Personalized screening policies for cervical cancer prevention

Official URL: https://risc01.sabanciuniv.edu/record=b2579566_(Table of contents)

Since the introduction of the first cervical cancer screening method, i.e., Pap smear, the disease burden has declined steadily and the outlook for women with cervical cancer has improved significantly. However, the reduction in mortality and incidence of the disease has not been uniform across the globe. While in the developed countries the disease is under control, in many low resource settings, the disease continues to inflict major health problems. Through organized screening programs, the abnormal changes in cervical cells can be detected at earlier and more treatable stages. Yet, determining the optimal frequency and type of screening test remains a challenge for the policy makers who want to balance the benefits and costs. To overcome this challenge, in many countries screening policies are presented in the form of general guidelines that dictate the frequency for the whole population, regardless of their risk profile. Essentially, since guidelines do not have the flexibility to address the individual patient’s risk profile prior to a screening decision, they fail to distinguish between different types of patients. Recently, personalized medicine (PM) has been adopted to answer this differentiation problem. Personalized medicine links the disease condition of an individual to the risk profile which helps to design a targeted treatment for a patient’s specific health condition. The application of PM has recently been extended to the screening decisions for many cancer types including cervical cancer. In this study, we propose a personalized partially observable Markov decision process (POMDP) approach to address the questions regarding the screening type and frequency at different ages for cervical cancer patients with different risk profiles. As a sequential decision making tool, POMDPs provide a rich framework to study further questions such as the impact of the patient’s risk profile and the false outcomes of the screening tests on the screening decisions. Specifically, we propose two POMDP models that incorporate the risk of the disease as well as the effect of test characteristics on the screening decisions. The first model considers a reduced state space with cotesting of cytology and HPV-DNA testing as a single combined test for screening as opposed to waiting. This constitutes a small model which can be solved exactly. The second model, uses cytology and HPV-DNA testing as two separate decisions along with the wait decision. We also expand the state space of the model for accurate representation of the natural history of the disease. Due to the large size of this model, exact methods are not applicable, and we propose an approximate solution method to solve the problem. Our results suggest that screening the whole population with the same frequency leads to health disparities and that screenings need to be tailored to the specific risk profile of the patients to improve the expected quality adjusted life years (QALYs) of the patients. Our results also confirm that screening programs will experience a paradigm shift towards HPV-DNA testing as the standalone testing for cervical cancer in accordance with what is proposed in the literature.