Survival outcome prediction in cervical cancer: Cox models versus deep-learning model.
Am J Obstet Gynecol. 2018 Dec 21;:
Authors: Matsuo K, Purushotham S, Jiang B, Mandelbaum RS, Takiuchi T, Liu Y, Roman LD
BACKGROUND: Historically, the Cox proportional hazard regression (CPH) model has been the mainstay for survival analyses in oncologic research. The CPH model is generally utilized based upon an assumption of linear association. However, it is likely that in reality there are many clinico-pathological features that exhibit a non-linear association in biomedicine.
OBJECTIVE: To compare the deep-learning neural network model and the CPH model in the prediction of survival in women with cervical cancer.
STUDY DESIGN: This is a retrospective pilot study of consecutive cases of newly diagnosed stage I-IV cervical cancer between 2000-2014. A total of 40 features including patient demographics, vital signs, laboratory test results, tumor characteristics, and treatment types were assessed for analysis and grouped into three feature sets. The deep-learning neural network model was compared to the CPH model and three other survival analysis models for progression-free survival (PFS) and overall survival (OS). Mean absolute error and concordance index were used to assess the performance of these five models.
RESULTS: There were 768 women included in the analysis. The median age was 49 years, and the majority were Hispanic (71.7%). The majority of tumors were squamous (75.3%) and stage I (48.7%). The median follow-up was 40.2 months, and there were 241 events for recurrence and progression and 170 deaths during follow-up. The deep-learning model showed promising results in the prediction of PFS when compared to the CPH model (mean absolute error, 29.3 versus 316.2). The deep-learning model also outperformed all the other models, including the CPH model, for OS (mean absolute error, CPH versus deep-learning, 43.6 versus 30.7). The performance of the deep-learning model further improved when including more features (concordance index for PFS: 0.695 for 20 features, 0.787 for 36 features, and 0.795 for 40 features). There were 10 features for PFS and three features for OS that demonstrated significance only in the deep-learning model but not in the CPH model. There were no features for PFS and three features for OS that demonstrated significance only in the CPH model but not in the deep-learning model.
CONCLUSION: Our study suggests that the deep-learning neural network model may be a useful analytic tool for survival prediction in women with cervical cancer, exhibiting superior performance compared to the CPH model. This novel analytic approach may provide clinicians with meaningful survival information that could potentially be integrated into treatment decision-making and planning. Further validation studies are necessary to support this pilot study.
PMID: 30582927 [PubMed – as supplied by publisher]