)

• outcomes other than diagnostic accuracy, such as staging accuracy, change in stage and impact on management decisions

Evidence Tables

Templates for evidence tables were created in Microsoft Excel® and Microsoft Word® Appendix B).^* One reviewer performed primary data abstraction of all data elements into the evidence tables, and a second reviewer reviewed the evidence tables for accuracy. Disagreements were resolved by discussion, and if necessary, by consultation with a third reviewer. When small differences occurred in quantitative estimates of data from published figures, the values obtained by the two reviewers were averaged.

Therapeutic Studies

The general approach to grading evidence developed by the U.S. Preventive Services Task Force (Harris et al. 2001) was applied. Quality of the abstracted studies was assessed by one reviewer and fact-checked by a second. Discordant quality assessments were resolved by discussion or by consultation with a third reviewer, if necessary. The quality criteria for randomized, controlled trials were as follows:

• Initial assembly of comparable groups: adequate randomization, including concealment and whether potential confounders (e.g., baseline characteristics, other concomitant care) were distributed equally among groups

• Maintenance of comparable groups (includes attrition, crossovers, adherence, contamination)

• Important differential loss to follow-up or overall high loss to follow-up

• Measurements: equal, reliable, and valid (includes masking of outcome assessment)

• Clear definition of interventions

• All important outcomes considered

• Analysis: adjustment for potential confounders, intention-to-treat analysis

Diagnostic Studies

The Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool underwent a rigorous development process by Whiting, Rutjes, Dinnes, et al. (2004) and includes the following items:

• Was the spectrum of patients representative of the patients who will receive the test in practice?

• Were selection criteria clearly described?

• Is the reference standard likely to classify the target condition correctly?

• Is the period between reference standard and index test short enough to be reasonably sure that the target condition did not change between the two tests?

• Did the whole sample or a random selection of the sample, receive verification using a reference standard of diagnosis?

• Did patients receive the same reference standard regardless of the index test result?

• Was the reference standard independent of the index test (i.e. the index test did not form part of the reference standard)?

• Was the execution of the index test described in sufficient detail to permit replication of the test?

• Was the execution of the reference standard described in sufficient detail to permit replication of the reference standard?

• Were the index test results interpreted without knowledge of the results of the reference standard?

• Were the reference standard results interpreted without knowledge of the results of the index test?

• Were the same clinical data available when test results were interpreted as would be available when the test is used in practice?

• Were uninterpretable/intermediate test results reported?

• Were withdrawals from the study explained?

Definition of Ratings Based on Criteria

The rating of therapeutic intervention studies encompasses the 3 quality categories described below. No analogous quality categories have been incorporated into the QUADAS tool for assessing diagnostic accuracy studies. Rather, each of the 14 QUADAS items is considered individually.

Good:Meets all criteria: Comparable groups are assembled initially and maintained throughout the study (follow-up at least 80 percent); reliable and valid measurement instruments are used and applied equally to the groups; interventions are spelled out clearly; all important outcomes are considered; and appropriate attention to confounders in analysis. In addition, for randomized, controlled trials (RCTs), intention to treat analysis (i.e., all patients randomized were analyzed) is used.

Fair: Studies will be graded “fair” if any or all of the following problems occur, without the fatal flaws noted in the “poor” category below: Generally comparable groups are assembled initially but some question remains whether some (although not major) differences occurred with follow-up; measurement instruments are acceptable (although not the best) and generally applied equally; some but not all important outcomes are considered; and some but not all potential confounders are accounted for. Intention-to-treat analysis is done for RCTs.

Poor: Studies will be graded “poor” if any of the following fatal flaws exists: Groups assembled initially are not close to being comparable or maintained throughout the study; unreliable or invalid measurement instruments are used or not applied at all equally among groups (including not masking outcome assessment); and key confounders are given little or no attention. For RCTs, intention-to-treat analysis is lacking.

Overview

As summarized in Summary Table 2, 6 randomized, controlled trials (RCTs) made comparisons of alternating, concurrent and sequential TRTx for limited stage SCLC. Two trials (n=307) compared concurrent and sequential TRTx (Takada, Fukuoka, Kawahara, et al., 2002; Park, Kim, Jeong, et al., 1996). Two trials compared alternating to sequential TRTx (Gregor, Drings, Burghouts, et al., 1997; Sun, Zhang, Yin, et al., 1995; n=458). One trial compared alternating to concurrent TRTx (Lebeau, Urban, Brechot, et al., 1999, n=156). The Work and colleagues trial (Work, Nielsen, Bentzen, et al., 1997/Work, Bentzen, Nielsen, et al., 1996) compared early alternating and late alternating TRTx (n=199). Collectively, the 6 trials randomized 228 patients to concurrent treatment, 337 patients to alternating treatment, and 385 patients to sequential treatment. It is worth noting that these studies were generally small in size and likely underpowered to find small but clinically significant differences in survival.

