Precision medicine: Statistical methods for estimating adaptive treatment strategies

To address early and late treatment failures in older patients with diffuse large B-cell lymphoma (DLBCL), we designed a two-stage randomized trial of cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) versus rituximab plus CHOP (R-CHOP), with a second random assignment to maintenance rituximab (MR) or observation in responding patients. Untreated DLBCL patients who were 60 years or older were randomly assigned to R-CHOP (n = 318) or CHOP (n = 314); 415 responders were randomly assigned to MR (n = 207) or observation (n = 208). The primary end point was failure-free survival (FFS). All P values were two sided. Three-year FFS rate was 53% for R-CHOP patients and 46% for CHOP patients (P = .04) at a median follow-up time of 3.5 years. Two-year FFS rate from second random assignment was 76% for MR compared with 61% for observation (P = .009). No significant differences in survival were seen according to induction or maintenance therapy. FFS was prolonged with MR after CHOP (P = .0004) but not after R-CHOP (P = .81) with 2-year FFS rates from second random assignment of 77%, 79%, 74%, and 45% for R-CHOP, R-CHOP + MR, CHOP + MR, and CHOP, respectively. In a secondary analysis excluding MR patients, R-CHOP alone reduced the risks of treatment failure (P = .003) and death (P = .05) compared with CHOP alone. Rituximab administered as induction or maintenance with CHOP chemotherapy significantly prolonged FFS in older DLBCL patients. After R-CHOP, no benefit was provided by MR. These results, which are consistent with an additive effect of rituximab, suggest that future studies could focus on maintenance strategies with novel agents as well as new induction therapies.

There is increasing interest in discovering individualized treatment rules for patients who have heterogeneous responses to treatment. In particular, one aims to find an optimal individualized treatment rule which is a deterministic function of patient specific characteristics maximizing expected clinical outcome. In this paper, we first show that estimating such an optimal treatment rule is equivalent to a classification problem where each subject is weighted proportional to his or her clinical outcome. We then propose an outcome weighted learning approach based on the support vector machine framework. We show that the resulting estimator of the treatment rule is consistent. We further obtain a finite sample bound for the difference between the expected outcome using the estimated individualized treatment rule and that of the optimal treatment rule. The performance of the proposed approach is demonstrated via simulation studies and an analysis of chronic depression data.

For patients with relapsed or refractory aggressive lymphoma, we hypothesized that gemcitabine-based therapy before autologous stem-cell transplantation (ASCT) is as effective as and less toxic than standard treatment.