Identification of prognostic variables in advanced ovarian cancer based on real-world patients from big data analytics.

Abstract

e17104 Background: Prognosis of advanced ovarian cancer is dismal with most cases recurring after initial surgery. Current factors able to predict the evolution of the disease are limited to BRCA status and platinum sensitivity. We aim to explore the potential of different clinical variables as prognostic factors using big data analytics in currently available hospitalary departmental information systems. Methods: An observational study with two cohorts (one prospective and one retrospective) was designed. Inclusion criteria were adult patients ( > 18 years old) diagnosed with epithelial ovarian cancer stage IC or superior.Clinical and histological data were recorded by a central data manager to ensure homogeneity in data collection. Big data analytics consisted on building an approximation to the statistical distribution of the tests to distinguish different kinds of features (metric, categorical, free text). Bootstrap resampling allowed to characterize the confidence regions for proportion differences, average differences, and text-profile differences in exitus vs in non-exitus groups. Results: Up to 265 patients in four different hospitals were recruited. Median age was 59 years (range 20-87), stage distribution was 48 (18%) I, 20 (8%) II, 122 (46%) III, 41 (15%) IVand 34 (13%) NA. Histology distribution was 158 (60%) papillary serous, 31 (12%) endometrioid, 18 (7%) clear cell, 11 (4%) mucinous and 47 (18%) NA. 152 (58%) patients underwent upfront surgery, 76 (29%) interval surgery and 10 (4%) no surgical intervention, 27 (10%) NA. 207 (78%) achieved optimal cytoreduction. 232 (88%) received adjuvant chemotherapy, most with carboplatin plus paclitaxel 180 (68%) and 48 (18%) also with bevacizumab. Median follow up was 81.4 months (CI95% 64.1-98.7) The proposed Big Data analytics identified a higher frequency of upfront surgery (vs interval surgery) and bevacizumab administration (vs chemotherapy alone) in the non-exitus group. Conclusions: Our results point to the notion that performance of upfront surgery and bevacizumab administration could have a long term impact in ovarian cancer. Simple Big Data analytics can contribute to identify new prognostic factors and to assess their real impact on patients managed in daily practice.