PB1786 IMMUNOHISTOCHEMICAL TYPE ADJUSTED‐INTERNATIONAL PROGNOSTIC INDEX (I‐IPI) IS BETTER FOR PROGNOSIS OF OUTCOME THAN STANDARD IPI FOR PATIENTS WITH DIFFUSE LARGE B‐CELL LYMPHOMA

Abstract

Background:Diffuse large B‐cell lymphoma (DLBCL) is a heterogeneous disease with a wide range of outcomes. The different prognostic systems for predicting the overall survival (OS) have been developed for last 23 years. The IPI has been created in pre‐rituximab era, others (NCCN‐IPI, GELTAMO‐IPI, etc.) – in rituximab era. However, these prognostic indexes do not explore immunohistochemical markers such as GCB or non‐GCB types of DLBCL. In the same time some researchers confirmed the important prognostic role of immunohistochemical markers for DLBCL outcome. In clinical practice physicians often encounter with different characteristics of DLBCL including stage, age, extranodal sites involved, routine lab test and immunohistology results. All of them may be a part of the disease prognostic systems.Aims:The present study shows the creation and significance of new prognostic system for stratification DLBCL outcomes based on combination the standard IPI index and GCB / non‐GCB types.Methods:A total of 104 patients with a median age of 58 years (range, 28‐83) were included. All patients received R‐CHOP chemotherapy as a first line. The clinical outcomes were analyzed according to IPI together with Hans algorithm for immunohistochemical type of DLBCL. To generate the I‐IPI, we used a machine learning method which called Classification and Regression Tree (CART). This advanced statistical method allowed to find significant cut‐off points for predictors for optimal prognosis of OS. As predictors we considered group of risk from standard IPI (high, intermediate‐high, intermediate‐low, low), GCB‐type and non‐GCB‐type of DLBCL. In turn OS was assed using Kaplan‐Meier method and compared between obtained groups using log rank‐test. Hazard ratio (HR) was evaluated using cox‐regression model.Results:Thirty‐one percent of the cohort had GCB type of disease, 49% ‐ non‐GCB type. According to IPI, 24%, 41%, 15% and 20% patients were categorized as low, intermediate‐low, intermediate‐high and high risk, respectively. All the patients received R‐CHOP regimen as first line with a median follow up of 24 months (range, 1‐150). We estimated and compared OS according IPI and Hans algorithm. The 5‐year OS in GCB and non‐GCB groups were 85% and 58%, respectively (HR 6.8, 95%CI: 1.29‐29.3). The 2‐year OS according to IPI among low, intermediate‐low, intermediate‐high and high‐risk groups were 87%, 88%, 79% and 43%, respectively (log‐rank test, p = 0.00017). However, there wasn’t statistical significance between all groups excluding high‐risk. At the next step we used machine learning algorithm for creation combined prognostic model. It allowed to stratification patients in three groups with statistical significance differences in OS (Figure 1). So, we have got 3 classes according new I‐IPI: high (equals the high risk by standard IPI), intermediate (equals the non‐high risk by standard IPI + non‐GCB type) and low (equals the non‐high risk by standard IPI + GCB type). The 2‐year OS according to I‐IPI were 100% (in patients with low risk), 68% (in patients with intermediate risk) and 46% (in patients with high risk), log‐rank test, p < 0.0001. The HR for high risk was 5.1 (95%CI 2.1‐12.1).Summary/Conclusion:I‐IPI is a better model in predicting OS for DLBCL treated with R‐CHOP than standard IPI. The model is can predict the risk disease more discriminately which would suggest the tailor therapy approach. For instance, the I‐IPI is better in predicting patients with high risk in whom more effective than R‐CHOP treatment, is warranted to improve the cure rate.image