Within-Host Bayesian Joint Modeling of Longitudinal and Time-to-Event Data of Leishmania Infection

The host immune system plays a significant role in managing and clearing pathogen material during an infection, but this complex process presents numerous challenges from a modeling perspective. There are many mathematical and statistical models for these kinds of processes that take into account a wide range of events that happen within the host. In this work, we present a Bayesian joint model of longitudinal and time-to-event data of Leishmania infection that considers the interplay between key drivers of the disease process: pathogen load, antibody level, and disease. The longitudinal model also considers approximate inflammatory and regulatory immune factors. In addition to measuring antibody levels produced by the immune system, we adapt data from CD4+ and CD8+ T cell proliferation, and expression of interleukin 10, interferon-gamma, and programmed cell death 1 as inflammatory or regulatory factors mediating the disease process. The model is developed using data collected from a cohort of dogs naturally exposed to Leishmania infantum. The cohort was chosen to start with healthy infected animals, and this is the majority of the data. The model also characterizes the relationship features of the longitudinal outcomes and time of death due to progressive Leishmania infection. In addition to describing the mechanisms causing disease progression and impacting the risk of death, we also present the models ability to predict individual trajectories of Canine Leishmaniosis (CanL) progression. The within-host model structure we present here provides a way forward to address vital research questions regarding the understanding progression of complex chronic diseases such as Visceral Leishmaniasis, a parasitic disease causing significant morbidity worldwide. Author SummaryThe immune system is complex and its effectiveness against infection depends on a variety of host and pathogen factors. Despite numerous studies of Leishmania parasite infections, researchers are still discovering new connections between immune system components with hopes of better understanding how the immune system functions during Leishmania infection. The development of tools for understanding, preventing, and predicting Leishmania infection outcomes is the main goal of this work. We present a computational model made using field-collected data during canine Leishmania infections. The model considers the interplay between three main components: parasite load, antibody level, and disease severity. The model explores how key inflammatory and regulatory elements of the immune response affect these main components, including T cell proliferation and important cytokine expressions such as protective interferon-gamma (IFN-{gamma}) or inhibitory interleukin 10 (IL-10) [1]. Although the induction of CD4+ T helper 1 cell responses is considered essential for immunity against Leishmania, B cells and the production of Leishmania-specific antibodies have also been proposed to play an important role in disease progression [2]. In a simpler model, Pabon-Rodriguez et. al. [3] showed antibody levels are dependent on pathogen load and canine Leishmaniasis (CanL) disease presentation. These high levels of Leishmania specific antibodies are observed in subjects with visceral Leishmaniasis (VL) and other severe forms of Leishmanial disease, and there is accumulating evidence that B cells and antibodies correlate with pathology [4]. In Section 1, we introduce Canine Leishmaniasis and discuss the importance of host-pathogen interaction with the immune response. Next, in Section 2, we introduce the data collection study, the variables utilized in this model, and define the clinical signs of Leishmania infection. In addition, this section explains how the presented model was constructed based on different techniques. A summary of model parameters, model implementation details, convergence diagnostics, and sensitivity analysis are also included. In Section 3, we provide summary results of how different model variables interact with one another and disease progression forecasts. In Section 4, we discuss the results and provide further recommendations and considerations.