The role of secondary cytoreductive surgery in patients with recurrent epithelial ovarian, tubal, and peritoneal cancers: a comparative effectiveness analysis

Abstract

Background. All published reports concerning secondary cytoreductive surgery for relapsed ovarian cancer have essentially been observational studies. However, the validity of observational studies is usually threatened from confounding by indication. We sought to address this issue by using comparative effectiveness methods to adjust for confounding. Methods. Using a prospectively collected administrative health care database in a single institution, we identified 1,124 patients diagnosed with recurrent epithelial, tubal, and peritoneal cancers between 1990 and 2009. Effectiveness of secondary cytoreductive surgery using the conventional Cox proportional hazard model, propensity score, and instrumental variable were compared. Sensitivity analyses for residual confounding were explored using an array approach. Results. Secondary cytoreductive surgery prolonged overall survival with a hazard ratio (95% confidence interval) of 0.76 (range 0.66-0.87), using the Cox proportional hazard model. Propensity score methods produced comparable results: 0.75 (range 0.64-0.86) by nearest matching, 0.73 (0.65-0.82) by quintile stratification, 0.71 (0.65-0.77) by weighting, and 0.72 (0.63-0.83) by covariate adjustment. The instrumental variable method also produced a comparable estimate: 0.75 (range 0.65-0.86). Sensitivity analyses revealed that the true treatment effects may approach the null hypothesis if the association between unmeasured confounders and disease outcome is high. Conclusions. This comparative effectiveness study provides supportive evidence for previous reports that secondary cytoreductive surgery may increase overall survival for patients with recurrent epithelial, tubal, and peritoneal cancers.

Figure 1.

Kaplan-Meier plot for overall survival using different propensity score strategies. (A) Nearest neighbor one-to-one matching (caliper = 0.25σ; p = .001, log-rank test). (B) Mahalanobis metric matching without propensity score (p < .001, log-rank test). (C) Mahalanobis metric matching with propensity score (p < .001, log-rank test).

Figure 2.

Forest plot of the hazard ratio of overall survival with secondary cytoreductive surgery versus without secondary cytoreductive surgery. The estimate of the hazard ratio of each study corresponds to the middle of the squares and the horizontal line gives the 95% confidence interval (CI). The closed diamond shows the pooled total hazard ratio with its 95% CI. A test of heterogeneity by Q-statistic between the studies is given below the summary statistic. Abbreviations: CI, confidence interval; IPTW, inverse probability of treatment weighted.

Figure 3.

Sensitivity analysis using an array approach. The observed hazard ratio was set at 0.72 (from pooled results of Fig. 2). Prevalence of unmeasured confounders in the control group (not receiving surgery) was set at 0.3. Two factors were varied: the strength of the unmeasured confounder-outcome association (1.0–5.5) and the prevalence of the unmeasured confounder in the treatment group (received surgery, 0.0–0.5). A three-dimensional mesh plot was constructed to check the impacts on the fully adjusted hazard ratio by varied unmeasured confounder settings. Abbreviation: HR, hazard ratio.