Palliative Care

SS 10 - Palliative 2

71 - Optimized Survival Evaluation to Guide Bone Metastases Management:?Developing an Improved Statistical Approach

Monday, October 22
7:45 AM - 7:55 AM
Location: Room 007 C/D

Optimized Survival Evaluation to Guide Bone Metastases Management: Developing an Improved Statistical Approach
S. R. Alcorn1, J. Fiksel2, T. Smith3, J. L. Wright1, T. R. McNutt1, T. L. DeWeese1, and S. Zeger2; 1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 2Department of Biostatistics, Johns Hopkins School of Public Health, Baltimore, MD, 3Department of Palliative Care, Johns Hopkins University School of Medicine, Baltimore, MD

Purpose/Objective(s): In managing bone metastases, estimation of life expectancy is central for individualizing patient care given a range of radiation therapy (RT) treatment options. With access to larger volume and more complex patient data and statistical models, oncologists must develop methods for optimal decision support. Yet including more covariates may not improve predictions; studies show simple survival models of 2-3 variables perform nearly as well as 6-8 variable models in metastatic populations. Approaches incorporating many covariates should also identify complex interactions and manage missing data. To address these issues, we applied a statistical learning approach, random survival forests (RSF), to predict survival for patients with bone metastases using up to 26 predictor variables. We then compared our method to two simpler, validated Cox regression models.

Materials/Methods: For 397 patients evaluated in RT consultation for bone metastases from 1/2007 to 1/2013, data for 26 readily available clinical variables was collected. Primary outcome was time from consultation to death. Patients were randomly assigned to training (n=306) and validation sets (n=91). The training set was used to build a RSF model. To establish relative utility of our RSF model, we performed Cox regressions per Chow’s 3-item Survival Prediction Score (SPS) and Westhoff’s 2-item tool (W2). Predictive accuracy of the 3 models was compared using time-dependent area under the curve (tAUC). We obtained internal estimates of tAUC using the .632+ bootstrap method for the training set and external estimates for the validation set.

Results: Patient mean age was 62 years (SD 13). Median survival was 227 days. Table 1 shows tAUC at select time points by survival model. Using both internal and external tAUC estimates, RSF predictions out-performed simpler models at all times, with greatest difference at 30 days and performance most similar at the 1-year prediction horizon. For the RSF model, variable importance was greatest for performance status, blood cell counts, histology, age, recent chemotherapy, time from diagnosis, RT site, and neuraxis compromise.

Conclusion: For patients with bone metastases, our RSF model substantially improved survival predictions versus relatively simpler traditional models. As such, we have developed a web platform to facilitate ease of data entry and display predicted patient survival probabilities from our RFS to guide in selection of appropriate RT regimens. Our future work aims to further optimize estimates through inclusion of time-dependent covariates to better reflect the dynamic nature of health status and through investigation of deep learning models.
Table 1: Estimates of tAUC by time and survival model

Training set

Validation set

Days from consultation RSF SPS W2 RSF SPS W2
30 0.90 0.82 0.83 0.90 0.82 0.80
180 0.84 0.80 0.81 0.80 0.77 0.74
360 0.84 0.80 0.83 0.74 0.70 0.72

Author Disclosure: S.R. Alcorn: Research Grant; NIH. J. Fiksel: Research Grant; National Institute on Aging. T. Smith: None. J.L. Wright: None. T.R. McNutt: Research Grant; Elekta Oncology Systems, Philips Radiation Oncology Systems, Toshiba. Patent/License Fees/Copyright; Accuray-Tomotherapy, Sun Nuclear. President Elect; AAPM-MAC. T.L. DeWeese: None. S. Zeger: Honoraria; ASTRO.

Sara Alcorn, MD, MPH

The Johns Hopkins Hospital: assistant faculty: Employee

NIH: Research Grants


Send Email for Sara Alcorn


71 - Optimized Survival Evaluation to Guide Bone Metastases Management:?Developing an Improved Statistical Approach

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Optimized Survival Evaluation to Guide Bone Metastases Management:?Developing an Improved Statistical Approach