Principle investigator:
The notion of “principle investigator” is not applicable in our case.
We performed no studies with human subjects/patients. All data are
derived from published clinical studies.
Word count: 4000 words.
Table count: No tables in the main text.
Figures count: 8 figures.
Statement 1 : What is already known about this subject:
- Severity of haematotoxicity under chemotherapy is heterogeneous among
patients and difficult to predict
- Biomathematical models of thrombopoiesis were developed and showed
potential to describe individual therapy courses
- A tool for neutrophil-guided dose adaptation in chemotherapy based on
a semi-mechanistic haematopoiesis model is available
Statement 2 : What this study adds:
- We established a novel quantitative and clinically relevant method for
comparing prediction performance of different biomathematical models
of thrombopoiesis under chemotherapy.
- We applied this method to compare the individual prediction
performance of a comprehensive mechanistic and a semi-mechanistic
model regarding thrombocytopenia of grades III-IV.
- We analysed the contribution of baseline covariates to prediction
performance of the models using an independent validation data set
- We showed that the comprehensive mechanistic model has superior
predictive accuracy compared to the semi-mechanistic model
Summary
Aims: Thrombocytopoenia is a common major side-effect of
cytotoxic cancer therapies. A clinically relevant problem is to predict
an individual’s thrombotoxicity in the next planned chemotherapy cycle
in order to decide on treatment adaptation. To support this task, two
dynamical mathematical models of thrombopoiesis under chemotherapy were
proposed, a simple semi-mechanistic model and a comprehensive
mechanistic model. In this study, we compare the performance of these
models.
Methods: We consider close-meshed individual time series data
of 135 non-Hodgkin’s lymphoma patients treated with six cycles of
CHOP/CHOEP chemotherapies. Individual parameter estimates were derived
on the basis of these data considering a varying number of cycles per
patient. Parsimony assumptions were applied to optimize parameter
identifiability. Models are compared by determining deviations of
predicted and observed degrees of thrombocytopoenia in the next cycles.
Results: The mechanistic model results in superior fits of
individual time series data. Moreover, prediction accuracy of future
cycle toxicities by the mechanistic model is higher even if it used data
of two cycles, while the semi-mechanistic model used data of five cycles
for the corresponding calibrations.
Conclusions: We successfully established a quantitative and
clinically relevant method for comparing prediction performance of
biomathematical models of thrombopoiesis under chemotherapy. We showed
that the more comprehensive mechanistic model outperforms the
semi-mechanistic model. We aim at implementing the mechanistic model
into clinical practice to assess its utility in real life clinical
decision making.
Introduction
Thrombocytopoenia is a major side effect of many anti-cancer cytotoxic
drugs and is dose-limiting in some occasions . However, severity of this
condition is highly heterogeneous between patients with the effect that
a possibly small proportion of patients expressing excessive toxicity
limits the dosage for the entire patient population for safety reasons .
It is therefore of clinical importance to predict thrombotoxicity on an
individual level and to develop and apply personalized schedules of
chemotherapy and ameliorative treatment such as platelet transfusion or
growth factor applications.
A practically applied method to personalize chemotherapy are dose
adjustments which could be done either a priori or a
posteriori. A priori dose adjustment procedures are based on
pharmacogenetic, demographic and clinical information with known
predictive power regarding cytotoxic effects. However, although several
of such risk factors are known, their predictive power is poor, still
resulting in high heterogeneity within risk groups . Since current
chemotherapy regimens are based on multiple cycles , aposteriori dose adjustment referring to therapy control on the
basis of observed toxicity could be a promising alternative. A
posteriori Bayesian-guided dosing for several anticancer drugs was
applied by different groups as reviewed in Rousseau et al . Based on a
semi-mechanistic model of haematopoiesis developed by Friberg et al. , a
Bayesian a posteriori strategy of neutrophil-guided dose
adaptation was proposed and a predictive tool was developed . However,
it was shown, that inter-occasion variability (IOV) reduced
effectiveness of dose adaptations . The authors treated IOV as a purely
random effect affecting some of the individualized parameters. However,
IOV of haematotoxicity could result from (non-random) long-range
cumulative chemotherapy effects. To account for this effect,
modifications of the Friberg model were proposed . Among these, the
model of Henrich et al was designed to consider cumulative
chemotherapy-induced haematotoxicity, based on slow bone-marrow
exhaustion .
We recently proposed a comprehensive model of human thrombopoiesis under
chemotherapy and developed a method to fit it to individual patient data
using a Bayesian approach which exploits a large body of literature
information including patient data of other studies . The major aim of
the present work is to propose a framework to assess, improve and
compare the performance of different models regarding prediction of
next-cycle thrombocytopoenia at an individual level. We apply this
framework to compare our model with that of Henrich et al .
Materials and methods
Modelling concepts to be
compared
We compared the predictive potential of our comprehensive mechanistic
thrombopoiesis model with the semi-mechanistic model of general
haematopoiesis and bone marrow exhaustion proposed by Henrich et al .
Figure 1 provides and overview of both modelling concepts.