Predictors of response to subsequent chemotherapy in platinum pretreated ovarian cancer: A multivariate analysis of 704 patients

Background: The probability of response to chemotherapy following platinum based treatment in ovarian cancer has usually been related to the 'platinum-free interval'. However, in a recent European-Canadian trial of paclitaxel, serous histology, tumor bulk, and hemoglobin, but not treatment free interval, were predictors of response. To determine if these observations were unique to this study (or this drug), data from other active agents given as second-or third-line treatment in ovarian cancer were obtained and analyzed. Methods. In the first part of the study, results of trials in 1185 platinum pretreated ovarian cancer patients were obtained on six agents: paclitaxel, epirubicin, docetaxel, carboplatin, irinotecan, and gemcitabine. Response results according to histology, baseline hemoglobin, tumor size and time from last chemotherapy were determined for each agent and the Cochran-Mantel-Haenszel procedure was used to obtain an overall assessment of significance for each factor. In the second part of the study, individual data from 704 patients in four studies (three agents: paclitaxel, docetaxel and epirubicin) were pooled for univariate and multivariate analysis of factors predictive of response. Results: In the analysis of results of individual agents all factors examined were significant predictors of response: serous histology (P = 0.001), tumor size less than or equal to 5 cm (P = 0.02), normal baseline hemoglobin (P = 0.003), and greater than or equal to 6 mo since last treatment (P = 0.001). While these results were interesting, they did not supply definitive answers regarding independent response predictors. Therefore a multifactor analysis was undertaken on the 704 patients for whom individual data were available. Of the 11 factors examined in a univariate analysis, 10 met the criteria for inclusion in a stepwise logistic regression. In the final model only 3 factors remained as significant independent predictors of response: serous histology (P = 0.009), no. disease sites (P = 0.003), and tumor size (P = 0.001). Time from last treatment, when evaluated as a continuous variable, was not in the final model and was highly correlated with tumor size (P = 0.0005). Conclusions: On the basis of this analysis, we conclude that tumor burden (as assessed by size of the largest lesion and number of disease sites) and histology are factors of importance in response to subsequent chemotherapy in relapsed ovarian cancer. Time from last treatment was correlated with tumor size in this data set and its effect on response was dependent on whether it was examined as a categorical or continuous variable, so we conclude it is not the sole critical factor of biologic importance. We recommend description of these factors in reports of phase II studies, confirmation of these findings in other data sets and further investigation of the mechanism of sensitivity of serous tumors.