This review concluded that timing of administration of chemotherapy (postoperative versus preoperative) did not significantly affect overall and disease-free survival in patients with resectable lung cancer. The authors' conclusions should be interpreted with caution as they were based on indirect comparisons.

To indirectly compare the effect of preoperative versus postoperative chemotherapy on survival rates in patients with operable non-small cell lung cancer.

MEDLINE (1950 to August 2007), EMBASE (1996 to August 2007) and The Cochrane Library (2007, issue 3) were searched without language restrictions. Search terms were reported. The full research strategy was available on request. Trial registers on the internet, abstracts from major thoracic surgery scientific meetings (2005 to 2007) and reference lists of relevant studies were searched. Authors of previous studies were contacted for further information where required.

Randomised controlled trials (RCTs) that compared general chemotherapy plus surgery to surgery alone for non-small cell lung cancer were eligible for inclusion if they reported overall survival rates. The secondary outcome of interest was disease-free survival.

Where reported, mean/median age of patients in the included studies ranged from 53 to 65 years. Where reported, most studies included more than 50% males (range 34% to 98%). Most patients reported disease stage I to III. Chemotherapy regimens varied (no data on doses and frequencies) and included single or combination therapy. Most studies administered chemotherapy postoperatively (22 RCTs). Perioperative administration was defined as chemotherapy given before and after surgery (three RCTs). Some studies reported use of postoperative radiotherapy (nine RCTs). Between 50% and 100% of patients completed chemotherapy.

Three reviewers assessed English-language papers for inclusion in the review, with discrepancies resolved through consensus. Foreign-language papers were assessed by two reviewers for inclusion.

Downs and Black quality assessment tool (27 items) was used to evaluate the quality of included studies. The maximum score was 33 (higher scores indicated studies of higher quality).

The authors did not state how many reviewers performed quality assessment.

It appeared that two reviewers extracted hazard ratios (HRs) and their 95% confidence intervals (CIs) for overall and disease-free survival where reported in the original articles. Where hazard ratios were not reported, they were calculated from summary statistics or survival plots using the method of Parmar. Where there was more than one chemotherapy and surgery subgroup, results were entered separately as two trials.

A random-effects model was used to combine hazard ratios and their 95% CIs, weighted by the inverse of the variance and taking into account the timing of chemotherapy (preoperative versus postoperative) on survival. The authors reported that they also undertook fixed-effect meta-analysis, but only results using the random-effects model were presented. Statistical heterogeneity was assessed using the X² test and I² statistic.

Sensitivity analyses were conducted using meta-regression to determine whether significant differences may have resulted from inclusion of studies that: administered chemotherapy perioperatively; administered tegafur/uracil; were of patients with stage I disease only; administered radiotherapy; and commenced in different years. Sensitivity analysis also addressed substitution of hazard ratios obtained by studies of individual patient data meta-analyses, where available.

Cumulative data meta-analysis was undertaken to assess the impact of increasing information over time for preoperative chemotherapy trials and postoperative chemotherapy trials.

Publication bias was assessed using the Begg's test.

Thirty-two studies (33 comparative groups, n=10,811) were included in the review. Twenty two studies (23 arms) administered postoperative chemotherapy. Ten studies administered pre/perioperative chemotherapy. Sample sizes ranged from 26 to 1,867 participants. The overall mean quality score was 21 (range 15 to 24).

Hazard ratios for overall mortality were similar for postoperative chemotherapy (HR 0.80, 95% CI 0.74 to 0.87; 22 studies) and pre/perioperative chemotherapy (HR 0.81, 95% CI 0.68 to 0.97; 10 studies) when compared with a control. There was no evidence of statistical heterogeneity between the two groups (p=0.81) or for all included studies (p=0.12, I²=23%). Indirect evidence that compared the two chemotherapy regimens showed no statistically significant difference (HR 0.99, 95% CI 0.81 to 1.21).

Similar results were reported for disease-free survival. Postoperative chemotherapy studies reported a hazard ratio for death or recurrence of 0.76 (95% CI 0.68 to 0.85) and a hazard ratio of 0.80 (95% CI 0.66 to 0.92) for pre/perioperative chemotherapy compared with a control. There was evidence of statistical heterogeneity for postoperative studies (p=0.014, I²=50%), but not for preoperative studies (p=0.154, I²=36%). Indirect comparison supported this and showed no statistically significant difference in chemotherapy timing (HR 0.96, 95% CI 0.77 to 1.20).

Meta-regression analysis did not find any significant associations between indirect comparison of overall survival and any of the variables investigated.

There was evidence of publication bias using Begg's test (p=0.047).

Timing of administration of chemotherapy (postoperative versus preoperative) did not significantly affect overall and disease-free survival in patients with resectable lung cancer.

The review question and associated inclusion criteria were clear. A satisfactory search of the literature was undertaken without language restrictions, which reduced potential for language bias. A search for unpublished data was undertaken, which reduced the possibility that potentially relevant papers may have been missed. Publication bias was assessed and showed evidence of small-study effects. The authors performed study selection and data extraction in duplicate, but the process was unclear for validity assessment and so potential for reviewer error and bias could not be ruled out completely. Data were pooled using meta-analyses. The authors investigated statistical heterogeneity. Comparison between postoperative and pre-/perioperative therapy was based on indirect comparisons, which may not produce reliable results and so the authors' conclusions should be interpreted with caution.

The authors did not state any implications for practice or research.

None stated.

This is a critical abstract of a systematic review that meets the criteria for inclusion on DARE. Each critical abstract contains a brief summary of the review methods, results and conclusions followed by a detailed critical assessment on the reliability of the review and the conclusions drawn.