Study populations and treatment protocols are summarized in Summary Table 3. Additional details are in Appendix Tables 1A–D, 1H.^* Information in the tables came exclusively from articles except for the Park, Kim, Jeong, et al. (1996) study. Park, Kim, Jeong, et al. (1996) did not report survival probabilities at yearly intervals, so an author was contacted directly and additional data were sought. The data obtained from the author represented a larger patient sample than described in the article.

Concurrent vs. Sequential

Interventions. Two trials compared concurrent and sequential TRTx. Radiation dose in the Takada, Fukuoka, Kawahara, et al. (2002) study was 45 Gy, while it varied between 40 and 50 Gy in the Park, Kim, Jeong, et al. (1996) study. Both studies gave concurrent TRTx in weeks 1–3. Sequential TRTx occurred in weeks 13–15 in the Takada, Fukuoka, Kawahara, et al. (2002) trial and between weeks 19 and 24 in the Park, Kim, Jeong, et al. (1996) study. Takada, Fukuoka, Kawahara, et al. (2002) delivered 2 daily fractions of TRTx in both groups, while Park, Kim, Jeong, et al. (1996) gave it to the concurrent group. Both studies gave prophylactic cranial irradiation (PCI). Platinum-based chemotherapy was used by Park, Kim, Jeong, et al. (1996), but not by Takada, Fukuoka, Kawahara, et al. (2002).

Populations. Groups were well-balanced on age, gender and performance status in the Takada, Fukuoka, Kawahara, et al. (2002) study (n=228). The sample of 51 patients in the Park, Kim, Jeong, et al. (1996) article was also well-balanced on these characteristics, but the survival data represented 79 patients and no comparison of baseline characteristics is available for all.

Quality and Reporting. The Takada, Fukuoka, Kawahara, et al. (2002) trial was rated as being of good quality. The Park, Kim, Jeong, et al. (1996) study was rated as poor due to insufficient information about assembly and maintenance of comparable groups, in addition to uncertainty about full accounting of subjects in data analysis.

Results. Survival outcomes are shown in Summary Table 4 and adverse events in Summary Table 5. More detailed results are in Appendix Tables 1E–1G.^* Both studies showed survival results favoring concurrent TRTx, but were generally not statistically significant. Unadjusted overall survival did not differ significantly between concurrent and sequential TRTx, although p values were nearly significant. Overall median survival favored concurrent therapy by 5.1 months (Park, Kim, Jeong, et al., 1996) and 7.5 months (Takada, Fukuoka, Kawahara, et al., 2002). A Cox proportional hazards model regression found that treatment was a significant predictor of survival, producing a hazard ratio of 0.70 (95 percent confidence interval [CI]: 0.52–0.94) for concurrent relative to sequential TRTx. Takada, Fukuoka, Kawahara, et al. (2002) also reported that median progression-free survival favored the concurrent group by 2 months (p=0.084), but Park, Kim, Jeong, et al. (1996) did not report on progression.

Neither study reported on quality of life, but both reported tumor response data. Both found nonsignificantly higher overall response rates (ORRs) in the concurrent group, although the Park, Kim, Jeong, et al. (1996) study found a fairly large difference in rates that approached significance. In the Takada, Fukuoka, Kawahara, et al., 2002) study, ORRs were 96.5 percent; for concurrent and 92.1 percent for sequential (p=0.25). The complete responses (CRs) were higher in the concurrent group (39.5 percent) than in the sequential group (27.2 percent, p=0.07). In the Park, Kim, Jeong, et al. (1996) trial, the concurrent group achieved an ORR of 88 percent, versus 63 percent for sequential (p=0.13). Mean response duration was longer in the sequential group than in the concurrent group (395 days vs. 180 days, p=0.03).

Among 12 categories of adverse events, 5 were reported by both studies (Summary Table 6). Significant between-group differences were not found in either trial for anemia, thrombocytopenia, infection and fever. Both studies found significantly higher risks of leukopenia for those in the concurrent arm. In the Takada, Fukuoka, Kawahara, et al. (2002) study, grade 3 or 4 leukopenia was seen in 88.4 percent of concurrent-arm patients and in 53.6 percent of sequential-arm patients (p=0.001). The risk of higher grade leukopenia among concurrent TRTx patients in the Park, Kim, Jeong, et al. (1996) study was 51.8 percent, compared with 16.7 percent of sequential TRTx patients (p=0.02). One study reported data on each of 7 adverse events, none of which was marked by significant differences between concurrent and sequential TRTx: treatment-related mortality, nausea/vomiting, esophagitis, renal toxicity; alopecia, arrhythmias, and hepatic toxicity.

Summary. These 2 studies suggest better efficacy outcomes for concurrent TRTx than for sequential TRTx, with inconsistent statistical significance, along with similar rates of adverse events of all types except leukopenia, which was more common for concurrent TRTx. Unadjusted analyses of overall survival found nearly significant differences favoring concurrent over sequential TRTx in 2 studies. One study using adjustment by Cox regression found a significant treatment effect for concurrent TRTx. One study that analyzed progression-free survival reported a nearly significant difference in favor of concurrent TRTx. CRs were more common with concurrent therapy in both studies, but not significantly so (p values were 0.07 and 0.13). One study found a significantly longer response duration for concurrent TRTx. Only 1 of 11 types of adverse events showed significant between-group differences. Leukopenia was more common for concurrent TRTx in both studies.

