Ched and unmatched groups, and found that the two subgroups had identical distributions for all these variables. Therefore, the COXEN matched group showed a significantly longer survival than the unmatched group while both groups had identical clinical characteristics except that the former group was treated with the predicted effective drugs. Avoiding potential bias on a cohort from a single site, we thus believe that our observations on the TCGA cohort from .10 clinical centers could reasonably reflect the outcomes from the use of these predictors on the general EOC patient population. Also, note that the patient characteristics and survival statistics of this TCGA cohort have been confirmed to be well matched with those in the general EOC population [14]. It is worthwhile to note several limitations of our current study. In this study we were able to perform our COXEN analysis only on the three standard chemotherapy drugs for which we had multiple patient data sets for our rigorous statistical prediction modeling, independent evaluation, and external validation. We employed this strict statistical principle to avoid many pitfalls from a genomic-based biomarker study, which resulted in a very limited set of drugs for our analysis. Despite such a limitation, we found that a comparative effectiveness-based selection only among the three drugs could still potentially provide a survival benefit compared to the current unselective use of many standard agents for recurrent EOC. Thus, we believe that, if further validated in a prospective setting, this kind of comparative drug selection strategy based on multiple therapeutic biomarker predictors may be proven to be highly effective to improve patient outcomes. This can then be expanded to a more comprehensive prediction capability among other commonly used chemotherapy agents, such as liposomal doxorubicin, and even different administrationPLOS ONE | www.plosone.orgSurvival Improvement by Personalized Chemotherapyschedules, including weekly paclitaxel. Unfortunately, current patient data with which we can assess such comparative effectiveness are very limited. As such, our study was based only on the estimated efficacy among limited drug selections. Also, our statistical estimation of the positive predictive value (PPV) curves for the drug predictors could be further improved by a nonparametric estimation to correlate their predicted scores more precisely with patient 5-year survival Quinagolide (hydrochloride) web probabilities if larger numbers of patients were available for these drugs. Finally, we note that even if our retrospective analysis has showed some evidence for an improved survival of advanced EOC by a selective use of several standard chemotherapy drugs, these findings must be confirmed in a prospective study, which may also allow us to refine our comparative drug selection strategy among the drugs.platinum-sensitive patients. (TIF)patients,(C)platinum-resistantFigure S6 Kaplan-Meier survival analysis for the validation of not being prognostic prediction on patients not treated with each drug. (A) paclitaxel predictor prediction, (B) Win 63843 supplier cyclophosphamide predictor prediction, (C) topotecan predictor prediction. (TIF) Figure S7 Comparative effectiveness of the COXEN predictors. Five-year survival positive predicted values (PPVs) are plotted against the predictor cutoff values. Paclitaxel and cyclophosphamide predictors provided higher five-year survival chances (PPVs) than topotecan predictors when a patient had s.Ched and unmatched groups, and found that the two subgroups had identical distributions for all these variables. Therefore, the COXEN matched group showed a significantly longer survival than the unmatched group while both groups had identical clinical characteristics except that the former group was treated with the predicted effective drugs. Avoiding potential bias on a cohort from a single site, we thus believe that our observations on the TCGA cohort from .10 clinical centers could reasonably reflect the outcomes from the use of these predictors on the general EOC patient population. Also, note that the patient characteristics and survival statistics of this TCGA cohort have been confirmed to be well matched with those in the general EOC population [14]. It is worthwhile to note several limitations of our current study. In this study we were able to perform our COXEN analysis only on the three standard chemotherapy drugs for which we had multiple patient data sets for our rigorous statistical prediction modeling, independent evaluation, and external validation. We employed this strict statistical principle to avoid many pitfalls from a genomic-based biomarker study, which resulted in a very limited set of drugs for our analysis. Despite such a limitation, we found that a comparative effectiveness-based selection only among the three drugs could still potentially provide a survival benefit compared to the current unselective use of many standard agents for recurrent EOC. Thus, we believe that, if further validated in a prospective setting, this kind of comparative drug selection strategy based on multiple therapeutic biomarker predictors may be proven to be highly effective to improve patient outcomes. This can then be expanded to a more comprehensive prediction capability among other commonly used chemotherapy agents, such as liposomal doxorubicin, and even different administrationPLOS ONE | www.plosone.orgSurvival Improvement by Personalized Chemotherapyschedules, including weekly paclitaxel. Unfortunately, current patient data with which we can assess such comparative effectiveness are very limited. As such, our study was based only on the estimated efficacy among limited drug selections. Also, our statistical estimation of the positive predictive value (PPV) curves for the drug predictors could be further improved by a nonparametric estimation to correlate their predicted scores more precisely with patient 5-year survival probabilities if larger numbers of patients were available for these drugs. Finally, we note that even if our retrospective analysis has showed some evidence for an improved survival of advanced EOC by a selective use of several standard chemotherapy drugs, these findings must be confirmed in a prospective study, which may also allow us to refine our comparative drug selection strategy among the drugs.platinum-sensitive patients. (TIF)patients,(C)platinum-resistantFigure S6 Kaplan-Meier survival analysis for the validation of not being prognostic prediction on patients not treated with each drug. (A) paclitaxel predictor prediction, (B) cyclophosphamide predictor prediction, (C) topotecan predictor prediction. (TIF) Figure S7 Comparative effectiveness of the COXEN predictors. Five-year survival positive predicted values (PPVs) are plotted against the predictor cutoff values. Paclitaxel and cyclophosphamide predictors provided higher five-year survival chances (PPVs) than topotecan predictors when a patient had s.