DeMora, Lyudmila

PURPOSE To compare the predictive ability of mapping algorithms derived using cross-sectional and longitudinal data. METHODS This methodological assessment used data from a randomized controlled noninferiority trial of patients with low-risk prostate cancer, conducted by NRG Oncology (ClinicalTrials.gov identifier: NCT00331773 ), which examined the efficacy of conventional schedule versus hypofractionated radiation therapy (three-dimensional conformal external beam radiation therapy/IMRT). Health-related quality-of-life data were collected using the Expanded Prostate Cancer Index Composite (EPIC), and health utilities were obtained using EuroQOL-5D-3L (EQ-5D) at baseline and 6, 12, 24, and 60 months postintervention. Mapping algorithms were estimated using ordinary least squares regression models through five-fold cross-validation in baseline cross-sectional data and combined longitudinal data from all assessment periods; random effects specifications were also estimated in longitudinal data. Predictive performance was compared using root mean square error. Longitudinal predictive ability of models obtained using baseline data was examined using mean absolute differences in the reported and predicted utilities. RESULTS A total of 267 (and 199) patients in the estimation sample had complete EQ-5D and EPIC domain (and subdomain) data at baseline and at all subsequent assessments. Ordinary least squares models using combined data showed better predictive ability (lowest root mean square error) in the validation phase for algorithms with EPIC domain/subdomain data alone, whereas models using baseline data outperformed other specifications in the validation phase when patient covariates were also modeled. The mean absolute differences were lower for models using EPIC subdomain data compared with EPIC domain data and generally decreased as the time of assessment increased. CONCLUSION Overall, mapping algorithms obtained using baseline cross-sectional data showed the best predictive performance. Furthermore, these models demonstrated satisfactory longitudinal predictive ability.

265 Background: A recent study ( Fossati et al, 2018) found that higher lymph node count at radical prostatectomy was associated with improved outcomes in patients treated with salvage radiation for elevated prostate-specific antigen (PSA) after surgery. We sought to validate these results in NRG/RTOG 9601, a randomized controlled trial of men with pT2/T3 disease who underwent either radiation (RT) alone or RT+antiandrogen (bicalutamide) therapy for PSA elevation following radical prostatectomy from 1998-2003. Methods: We reviewed available pathology reports for all patients in NRG/RTOG 9601 to determine the nodal count at radical prostatectomy. Clinical data was as of 11/5/2015, same as the primary endpoint for the trial. Cox proportional hazards models were used to assess the effect of number of positive lymph nodes, treatment arm (RT alone or RT+bicalutamide), Gleason score, positive margins, and seminal vesicle invasion on the following endpoints: times to local and distant failure and overall and disease specific survival. Results: Out of the 760 patients originally eligible in the trial, 552 (73%, 276 in each arm) had complete data available. Median node count in the entire cohort was 6 (range 0-33, Q1-Q3 3-9). There were no significant differences between treatment arms in terms of patient demographic or clinical characteristics, including total lymph nodes removed in either arm (RT alone vs RT+bicalutamide median 5 vs 6, p = 0.11). There was no significant association between total lymph nodes and overall survival with both arms combined (HR = 1.00, 95% CI:0.97-1.03, p = 0.87) or in the individual arms alone (RT+Casodex: HR = 1.01, 95% CI:0.97-1.05, p = 0.65; RT+Placebo: HR = 0.98, 95% CI: 0.94-1.03, p = 0.45). There was also no significant association between total lymph nodes and disease-specific survival with both arms combined (HR = 1.00, 95% CI:0.95-1.04, p = 0.84) and in the arms alone (RT+Casodex: HR = 1.00, 95% CI:0.95-1.05, p = 0.92; RT+Placebo: HR = 0.99, 95% CI: 0.92-1.07, p = 0.86). In multivariable analysis performed on the two arms, Gleason score was the only feature associated with worse overall and disease-specific survival, seen only in the RT alone arm. Similar findings were seen when evaluating times to local and distant failure. Conclusions: Lymph node yield in NRG/RTOG 9601 did not show any association with adverse outcomes in the entire cohort, or in either treatment arm alone. The therapeutic benefit of an extensive lymph node dissection in this population remains uncertain. Clinical trial information: NCT00002874.