Alternating vs. Sequential

Interventions. Both Sun, Zhang, Yin, et al. (1995) and Gregor, Drings, Burghouts, et al. (1997) delivered TRTx in single fractions per day. There was a wide range of total doses in the Sun, Zhang, Yin, et al. (1995) study (30–60 Gy), while the Gregor, Drings, Burghouts, et al. (1997) study gave 50 Gy to all patients. Alternating TRTx was given between weeks 4 and 9 in the Sun, Zhang, Yin, et al. (1995) study, whereas Gregor, Drings, Burghouts, et al. (1997) administered it every 4 weeks between 7 and 20 weeks. In the Gregor, Drings, Burghouts, et al. (1997) study, 4 weeks of TRTx in the alternating arm was given over a period of 13 weeks wherease sequential TRTx was given over 4 consecutive weeks. Sun, Zhang, Yin, et al. (1995) provided TRTx over 6 consective weeks in both the alternating and sequential arms. That study also used platinum-based chemotherapy in the later period of the trial, but no patients received it in the Gregor, Drings, Burghouts, et al. (1997) study. Neither report made clear whether patients received PCI.

Populations. The Sun, Zhang, Yin, et al. (1995) article did not report any baseline patient characteristics; it simply stated that 123 patients had localized disease. Patient groups in the Gregor, Drings, Burghouts, et al. (1997) study (n=335) were well-matched on the 3 key characteristics: age, gender and performance status.

Quality and Reporting. Gregor, Drings, Burghouts, et al. (1997) received a good study quality rating. Sun, Zhang, Yin, et al. (1995) was rated as poor because details were lacking for all quality domains.

Results. Gregor, Drings, Burghouts, et al. (1997) did not find a statistically significant difference between groups in adjusted survival. Median survival was 15 months in the sequential group and 14 months in the alternating group. The entire survival curve for the sequential TRTx group was slightly higher than that of the alternating group. Between 1 and 4 years, survival probabilities differed by 4 percent or less, while the difference was 8 percent at 5 years. In the Sun, Zhang, Yin, et al. (1995) study, statistical test results for survival were missing. At 1 year, the survival probability was higher in the sequential group by 1.5 percent, whereas at 2 an 3 years, it was higher for the alternating group by 14.4 percent and 4 percent. Relative risks (RR) for death at 2 years and 3 years were computed for purposes of meta-analysis. At 2 years, the RR of 0.831 significantly favors alternating TRTx (95 percent CI: 0.692–0.999). The difference is smaller and in the same direction at 3 years, with an RR of 0.957, but nonsignificant (95 percent CI: 0.831–1.102). The difference in progression-free survival favoring sequential TRTx in the Gregor, Drings, Burghouts, et al. (1997) study approached statistical significance (p=0.07). Neither study reported on tumor response or quality of life.

Sun, Zhang, Yin, et al. (1995) reported no data on adverse events (Summary Table 7), while Gregor, Drings, Burghouts, et al. (1997) gave data on 6 types. There were no between-group differences in the incidence of nausea/vomiting, acute esophagitis, or late pulmonary fibrosis. Late esophagitis was significantly less frequent in the alternating group, compared to the sequential group (p=0.01). Both thrombocytopenia and leukopenia were more common (p<0.001) in the alternating group.

Summary. Results are mixed on the relative impact on outcomes for alternating and sequential TRTx. One study reported that the survival curve for sequential TRTx was always higher than that for alternating TRTx, but the difference was not significant. The other study showed a significant difference in the RR of death at 2 years favoring alternating TRTx. The study reporting progression-free survival found a nearly significant advantage for sequential TRTx. Late esophagitis was more common for the sequential group, but thrombocytopenia and leukopenia were more frequent in the alternating group. These data do not show a clear advantage for either sequential or alternating TRTx.

Alternating vs. Concurrent

Interventions. The Lebeau, Urban, Brechot, et al. (1999) study delivered doses of 55 Gy to the alternating TRTx group and 50 Gy to the concurrent TRTx group.^* Radiation was given in once daily fractions to both groups. Concurrent TRTx was offered across 5 weeks from week 5 through 9, while alternating TRTx occurred across 11 weeks during weeks 6–7, 10–11 and 14–16. Both groups received PCI. Non-platinum chemotherapy was administered.

Populations. The 2 groups of patients in this study (n=156) were well-matched on baseline characteristics.

Quality and Reporting. This trial received a good study quality rating.

Results. The entire survival curve for alternating TRTx lies slightly above that for concurrent TRTx, but the difference overall was not significant. Differences in survival probabilities ranged from a high at 1 year of 9 percent to a low of 2 percent at 5 years. Progression-free survival and quality of life was not reported. There was no statistically significant difference between groups in tumor response rates.

Four types of adverse events (Summary Table 8) were noted by Lebeau, Urban, Brechot, et al. (1999). The only outcome that showed a statistically significant between-group difference was deaths from pulmonary fibrosis, which were more common in the concurrent TRTx group (p=0.05).

