Evaluation of breast cancer mammography screening policies considering adherence behavior

Highlights

• We incorporate women's nonstationary adherence behavior into breast cancer screening decision modeling.

• We develop a partially observable Markov chain model to incorporate imperfectness of screening tests.

• Both screening mammography and breast self examination are considered as methods of breast cancer detection.

• Interval cancer is incorporated in the model.

• Our results can facilitate healthcare providers to tailor screening recommendations for patients.

Abstract

Mammography is known to be the most effective way of breast cancer detection. The efficacy of mammography screening guidelines is highly associated with women's compliance with these recommendations. Currently, none of the existing policies takes women's behavior into consideration; instead, perfect adherence is assumed. However, an earlier longitudinal data analysis has revealed that women's compliance with mammography guidelines remained low in recent years. In this study, we develop a randomized discrete-time finite-horizon partially observable Markov chain model to evaluate a wide range of screening mammography policies, incorporating heterogeneity in women's adherence behaviors. Considering potential harms of mammography tests (e.g., risk of developing radiation-induced breast cancer, false negatives, false positives, etc.), policies with varying starting age, ending age and frequency of screening mammograms at different ages are compared in terms of total quality adjusted life years (QALYs) and lifetime breast cancer mortality risk. Our results show that women with perfect adherence do not always experience higher QALYs. In fact, for some policies, including the American Cancer Society (ACS) policy, the general population with various adherence levels has higher QALYs than women with perfect adherence. However, in terms of lifetime breast cancer mortality risk, higher/perfect adherence always results in lower risk of dying from breast cancer. This implies that the benefits of mammography in decreasing death from breast cancer outweigh the increased risk of developing radiation-induced breast cancer from mammographic screening. This study can facilitate healthcare providers to tailor screening mammography recommendations based on their patients' estimated adherence levels.

Introduction

Breast cancer is the most common cancer worldwide in women contributing more than 25 percent of the total number of cancer cases diagnosed in 2012 (World Cancer Research Fund International, 2014). It is also the most common non-skin cancer affecting women in the United States. According to the American Cancer Society, about one in eight US women will develop breast cancer in her lifetime. In 2015, an estimated 231,840 invasive breast cancer cases will be diagnosed, and about 40,290 will die from breast cancer (American Cancer Society, 2015).

Mammography plays an important role in detection of breast cancer when the cancer is in its early stage. It is known to be an effective method and the current standard practice for early diagnosis of breast cancer. On average, mammography can detect breast cancer 1.7 years before a woman can feel a lump in her breast (Center for Disease Control and Prevention, 2013). In addition, it can reduce breast cancer mortality rate by a reasonable estimate of 15 percent (Gøtzsche & Nielsen, 2009).

However, mammography screening recommendations have been the subject of significant debate in recent years.  There are varying screening guidelines from different organizations about when to start and end mammograms and how frequently a woman should undergo mammography screenings. For example, the American Cancer Society (ACS), the department of Health and Human Services (HHS), the American Medical Association (AMA), and the American College of Radiology (ACR) recommend screening mammography every year, beginning at age 40. In 2009, the U.S. Preventive Services Task Force (USPSTF) issued revised screening mammography guidelines, which resulted in a significant controversy. According to the USPSTF guidelines, screening mammograms should be done every two years between age 50 and 75 for women at average risk of breast cancer (The US Preventive Services Task Force, 2013).

Although mammography is known to be one of the most effective methods of detecting breast cancer, there are many concerns on its adverse effects on women's quality of life. Unreliability of mammography, i.e., false negative and false positive results and over-diagnosis are among the deficiencies of mammography (Epstein, Bertell, & Seaman, 2001). A prospective cohort study of 7 mammography registries of the Breast Cancer Surveillance Consortium (BCSC) showed that the adjusted sensitivity increased with age from 68.6 percent in women ages 40–44 to 83.3 percent in women ages 80–89 (Breast Cancer Surveillance Consortium, 2013). Similarly, the specificity rate increased from 88.2 percent for women aged 40 through 44 years to 93.4 percent in women older than 75 (Carney et al., 2003). Moreover, about one-third of all aggressive cancers are diagnosed in the interval between successive annual mammograms (Epstein, Steinman, & LeVert, 1998). These cancers are known as interval cancers and are defined as cancers detected within 12 months after a negative mammogram (Burhenne et al., 1994). Some of the interval cancers are present at the time of mammography screening (false-negatives) while others grow rapidly in the interval between a mammogram and detection (Hébert-Croteau, Théberge, Langlois, Major, & Brisson, 2005). Generally, interval cancers grow rapidly and are frequently diagnosed at advanced stages (Caumo et al., 2010). In addition, studies have shown that receiving mammograms increases a woman's chance of developing breast cancer due to exposure to radiation (Epstein, Bertell and Seaman, 2001, Yaffe and Mainprize, 2011, Gøtzsche and Nielsen, 2009). Each radiation-absorbed dose (rad) of exposure increases breast cancer risk by 1 percent (Epstein et al., 2001). Moreover, Bleyer and Welch (2012) also found that up to a third of diagnosed breast cancers are overdiagnosed cases and do not need treatment.

