Prediction of early death among patients enrolled in phase I trials: development and validation of a new model based on platelet count and albumin

Abstract

Background:Selecting patients with ‘sufficient life expectancy’ for Phase I oncology trials remains challenging. The Royal Marsden Hospital Score (RMS) previously identified high-risk patients as those with ⩾2 of the following: albumin <35 g l−1; LDH > upper limit of normal; >2 metastatic sites. This study developed an alternative prognostic model, and compared its performance with that of the RMS.Methods:The primary end point was the 90-day mortality rate. The new model was developed from the same database as RMS, but it used Chi-squared Automatic Interaction Detection (CHAID). The ROC characteristics of both methods were then validated in an independent database of 324 patients enrolled in European Organization on Research and Treatment of Cancer Phase I trials of cytotoxic agents between 2000 and 2009.Results:The CHAID method identified high-risk patients as those with albumin <33 g l−1 or ⩾33 g l−1, but platelet counts ⩾400.000 mm−3. In the validation data set, the rates of correctly classified patients were 0.79 vs 0.67 for the CHAID model and RMS, respectively. The negative predictive values (NPV) were similar for the CHAID model and RMS.Conclusion:The CHAID model and RMS provided a similarly high level of NPV, but the CHAID model gave a better accuracy in the validation set. Both CHAID model and RMS may improve the screening process in phase I trials.