Summary. The single study comparing alternating and concurrent TRTx does not suggest a meaningful improvement in survival associated with alternating TRTx. Overall survival did not differ significantly, with a difference between medians of only 0.5 months favoring alternating TRTx. Deaths from pulmonary fibrosis were more frequent in the concurrent TRTx group.

Early Alternating vs. Late Alternating

Interventions. The dose given to both groups in the early phase of the Work and colleagues study (Work, Nielsen, Bentzen, et al., 1997/Work, Bentzen, Nielsen, et al., 1996) was 40 Gy; it was increased later to 45 Gy. Radiation was delivered as a single daily fraction in both treatment arms. TRTx was given during weeks 1–2 and 6–7 in the early-alternating group and in weeks 18–19 and 23–24 in the late-alternating group. Given the somewhat lower total dose in this study compared with other studies addressed above and administration in split-course fashion, TRTx was given at a relatively low dose rate. Both groups received PCI. The chemotherapy regimen for all patients was platinum-based; however the regimen was given in an unusual schedule and the doses of drugs actually delivered is unclear..

Populations. Groups receiving early and late alternating TRTx were well-balanced on baseline patient characteristics.

Quality and Reporting. This study was rated as fair in quality. The key deficiency was an inadequate description of the randomization method.

Results. Work and colleagues (Work, Nielsen, Bentzen, et al., 1997/Work, Bentzen, Nielsen, et al., 1996) reported that median survival was slightly longer in the late-alternating group (1.5 months), while 2- and 3-year survival probabilities were slightly higher in the early-alternating group. Overall, there was no significant difference in survival between groups. There was an 18 percent difference at 1 year in percentage without in-field recurrence (PWIFR) favoring late-alternating TRTx, but differences at later times were much smaller and the groups did not differ significantly. Tumor response rates did not differ for the 2 patient groups. No quality of life data were collected.

Of the 6 categories of adverse events, only leukopenia showed a difference between groups (Summary Table 9). This outcome was significantly more common among those receiving early alternating TRTx.

Summary. The single study comparing early and late alternating is does not support conclusions about the relative effectiveness of these approaches to TRTx. There was no significant difference between groups on overall survival, percentage without in-field recurrence and tumor response. Of 6 types of adverse events reported, groups differed only in the frequency of leukopenia, which was significantly higher among those receiving early alternating TRTx.

Conclusions

Among 6 studies meeting selection criteria for Key Question 1, two trials (n=307) compared concurrent and sequential TRTx. Two trials compared alternating to sequential TRTx (n=458). One trial compared alternating to concurrent TRTx (n=156). The final trial compared early alternating and late alternating TRTx (n=199). Comparing these trials with others addressing TRTx delivery, survival is generally lower is this set relative to studies assessing the effect of hyperfractionation. Although an explanation is not readily apparent, possible reasons include patient selection, stage drift and the adequacy of chemotherapy.

Concurrent vs. Sequential. Results are not conclusive but suggest better outcomes for concurrent TRTx. Although not statistically significant, unadjusted overall survival and CR rates favored concurrent TRTx in both studies. However, adjusted overall survival in the larger study was significantly in favor of concurrent TRTx. A smaller study found significantly longer response duration for concurrent TRTx in 1 study. Out of 11 types of adverse events, only leukopenia occurred significantly more frequently, in the concurrent TRTx group in both studies.

Alternating vs. Sequential. Inconsistent findings were observed in the 2 studies and no conclusions can be drawn that one is superior to the other. The direction of the advantage on overall survival differed in the 2 studies.

Alternating vs. Concurrent. In the single study comparing alternating and concurrent TRTx, there was no statistically significant effect on survival and no conclusions of differential efficacy could be drawn.

Early Alternating vs. Late Alternating. In the single study comparing early versus late alternating TRTx, there was no statistically significant difference in survival, thus no conclusions of differences in efficacy can be reached.

Overview

Six randomized, controlled trials (RCTs) compared outcomes of alternate times to administer TRTx concurrently in first-line therapy for limited stage SCLC (N=1,177). Summary Table 10 summarizes selected study variables; further details are in Summary Table 11 and Appendix Tables 2A–C, 2H.^* Each of the three larger trials randomized from 125 to 166 patients per arm (Murray, Coy, Pater, et al., 1993/Coy, Hodson, Murray, et al., 1994/Feld, Payne, Hodson, et al., 1988 [hereafter referred to as “Murray-Coy-Feld”]; Perry, Eaton, Propert, et al., 1987/Ahles, Silberfarb, Rundle, et al., 1994/Perry, Herndon, Eaton, et al., 1988; [hereafter referred to as “Perry-Ahles-Perry”]; James, Spiro, O'Donnell, et al., 2003). Together, the three smaller trials included less than one-fourth of all patients studied (Jeremic Shibamoto, Acimovic, et al., 1997; Qiao, Zhou, Xin, et al., 2004; Skarlos, Samantas, Briassoulis, et al., 2001).

