Development of a biomarker-based prediction model of disease progression during chemotherapy in children and adolescents

Abstract

Aims: This study aims to develop a quantitative semi-mechanistic model to describe the disease progression of leukemia during chemotherapy using circulating biomarkers in Korean children and adolescent population.

Methods: A routine clinical data set for 74 patients who were diagnosed as acute lymphoblastic leukemia (ALL) for the first time between the age of 1 – 19 were collected from Severance hospital electric medical records (EMR) system. Absolute lymphocyte count (ALC) and platelet count (PLT) were dependent variables. Age, WBC count, bone marrow transplantation (BMT) and other laboratory results such as creatinine, AST and ALT levels were possible covariates to be tested. A (semi) mechanistic model was used to describe ALC and PLT changes with time during chemotherapy. A K-PD model was used to describe drug kinetics as blood concentration data were not available. ALC and PLT production were described by a single compartment representing proliferative cells, 3 transit compartments representing biomarker maturation, and a single compartment representing biomarker concentrations in blood, where ALC production was assumed to be positively influenced by linear disease progression of ALL and reduced by chemotherapy and vice versa for PLT. Covariate search was then carried out for biomarkers. Population modeling approach was performed using NONMEM 7.3. 3 year overall survival analysis was performed using variables from biomarker disease progression models as predictors of overall survival of ALL patients.

Results: Due to data’s heterogeneity, patients treatment options were grouped into risk based protocol, standard risk (SR) and high risk (HR) based on the clinical practice. Drug effect was described by an Emax model of effect-site concentration obtained from K-PD model, where drug doses were BSA standardized due to the use of multiple cytotoxic drugs during the treatment. In addition, chemotherapy resistance was described the best with inhibitory Imax model where the chemotherapy exposure inhibits the drug effect. Estimated EC50 of standard and high risk groups are 0.00026 and 0.711 respectively in ALC, and 1.57 and 3.1 respectively in PLT. Estimated mean transit time (MTT) was 5.34 days and 17.0 days for lymphocyte and platelet respectively. Estimated gamma for ALC and PLT were 0.0973 and 0.215 respectively, which denotes the power of feedback component. For 3 year survival analysis of ALL patients, predicted latent disease slope of PLT at day 30 (DPLT), history of bone marrow transplantation (BMT) and predicted ALC/baseline ratio averaged over the period of day 1 to day 22 (RALC) were statistically significant predictors of 3 yr survival time in ALL patients in children and adolescents.

Conclusion: The proposed semi-mechanistic disease progression model with circulating biomarkers provide platforms to predict survival time of ALL patients which can be a powerful tool to help clinicians to monitor disease in ALL patients in children and adolescents.

목적 본 연구의 목적은 급성 림프구성 백혈병 소아 환자들에게서 바이오마커 기반 질병진행을 예측할 수 있는 모형을 모델링 하고 모델 매개변수 추정치와 환자의 공변량을 이용하여 환자의 생존 기간을 예측하여 향후 환자 개인별 특성을 고려한 급성 림프구성 백혈병 환자의 질병관리 또는 적절한 항암요법을 선택하는데 활용하고자 한다. 방법 본 연구는 후향적 연구로서 2000년 1월 1일 ~ 2015년 7월 31일의 기간 동안 연세의료원 세브란스병원에서 급성 림프모구 백혈병 소아암으로 진단받고 치료받은 환자들의 전자의무기록 (electronic medical record, EMR) 에서 수집된 총 74명의 자료를 분석하였다. 유효성 평가를 위한 biomarker 은 기존 연구에서 급성 림프모구 백혈병 예후인자로 알려진 혈소판 및 림프구 수치를 사용하였다. 질병이 진행됨에 따라 혈소판은 감소하고 림프구는 증가한다는 가정하에 항암치료 효과는 disease progression model로 모델링 하였다. 공변량 분석에는 stepwise 방법을 사용하여 환자의 나이, 백혈구 수, creatinine, AST, ALT 등의 유의성을 평가하였으며 forward inclusion 과 backward elimination 의 유의도는 각각 0.05와 0.01 로 설정하였다 37. 자료의 분석 및 모형 구축에는 비선형혼합효과 모델링에 기반을 둔 NONMEM 7.3 version (ICON Development Solutions, MD)이 사용되었으며 필요에 따라 Wings for NONMEM, Perl-Speaks-NONMEM (PSN), R 등의 프로그램이 사용되었다. 개발된 모형의 평가는 Visual predictive check (VPC) 를 통한 시뮬레이션 자료와 실제 자료 사이의 유사성으로부터 모형의 예측능력을 확인하는 것으로 진행하였다.31 아울러 생존 모형은 각 시간에서의 환자의 실제 생존 여부와 모형에서 예측된...