Use of serum proteins identified by proteomics to predict prognosis in patients with metastatic renal cell cancer (mRCC) and comparison to the currently used MSKCC survival model

Abstract

11078 Background: In mRCC the Memorial Sloan-Kettering Cancer Center (MSKCC) risk model (Motzer et al.,JCO2002) is widely used for clinical trial design and patient management. To improve this model, we searched for serum proteins associated with overall survival (OS) using proteomics and determined their potential contribution to the composition of better prognostic models. Methods: Sera from 125 mRCC patients, mostly interferon-treated, collected between 2001–2006 in three Dutch Cancer Centers, were screened by SELDI-TOF mass spectrometry (MS). Identified proteins were analyzed for their association with OS by Cox regression analysis and Kaplan-Meier curves. Three proteins were subsequently validated with conventional ELISAs and immunoturbidimetry. Computed Cox PH regression models were individually compared with the Akaike's Information Criteria (AIC). The final model was statistically bootstrapped to strengthen its validity. Results: SELDI-TOF MS successfully detected eleven prognostic proteins associated with OS. The strongest prognostic proteins, apolipoprotein A-II (ApoA2), serum amyloid alpha (SAA), and transthyretin were validated by alternative methods and confirmed for their association with OS (p=7.4×10-9, p=3.2×10-8 and p=0.0002, respectively). Stepwise multivariable Cox regression analysis provided an optimal combination of prognostic factors; ApoA2, SAA, LDH, performance status and number of metastasis sites. Subsequent categorization of patients into three risk groups generated a novel survival model that robustly predicts patient prognosis (AIC=684 and p=5.8×10-15) with increased accuracy compared to the MSKCC model (AIC=719, p=2.0×10-7). Bootstrapping our model resulted in an optimism correction of 0.04 (R2=0.45). Initiating systemic therapy according to novel proposed model instead of the MSKCC model would result in a change of treatment in 14% of the patients. Conclusions: Using proteomics, serum proteins associated with OS in mRCC patients can be identified. Incorporating some of these factors together with traditional risk factors yields a prognostic model outperforming the MSKCC risk model thereby further improving patients’ management. No significant financial relationships to disclose.