Interventions. Available studies did not uniformly define early and late concurrent therapy, with respect to either the chemotherapy cycle or weeks during which they administered TRTx. Most trials (4 of 6) began TRTx in chemotherapy cycle 1 (i.e., week 1) for those randomized to early concurrent therapy; two waited until cycle 2 (week 4) (Murray-Coy-Feld; James, Spiro, O'Donnell, et al., 2003). Those randomized to late concurrent therapy began TRTx in cycle 3 (week 6) in one trial (Jeremic Shibamoto, Acimovic, et al., 1997), cycle 4 (week 10 or 12) in three trials (Perry-Ahles-Perry; Qiao, Zhou, Xin, et al., 2004; Skarlos, Samantas, Briassoulis, et al., 2001), and cycle 6 (week 16) in the remaining two trials (Murray-Coy-Feld, James, Spiro, O'Donnell, et al., 2003).

Five of six RCTs used platinum-etoposide chemotherapy regimens, including two of three larger trials; Perry-Ahles-Perry was the exception. Total TRTx dose was ≥40 Gy in each RCT, and only two used doses greater than 50 Gy (Jeremic Shibamoto, Acimovic, et al., 1997; Qiao, Zhou, Xin, et al., 2004). Three trials gave TRTx over a 3-week period (Murray-Coy-Feld; Skarlos, Samantas, Briassoulis, et al., 2001; James, Spiro, O'Donnell, et al., 2003 2003), one gave TRTx over four weeks (Jeremic Shibamoto, Acimovic, et al., 1997), and two gave TRTX over five or six weeks (Perry-Ahles-Perry; Qiao, Zhou, Xin, et al., 2004). Thus, weekly doses were 10 Gy in two trials (Perry-Ahles-Perry; Qiao, Zhou, Xin, et al., 2004), 13.35 Gy in two trials (Murray-Coy-Feld; James, Spiro, O'Donnell, et al., 2003), and 15 Gy in two trials (Jeremic Shibamoto, Acimovic, et al., 1997; Skarlos, Samantas, Briassoulis, et al., 2001). Four trials administered single daily fractions (Murray-Coy-Feld; Perry-Ahles-Perry; Qiao, Zhou, Xin, et al., 2004; James, Spiro, O'Donnell, et al., 2003) and two gave two fractions per day (Jeremic Shibamoto, Acimovic, et al., 1997; Skarlos, Samantas, Briassoulis, et al., 2001). Five of six trials included PCI for each arm; Qiao, Zhou, Xin, et al. (2004) did not report PCI use.

Study Populations. Most trials studied patients with relatively favorable baseline characteristics, and were nearly always well-balanced across arms for consistently-reported factors (Summary Table 11). In four of six trials, performance status (PS) was 0–1 at enrollment for 75 percent to 91 percent of patients across arms (Murray-Coy-Feld; Perry Ahles-Perry; Skarlos, Samantas, Briassoulis, et al., 2001; James, Spiro, O'Donnell, et al., 2003). PS also was well balanced across arms in Jeremic Shibamoto, Acimovic, et al. (1997), but many patients (~50 percent) had Karnofsky scores of 50–80. Qiao, Zhou, Xin, et al. (2004) excluded patients with Karnofsky PS <60, but did not report PS distribution by arm. For all six trials, the median or mean age ranged from approximately 55 to 62 years, and was balanced across arms. Each trial enrolled mostly men (8.6 percent women in one trial, 33 percent to 43 percent across five others), and had similar proportions of women in each arm.

Other prognostic factors and baseline characteristics were reported inconsistently (Appendix Table 2B ^*). Only three trials reported the proportion with weight loss at entry (Perry-Ahles-Perry; Jeremic Shibamoto, Acimovic, et al., 1997; Skarlos, Samantas, Briassoulis, et al., 2001). Only three trials reported the proportion with disease outside the lung (Murray-Coy-Feld, Qiao, Zhou, Xin, et al., 2004; Skarlos, Samantas, Briassoulis, et al., 2001). Only one trial reported the proportion of former smokers (Skarlos, Samantas, Briassoulis, et al., 2001). No trials reported racial distributions.

Study Quality and Reporting. The larger studies included one good-quality and one fair-quality multicenter trial, each published in full. The third large trial also was multicenter, but was reported only in abstract and information to rate quality was lacking. Two of three small trials were single-center and one was multicenter; each was of fair quality and published in full.

Results

In one large and two smaller trials, results significantly favored the early TRTx arms for overall (OS), progression-free (PFS), or local recurrence-free (LRFS) survival (Summary Table 12). Murray-Coy-Feld (n=308) reported significantly longer median OS (21.2 versus 16.0 months; p=0.008) and greater 2- and 3-year survival (40 percent versus 33.7 percent and 29.7 percent versus 21.5 percent, respectively) with early TRTx. Murray-Coy-Feld also reported significantly greater PFS with early TRTx (median, 15.4 versus 11.8 months; 26 percent versus 19 percent at 3 years; p=0.036). Qiao, Zhou, Xin, et al. (2004; n=90) reported longer median OS (26 versus 19 months; p<0.05) and greater 3-year survival (33 percent versus 22 percent) with early TRTx, but did not report an outcome related to progression or recurrence. Jeremic Shibamoto, Acimovic, et al. (1997; n=103) reported significantly greater 2- and 3-year LRFS with early TRTx (90 percent versus 69 percent and 73 percent versus 61 percent, respectively; p=0.011). While median OS (34 versus 26 months) and 2- and 3-year survival also favored early TRTx in the Jeremic Shibamoto, Acimovic, et al. (1997) trial, these results were just barely statistically nonsignificant (p=0.052).

