Multivariate screening of prognostic factors to identify a novel, simple, and objective marker to optimize patient selection and predict survival benefit in early-phase trials.

Abstract

2608 Background: Several prognostic indices have been devised to optimize patient selection for phase I trials. However, there is no consensus as to the optimal score and none qualifies as a marker of response. We developed a simple score through a multivariate screening of individual variables and compared its performance with existing prognostic scores including the Royal Marsden, Nijmegen and Prince Margaret Hospital scores. Methods: We retrospectively analyzed characteristics and outcomes of 120 referrals to our phase I center (2007 - 2011). Independent predictors for overall survival (OS) were identified from univariate (Kaplan Meier) and multivariate (Cox regression) analyses and used to create the Hammersmith Hospital score (HS). This was compared with the other indices for predicting progression free survival (PFS), OS and 90 day mortality (90 DM). Multivariate logistic regression and ROC curves were used to estimate 90 DM and c-index was used to estimate the prognostic ability of the different indices. Changes in HS following treatment (ΔHS) were calculated at 6 weeks RECIST in a subset of patients receiving targeted therapies (n= 50). Results: Median age was 62 years (range: 28 – 80); median OS 4.3 months (range: 0.2 – 39); 34% male; 25% PS>1. Multivariate screening identified albumin <35 g/L, lactate dehydrogenase (LDH) >450 U/L and sodium <135mmol/L as the strongest independent predictors of OS (p<0.05). These were entered into a 3-point score (HS) that classifies patients as being at high risk (score 2-3) vs. low risk (score 0-1) of worse OS (HR= 5.8, p<0.001), PFS (HR= 2.7, p= 0.04) and 90 DM (OR= 5.9, p= 0.001). All scores predicted for OS on univariate analysis (p<0.05). On multivariate analysis HS performed best to predict OS (HR= 3.9, p<0.001 C-index score= 0.71), PFS (HR= 2.6 p= 0.04) and 90 DM with area under the ROC curve 0.71. ΔHS independently predicted for OS (p<0.001), with worsening of the score reflecting poorer OS. Prospective validation is ongoing. Conclusions: HS is a simple score that outperforms other prognostic indices. This can be used to select individuals for future studies as well as an additional marker of response.