60 Background: Mapping algorithms informing economic evaluations are often derived using baseline data from clinical trials. It is unclear if these algorithms can predict health utilities accurately in post-intervention data. Thus, this study examines the longitudinal predictive ability of mapping algorithms derived from baseline trial data and explores the factors associated with prediction errors. Methods: This methodological study utilized data from an international, multicenter, randomized controlled trial of patients with low-risk prostate cancer (PC), conducted by NRG Oncology (NCT00331773). In addition to patient demographic and clinical data, this study utilized PRO data collected at baseline and 6, 12 and 24 months post-intervention. The Expanded Prostate Cancer Index Composite (EPIC) questionnaire measures health-related quality-of-life (HRQoL) and has four domains (urinary, sexual, hormonal, and bowel) and two subdomains per domain (function and bother); EuroQOL-5D-3L (EQ5D) captures health utilities. Ordinary Least Squares (OLS) regression models were used to map EPIC scores to EQ5D utilities in the baseline data through 5-fold cross-validation. Predictive performance was tested in the post-intervention data; predicted and reported utilities were compared using t-tests, and the absolute prediction error was modeled using fixed effects, as a function of baseline demographic and clinical covariates, as well as observed and predicted EQ5D utilities. Results: A total of 267 (199) patients had complete EQ5D and EPIC domain (or subdomain) data at baseline and all subsequent assessments. In the EPIC domain sample, mean ± standard deviation observed EQ5D utility was 0.90±0.13 at baseline, 0.92±0.11 at 6 months, 0.90±0.13 at 12 months and 0.89±0.14 at 24 months. Mean absolute differences (MDs) between reported and predicted were lower for models using EPIC subdomain data compared to EPIC domain data, and generally decreased as the time of assessment increased. The mapping functions over-predicted utilities for patients in perfect health while the prediction errors were increasingly negative for lower reported EQ5D scores. According to the fixed effects model for EPIC domain data, lower observed and predicted baseline EQ5D scores, and time of assessment were significant predictors of the absolute prediction error; for EPIC subdomain data, lower observed and predicted baseline EQ5D scores, hormonal bother and function, and bowel function significantly predicted the absolute prediction error. Conclusions: This study is the first to demonstrate the longitudinal validity of EPIC questionnaire, and builds upon existing research on longitudinal validity of mapping functions. The low MDs in prediction errors in post-intervention data indicate that the mapping functions are sensitive to treatment effect, thereby increasing confidence in their use in economic evaluations in PC.

Abstract PURPOSE Previous secondary analysis of NRG/RTOG 0933 provided hypothesis-generating data supporting a relationship between larger volumes of MR-determined pre-treatment WMI and developing neurocognitive decline following HA-WBRT. The current study examines the relationship between pre-treatment WMI and neurocognitive function (NCF) following WBRT+memantine +/-HA in a substantially larger cohort. METHODS NCF testing was performed at baseline,2,4,6,and 12 months post-WBRT, and included Hopkins Verbal Learning Test–Revised (HVLT-R), Trail Making Test (TMT) Parts A and B, and Controlled Oral Word Association (COWA). Pre-treatment WMI was measured by FLAIR volume corrected for whole brain volume and corrected for the FLAIR volume associated with metastases (FLAIR/(whole brain volume – metastasis FLAIR volume). Pearson correlation coefficients were used to assess association between pre-treatment WMI and change from baseline for each standardized NCF score. RESULTS Of 518 randomized patients, 442 (217,WBRT+Memantine; 225,HA-WBRT+Memantine) had WMI data and were included. In the entire cohort, mean FLAIR volume was 9.3cc (0.1-68.2cc), mean metastases FLAIR volume was 61.5cc (0-423.5cc), mean Whole Brain volume was 1336.4cc (949.4-2397.8cc). At 2 months, there were no significant correlations between neurocognitive test change scores and pre-treatment WMI volume. However, at 4 months, both HVLT-R Total Recall and TMT Part B change score and pre-treatment WMI volume were significantly negatively correlated on the HA-WBRT+Memantine arm (ρ=-0.22 p=0.042 and ρ=-0.27, p=0.013). At 12 months, both TMT Part A and TMT Part B change scores and pre-treatment WMI volume were significantly negatively correlated on the HA-WBRT+Memantine arm (ρ=-0.30, p=0.046 and ρ=-0.53, p< 0.001). CONCLUSIONS Pre-treatment WMI volume was a significant imaging-biomarker predictor of post-treatment neurocognitive decline at 4-and 12-months following HA-WBRT+Memantine. This suggests patients with greater pre-treatment WMI were more susceptible to neurocognitive decline, specifically when undergoing HA-WBRT, but not following standard WBRT. Dose heterogeneity inherent to HA-WBRT delivery may contribute to these findings and are hypothesis generating.

55 Background: To compare the predictive ability of health utility mapping algorithms derived using cross-sectional and longitudinal data specific to the Expanded Prostate Cancer Index Composite (EPIC). Methods: This mapping study utilized data from an international, multicenter, randomized controlled trial of patients with low-risk prostate cancer conducted by NRG Oncology (NCT00331773). Health-related quality-of-life (HRQoL) data were collected using EPIC, and health utilities were obtained using EuroQOL-5D (EQ5D) at baseline and 6, 12 and 24 months post-intervention. Data were split into an estimation sample (70%) and a validation sample (30%). Ordinary Least Squares (OLS) regression models were estimated using baseline cross-sectional data as well as pooled data from all assessment periods. Random effects (RE) specifications that explicitly model the longitudinal nature of the data were also estimated. Candidate models were selected based on root mean square error (RMSE). Results: A total of 196 (147) patients in the estimation sample had complete EQ5D and EPIC domain (subdomain) data at all time points. OLS models using combined data outperformed the counter-part RE models as well as OLS models using baseline data in the five-fold cross-validation. Addition of covariates to the models resulted in improved predictive ability. In the external validation, when only EPIC domain/ subdomain data are available, the OLS model using combined data predicted EQ5D utilities better than the counterpart RE model and OLS model using baseline data (RMSE=0.121108 & 0.111345). OLS model using baseline data outperformed other model types for algorithms with EPIC domains and demographics (RMSE=0.121757), while RE models outperformed the other two model types for algorithms with EPIC subdomains and demographic data, (0.112782) and for algorithms with EPIC domains/ subdomains, demographics, and clinical covariates (RMSE=0.123589 & 0.163093). Conclusions: While algorithms using pooled data outperformed other model types in internal validation, RE models showed better predictive ability in external validation for algorithms with covariates. Clinical trial information: NCT00331773.