Between-arm differences in response rates were not statistically significant in any trial (Appendix Table 2F).

In two large and one smaller RCTs, OS and time to treatment failure (TTF) did not differ significantly between arms randomized to early or late TRTx (Summary Table 12). Perry-Ahles-Perry (n=270), the only trial that did not use platinum, reported non-significant differences in OS (P=0.144) and TTF (p=0.238). James, Spiro, O'Donnell, et al. (2003; n=325), the sole trial published as an abstract, only reported OS and also found no significant difference (p=0.18). Skarlos, Samantas, Briassoulis, et al. (2001; n=81) reported nonsignificant differences for median OS (p=0.65) and median TTF (p=0.6).

A small pilot sub-study from one RCT reported the only data comparing quality of life outcomes after early versus late TRTx. Ahles et al. (1994) scored responses to measures of mood, psychosocial function, and cognitive function for 14–17 patients (of n=121) given early TRTx and 10–12 (of n=141) given late TRTx in the Perry-Ahles-Perry trial (Appendix Table 2F).^*

They compared these with scores for another group randomized to chemotherapy without TRTx (not abstracted). Results suggested larger decrements of mood and psychosocial function after chemotherapy plus TRTx than after chemotherapy alone. However, they found no meaningful differences in magnitude of decrement between early and late TRTx groups.

Table 13 shows that leukopenia/neutropenia and esophagitis were the only adverse events consistently reported by all six trials. Although leukopenia/neutropenia was more common in the early arm of five RCTs, only Qiao, Zhou, Xin, et al. (2004; p<0.05) and James, Spiro, O'Donnell, et al. (2003; p=0.006) reported that differences were statistically significant (Summary Table 14). Of four reporting RCTs, only Murray-Coy-Feld reported significantly more anemia in the early treatment arm (49 percent versus 37 percent, p=0.03). Skarlos, Samantas, Briassoulis, et al. (2001) reported significantly more grade 3 esophagitis with late than with early TRTx. However, the arms did not differ significantly when grades 1–3 were pooled, and the other five trials reported no significant differences in grade 3 or 3+4 combined.

Between-arm differences in treatment-related mortality (3 reporting trials), nausea/ vomiting (5 reporting trials), neurosensory effects (3 reporting trials), bronchopulmonary effects or pneumonitis (3 reporting trials each), thrombocytopenia (5 reporting trials), and infections (4 reporting trials) were not statistically significant.

Conclusions

Overall, the evidence is equivocal, either finding no difference or a small advantage for early TRTx. One larger trial of good quality significantly favored concurrent therapy given in an early cycle (Murray-Coy-Feld; median OS 21.2 versus 16.0 months; p=0.008), as did 2 smaller trials. Of the two larger trials that that found no significant difference, one did not use platinum chemotherapy and the other has not been published in full text. Meta-analysis on the question of early versus late TRTx was performed to attempt to obtain clearer results.

Leukopenia/neutropenia appeared to be more common with early concurrent TRTx, although differences were statistically significant in only two of six reporting trials. Other events do not appear to be more frequent with either early or late TRTx. However, evidence is limited as adverse events were not reported consistently across all trials.

Overview

All but one of the studies selected for Key Questions 1 and 2 can be viewed as comparing early and late TRTx. Key Question 2 is limited to studies in which both early and late TRTx were given concurrently with chemotherapy, while Key Question 1 included those with arms defined by TRTx given either concurrently, sequentially or in alternating fashion. Only the study by Lebeau, Urban, Brechot, et al. (1999) is excluded because only 1 week separated the start of TRTx in the study's 2 arms. Four previous meta-analyses have compared the impact of early and late TRTx, but none have included all 11 studies reviewed here. Three meta-analyses are summarized in Summary Table 25; a meta-analysis by Cancer Care Ontario (2003) is omitted because it used much more restrictive study selection criteria, including only 4 studies. The meta-analysis of 8 studies by Fried, Morris, Poole, et al. (2004) used the most rigorous methods and comprised the largest pool of the previous meta-analyses. The present meta-analysis addresses whether the findings of Fried, Morris, Poole, et al. (2004) can be reproduced in light of a larger study pool and different meta-analytic techniques.

