Early dynamics of chronic myeloid leukemia on nilotinib predicts deep molecular response

Yuji Okamoto, Mitsuhito Hirano, Kai Morino, Masashi Kajita, Shinji Nakaoka, Mayuko Tsuda, Keiji Sugimoto, Shigehisa Tamaki, Junichi Hisatake, Hisayuki Yokoyama, Tadahiko Igarashi, Atsushi Shinagawa, Takeaki Sugawara, Satoru Hara, Kazuhisa Fujikawa, Seiichi Shimizu, Toshiaki Yujiri, Hisashi Wakita, Kaichi Nishiwaki, Arinobu Tojo, Kazuyuki AiharaNpj Systems Biology and Applications 2022 · DOI

TL;DR

This paper introduces a simple mathematical model that uses early dynamics of BCR-ABL1 transcript levels (IS) and white blood cell counts to predict whether chronic myeloid leukemia (CML) patients will achieve a deep molecular response (DMR) on nilotinib therapy. The model achieves high accuracy using only three initial time points, offering a valuable tool for personalized treatment adjustments.

What Problem Does This Paper Address?

Chronic myeloid leukemia (CML) is driven by the BCR-ABL1 tyrosine kinase. While targeted therapies like nilotinib have revolutionized CML treatment, individual patient responses vary. Achieving a deep molecular response (DMR) is crucial for long-term remission and potential treatment-free remission. However, current scoring systems and guidelines for predicting DMR are often retrospective, requiring at least 12 months of data, or lack sufficient discriminatory power. This leaves a critical gap in identifying patients who are unlikely to achieve DMR early on, preventing timely intervention or treatment adjustments. The authors aim to develop a method for early prediction of DMR using readily available clinical data.

Methods

The core of this work is a simple dynamical model that describes the time evolution of CML cells and normal white blood cells (WBCs) in peripheral blood under TKI therapy. The model uses two key features derived from the dynamics of BCR-ABL1 transcript levels (IS): the decreasing rate and the convergence value of IS. These features are then linked to the reduction rate and convergence value of CML cells.

The authors developed a mathematical framework to:

  1. Estimate model parameters: Using a simple set of differential equations, they model the exponential changes in normal and CML cell counts. They then use measured WBC counts and IS values to estimate the model's parameters, including the reduction rate and convergence value of CML cells.
  2. Classify patients: They identified that the reduction rate and convergence value of CML cells effectively distinguish between DMR and non-DMR patients. By plotting these two parameters, they found a clear separation.
  3. Predict DMR: They propose a classification method based on these parameters. Crucially, they demonstrate that these parameters can be estimated using only three initial time points (0, 3, and 6 months) of nilotinib therapy. They used leave-one-out cross-validation to determine optimal threshold values for predicting DMR, optimizing for accuracy, sensitivity, or specificity.

Datasets

The study utilized data from the N-road trial, a phase II, single-arm, open-label, multi-center clinical study for newly diagnosed chronic-phase CML patients treated with nilotinib. The dataset included 53 patients, of which 32 were eligible for this analysis due to complete data availability for WBC counts and IS measurements at specified time points.

Key Insights and Ideas

  • Early dynamics are predictive: The study highlights that the way IS levels decrease in the initial months of treatment, not just the absolute values, holds significant predictive power for achieving long-term DMR.
  • Simplicity is key: The authors demonstrate that a relatively simple dynamical model, relying on just two key features (reduction rate and convergence value of CML cells), can achieve high predictive performance, making it practical for clinical use.
  • Minimal data requirement: The ability to predict DMR using only three early time points (0, 3, and 6 months) is a significant advantage over existing methods that require much longer observation periods.
  • Model applicability: The proposed model is expected to be applicable to other TKIs like imatinib, but potentially not dasatinib due to its unique lymphocyte mobilization effect.

Key Results

  • The proposed mathematical model accurately reproduced the time series of IS for individual CML patients.
  • The model successfully classified DMR and non-DMR patients with 100% accuracy when using all available data points to estimate parameters.
  • Using only three initial time points (0, 3, and 6 months), the model achieved predictive performance comparable to or better than current scoring systems (EUTOS, ELTS) and ELN guidelines at 12 months.
  • The method achieved high specificity, meaning it was very good at correctly identifying patients who would not achieve DMR, which is crucial for timely treatment adjustments.
  • The authors' approach showed enhanced specificity and F1 scores compared to EUTOS and ELN guidelines.

Limitations and Future Work

  • Dasatinib exception: The model's applicability to dasatinib is questioned due to its unique lymphocyte dynamics, suggesting further investigation is needed for this specific TKI.
  • Drug resistance: The current model does not explicitly account for the acquisition of drug resistance, which can lead to IS reoccurrence. Further investigation with longer-term data might be needed to address this.
  • Outlier patients: The study identified outlier patients who did not fit neatly into DMR or non-DMR categories. Further clinical analysis of these outliers could reveal new characteristic features.
  • Future directions: The authors suggest employing more advanced machine learning methods for parameter optimization on larger datasets and exploring the prediction of other characteristic features of CML patients. They also note that predicting relapse might be another potential application of their framework.

Summary updated

Abstract

Chronic myeloid leukemia (CML) is a myeloproliferative disorder caused by the BCR-ABL1 tyrosine kinase. Although ABL1-specific tyrosine kinase inhibitors (TKIs) including nilotinib have dramatically improved the prognosis of patients with CML, the TKI efficacy depends on the individual patient. In this work, we found that the patients with different nilotinib responses can be classified by using the estimated parameters of our simple dynamical model with two common laboratory findings. Furthermore, our proposed method identified patients who failed to achieve a treatment goal with high fidelity according to the data collected only at three initial time points during nilotinib therapy. Since our model relies on the general properties of TKI response, our framework would be applicable to CML patients who receive frontline nilotinib or other TKIs.
BibTeX
@inproceedings{okamoto2022early,
  author = {Yuji Okamoto and Mitsuhito Hirano and Kai Morino and Masashi Kajita and Shinji Nakaoka and Mayuko Tsuda and Keiji Sugimoto and Shigehisa Tamaki and Junichi Hisatake and Hisayuki Yokoyama and Tadahiko Igarashi and Atsushi Shinagawa and Takeaki Sugawara and Satoru Hara and Kazuhisa Fujikawa and Seiichi Shimizu and Toshiaki Yujiri and Hisashi Wakita and Kaichi Nishiwaki and Arinobu Tojo and Kazuyuki Aihara},
  title = {{Early dynamics of chronic myeloid leukemia on nilotinib predicts deep molecular response}},
  booktitle = {Npj Systems Biology and Applications},
  year = {2022},
  doi = {10.1038/s41540-022-00248-3},
}
Figure