Synthesizing the Evidence
Studies of combined modality radiotherapy and chemotherapy can generally be categorized as using a concomitant or sequential approach based on the relative timing of the radiotherapy and chemotherapy, with different biological bases behind their designs. In this report, the trials that used a concomitant approach were described and analyzed separately from trials using a sequential approach. When data from trials of sequential and concomitant approaches were examined together, the pooled data were heterogeneous, suggesting that the studies are different in nature. Thus, a combined analysis of both approaches was rejected.
Data on survival and local recurrence were pooled and the results were examined for statistical heterogeneity. For each meta-analysis, data were pooled at a common time-point (e.g., mortality at one-year). The time point selected for meta-analyses must be clinically credible and relevant but not so far along the survival curve that wide confidence intervals result from fewer patients contributing to the estimate. Since time points prior to the median will generally ensure that there is sufficient data to be credible, the median survival times, weighted by the size of the treatment arms, were calculated to determine an appropriate time point for each meta-analysis. Pooling was conducted using one-year mortality data for all meta-analyses because the weighted median survival time was less than one year for both the concomitant and sequential groups.
The study results were pooled using Review Manager 4.0.3 (Metaview© Update Software), which is available through the Cochrane Collaboration. The random effects model was used as the more conservative estimate of effect. Results were expressed as odds ratios (OR) with 95% confidence intervals (CI). An odds ratio less than 1.0 favours the experimental treatment (i.e., radiochemotherapy [RTCT]) and an odds ratio greater than 1.0 favours the control (i.e., radiotherapy [RT] alone). In addition, the absolute difference is presented as percent difference in outcome, calculated from the pooled event rates. The number of patients that need to be treated with RTCT for one additional patient to benefit (NNT) was also calculated.
Results for adverse effects were not pooled because the primary authors of eligible trials reported data on adverse effects using different scoring systems and symptom categories. The presentation of the incidence of adverse effects (as opposed to the numbers of patients affected within each toxicity grade) makes a quantitative summary statistic difficult. The results were summarized in a descriptive fashion for this review based on the incidence of grade of toxicity for acute and late adverse effects, where available across the studies, to allow for a qualitative comparison.
Data extraction was performed independently by two members of the Gastrointestinal Cancer Disease Site Group (DSG). Discrepancies were resolved through consensus.
Subgroup Analysis
It was hypothesized a priori that the use of cisplatin versus non-cisplatin chemotherapy would have an impact on the effectiveness of treatment, and a subgroup analysis was planned to examine this hypothesis. The two most commonly employed chemotherapy regimens in Canada are 5-fluorouracil (5FU)/mitomycin and 5FU/cisplatin, and one of the major decisions facing clinicians is what type of chemotherapy to use if the combined modality approach is adopted. Furthermore, cisplatin-based chemotherapy has been used in combination with radiotherapy in many other disease systems resulting in significant improvement in outcome. It is, therefore, important to explore the impact of cisplatin versus non-cisplatin chemotherapy within the context of combined modality.
Potential Sources of Heterogeneity and Sensitivity Analysis
The following factors were postulated a priori to be potential sources of heterogeneity: study quality using scores on the Jadad scale (>2 versus <2); dose of radiotherapy (biological effective [or equivalent] dose (BED)* >60 versus BED <60); and type of chemotherapy (cisplatin-containing versus others). These factors were used to explore any significant heterogeneity of results across the trials. Heterogeneity of study results was assessed using a visual plot and by calculating the Breslow-Day statistic using a planned cut-off for significance of p<0.05. The robustness of our conclusions was examined through subsequent sensitivity analyses using these factors.
*Note: To facilitate comparison across trials, radiotherapy dose was converted to biological equivalent dose using the equation BED=nd (1+d/alpha/beta), where n=number of fractions, d=dose per fraction, and the assumption that alpha/beta=10 for tumour effect. Due to the limitations of this model, no allowance can be made for time gaps in split-course treatments.