Fried, Morris, Poole, et al. (2004) used the Mantel-Haenszel pooling method and found no significant heterogeneity at either 2 or 3 years; thus, fixed-effects models were employed. A significant increase in 2-year survival was found for early TRTx over late TRTx (RR: 1.17, 95 percent CI: 1.02–1.35). The effect was not significant at 3 years (RR: 1.13, 95 percent CI: 0.92–1.39). Subgroups of studies using hyperfractionation and platinum regimens had significant increases in 2- and 3-year survival favoring early TRTx, nonsignificant results were found for subgroups that did not use hyperfraction and platinum. Random effects meta-regression of risk differences (RDs) found that higher RDs were seen at both 2 and 3 years when studies used both hyperfractionation and platinum chemotherapy. Thus, larger effects of early over late TRTx were associated with combining hyperfractionation and platinum chemotherapy.

The present meta-analysis differs from that of Fried, Morris, Poole, et al. (2004) in the following ways: it included 3 additional studies; it used inverse variance weighting rather than the Mantel-Haenszel pooling method; and random effects meta-regression was carried out using RRs for this analysis and RDs by Fried, Morris, Poole, et al. (2004). In addition, Fried, Morris, Poole, et al. (2004) created 3 subgroups from the combination of hyperfractionation and platinum and used indicator variables for them in the meta-regression. This Review kept these variables separate. It could be argued that the hetereogeneity of comparisons across studies is too great to warrant pooling them. Like previous meta-analyses on this topic, we address this concern by using influence analysis, subgroup/sensitivity analysis, and meta-regression to investigate whether potential sources of hetereogeneity are associated with different results.

2-Year Mortality. The funnel plot in Figure 2 shows asymmetry in the lower right portion, suggestive of publication bias. The Egger regression test (Summary Table 15) reveals that the intercept differs significantly from zero. These results suggest the presence of publication bias.

Summary Table 16 and Figure 3 show 2-year RRs for each individual trial, along with 95 percent confidence intervals (CIs). It should be noted that all RRs were computed based on data from articles for all studies except Park, Kim, Jeong, et al. (1996). The Park, Kim, Jeong, et al. (1996) article did not give survival probabilities at yearly periods so an author was contacted who provided data for a larger sample than was described in the original articles. The Sun, Zhang, Yin, et al. (1995) and Takada, Fukuoka, Kawahara, et al. (2002) trials both obtained 2-year RRs showing a significant reduction in the risk of mortality for early TRTx. One study (Perry, Herndon, Eaton, et al. 1998) found a slight nonsignificant increase in mortality for early TRTx and the remaining 6 studies yielded nonsignificant decreases in mortality for early TRTx.

The Q statistic value obtained here (Summary Table 17) exceeds the threshold for concluding that significant heterogeneity of treatment effects exists, therefore a random effects pooled estimate was computed (see forest plot in Figure 3). The pooled RR is 0.921 and the 95 percent CI overlaps the null value of 1.0 (0.844, 1.005). Figure 4 presents the results of influence analysis, in which each individual study is excluded from the random effects pooled estimate. This graph can be interpreted by finding the studies that depart to the greatest extent from the vertical line for the full pooled estimate RR of 0.92. When the Perry study is excluded, the lowest RR estimate, 0.898, is obtained. So Perry exerts the greatest influence of pulling the pooled estimate toward the null or an advantage for late TRTx. Exclusion of the Takada, Fukuoka, Kawahara, et al. (2002) study results in the highest RR estimate, 0.955. Takada, Fukuoka, Kawahara, et al. (2002) is the most influential study in drawing the pooled estimate in the direction favoring early TRTx. Exclusion of the Perry study was the only instance in which a significant pooled result was obtained. However, as a whole, excluding any individual study has little influence on the estimate of the pooled RR.

Summary Table 18 gives the results of subgroup/sensitivity analysis. Selection of variables defining subgroups was constrained by the relatively small pool of studies. Fine gradations of variables dilute the power to detect differences between early and late TRTx. Two concerns guide interpretation of subgroup results: the magnitude of differences in RR point estimates for different levels of a variable and whether statistical significance is achieved in any subgroup. Q statistic values exceeded the threshold for significant heterogeneity in 4 instances: among those studies that delivered the early TRTx at the earliest time (beginning in the first week of chemotherapy), those that did not use platinum chemotherapy, those that gave TRTx and chemotherapy concurrently, and those studies rated as being of good quality. These subgroups were pooled using random-effects models, while all other subgroups were pooled with fixed-effects.

Use of hyperfractionation was the variable with the greatest difference in point estimates of RR between subgroups of studies. Inclusion of studies using hyperfractionation produced a significant pooled RR of 0.815 (95 percent CI: 0.702–0.946). Studies that used once daily fractionation had a pooled RR much closer to the null, 0.972 (95 percent CI: 0.913–1.035). There was a moderate difference between point estimates of those studies that did and did not use platinum. Studies using platinum in chemotherapy regimens obtained a greater reduction in mortality, with a significant RR of 0.905 (95 percent CI: 0.829–0.987). Those not using platinum yielded an RR close to the null, 0.964 (95 percent CI: 0.848–1.096). The set of studies that offered the earliest early TRTx did not result in a statistically significant reduction in mortality at 2 years for early TRTx (RR=0.914, 95 percent CI: 0.792–1.054). The point estimate for those studies not among the earliest was nearly identical and also nonsignificant (RR=0.918, 95 percent CI: 0.841–1.001). Those using concurrent TRTx had a nonsignificant RR of 0.938 (95 percent CI: 0.799–1.100) and those not using concurrent TRTx had a significant RR of 0.920 (95 percent CI: 0.854–0.992). There was a considerable difference between studies of good quality and lesser quality, but pooled results were nonsignificant for both. Good quality studies produced a RR of 0.874 (95 percent CI: 0.744–1.027). Lesser quality studies had a RR of 0.975 (95 CI: 0.906–1.050). A random effects meta-regression (Table 19) found that no variables was a significant predict of differences in treatment effect at 2 years, but hyperfractionation was nearly significant (p=0.07).

3-Year Mortality. The funnel plot (Figure 5) suggests the presence of publication bias. Point estimates appear to be missing in the lower right region of the plot. Linear regression (Egger test, Table 20) shows that the intercept differs significantly from zero, confirming that publication bias may be present.

Three-year mortality RRs for individual trials are given in Table 21. Three trials obtained RR estimates favoring late TRTx, while the other 8 favor early TRTx. The 95 percent CIs all overlap the null value RR of 1.0.

Table 22 shows that the Q statistic value does not exceed the level for concluding that significant heterogeneity of effects is present. Thus, a fixed effects model was used to compute a pooled 3-year RR (see forest plot in Figure 6). The obtained estimate was 0.991 (95 percent CI: 0.955–1.029). Based on these results, it cannot be concluded that use of early TRTx significant reduces the risk of mortality at 3 years.

The influence analysis plot in Figure 7 shows only extremely small changes in the pooled RR estimate when individual studies are excluded. When the Perry study is excluded, the lowest pooled RR estimate is produced: 0.977. This study has the greatest impact on drawing the pooled RR toward the null or effects favoring late TRTx. The largest pooled RR is derived when the Murray study is excluded: 1.000. Murray has the strongest influence on pulling the pooled RR away from the null, favoring early TRTx. Point estimates changed very little when individual studies were excluded.

Results of subgroup/sensitivity analysis are presented in Summary Table 23. The subset of studies using hyperfractionation yielded a significant pooled RR of 0.908 (95 percent CI: 0.828–0.995). Those that used once daily fractionation had a nonsignificant pooled RR of 1.008 (95 percent CI: 0.968–1.050). No other subgroup produced a significant result. Studies in which platinum was part of chemotherapy regimens had an RR of 0.958 (95 percent CI: 0.910–1.009). Non-platinum studies produced an RR of 1.029 (95 percent CI: 0.975–1.085). The group of studies in which early TRTx was begun at the earliest time produced a nearly null-value RR (0.998, 95 percent CI: 0.953–1.045). Those studies that began early TRTx after the first week of chemotherapy produced a similar RR (0.980, 95 percent CI: 0.921–1.042). Studies that offered concurrent RTx had a similar pooled RR (0.997, 95 percent CI: 0.947–1.051) compared with those that did not (RR: 0.985, 95 percent CI: 0.935–1.038). There was a modest difference between studies of good quality versus lesser quality. Good quality studies had an RR of 0.948 (95 percent CI: 0.843–1.064), while fair and poor quality studies had an RR of 1.000 (95 CI: 0.956, 1.047). Results of random effects meta-regression are shown in Summary Table 24. Use of hyperfractionation (p=0.04) was the only significant predictor, while use of platinum (p=0.06) was nearly a significant predictor of differences in treatment effects.

Summary. This meta-analysis indicates that the findings of Fried, Morris, Poole, et al. (2004) are not reproducible when different pooling methods are used and 3 additional studies are included. We found evidence of publication bias at both 2 and 3 years, while Fried, Morris, Poole, et al. (2004) found it at neither time. Significant heterogeneity was observed at 2 years here but not at 3 years. Thus, we used a random effects model at 2 years and a fixed effects model at 3 years, but Fried, Morris, Poole, et al. (2004) did not find significant heterogeneity at either period and used only fixed effects models. While Fried, Morris, Poole, et al. (2004) reported a significant advantage for early TRTx at 2 years and nonsignificance at 3 years, nonsignificant results were obtained here at both periods.

Subgroups including studies using hyperfractionation or platinum yielded significant advantages for early TRTx at 2 years in both Fried, Morris, Poole, et al. (2004) and this meta-analysis. At 3 years, Fried, Morris, Poole, et al. (2004) reported that both subgroups retained significance, while here only hyperfractionation was significant. The current meta-regression found hyperfractionation to be nearly significant (p=0.07) at 2 years; hyperfractionation was significant at 3 years (p=0.04) and platinum was nearly significant at 3 years (p=0.06). Fried, Morris, Poole, et al. found that hyperfractionation and platinum predicted heterogeneity in risk differences.

As an exercise, we ran multiple variable meta-regression models, but none were significant at either period. In particular, hyperfractionation and platinum were not significant independent predictors here in multiple variable models. In contrast, Fried, Morris, Poole, et al. (2004) found larger effects when the variables were combined. Any meta-gresssion with multiple variables models is limited by the risk of overfitting when the pool of studies is small.