Various studies have been conducted to evaluate and compare different screening policies or to identify optimal policies. Kirch and Klein (1974) developed an inspection strategy for the detection of an age-dependent disease with the objective of minimizing detection delay. To illustrate their methodology, they developed optimal schedules for breast cancer examinations. Shwartz (1978) developed a mathematical model to evaluate life expectancy gain associated with different screening policies under different assumptions about the rate of disease progression, the characteristics of the screening technique, and the frequency of screenings. Parmigiani (1993) presented a continuous time non-Markovian stochastic process model of disease progression to analyze which age groups and what part of the population should undergo breast cancer screening, while minimizing the total expected loss or risk including financial costs, side effects, wasted time, stress due to false-negative test results, etc. Zelen (1993) formulated a screening scheduling problem to maximize a weighted utility function in a continuous time setting. In Zelen's model all the parameters were assumed to be stationary except for the incidence rate. Baker (1998) used a mathematical parametric model to assess different screening policies in terms of minimizing the cost of cancer plus the cost of carrying out any screening. Baker's model was developed based on a small data set used to estimate the model parameters, and compares a small set of cost optimal policies under different sets of constraints. Maillart, Ivy, Ransom, and Diehl (2008) formulated a partially observable Markov chain model to evaluate a wide range of dynamic mammography screening policies as well as current practices. They compared different policies in terms of the resulting lifetime breast cancer mortality risk and the expected number of mammograms women should undergo for each policy and generated a frontier of efficient policies. In their formulation, they considered age-based dynamics of breast cancer (i.e., increasing incidence, decreasing aggression). They also incorporated the imperfect nature of the screening outcomes and dynamics of test result accuracy (increasing sensitivity and specificity rates with age). Ayer, Alagoz, and Stout (2011) developed a finite-horizon partially observable Markov decision process (POMDP) model to determine the optimal personalized mammography screening strategy based on personal risk characteristics of women such as their prior screening history. Ahern, Cheng, and Shen (2011) developed a mathematical model to optimize cancer screening schedules, taking into account the trade-off between the benefits and costs of screenings in their proposed utility function. They considered two different optimization frameworks: optimize the number of screening examinations with equal screening intervals between exams but without a pre-fixed total cost; and optimize the ages at which screening should be given for a fixed total cost and prove that the optimal solution exists under each of the two frameworks. Li, Zhu, Klein, and Kong (2014) developed a finite-horizon discrete-time partially observable Markov chain model to assess colonoscopy screening strategies in terms of the expected cumulative quality-adjusted life years (QALYs) and the expected cumulative cost. They also calculated the incremental cost-effectiveness ratio for different screening policies.

However, none of the abovementioned studies has taken patient behavior into consideration. They all assume that patients adhere to the guidelines perfectly and undergo the prescribed screening mammograms. This is also true for the current screening recommendations which are based on the assumption of 100 percent adherence to the guidelines. However, not all women have the same attitudes toward breast cancer screening. A recent study by the Centers for Disease Control and Prevention (CDC) revealed that about 67.1 percent of women aged 40 and older had one mammogram between 2008 and 2010 (Center for Disease Control and Prevention, 2014). In another longitudinally data analysis of the BCSC data registry, more than 75 percent of women have had less than five mammograms in the 14-year interval, between 1996 and 2009, which implies a low compliance with the current guidelines (Breast Cancer Surveillance Consortium, 2013).

There are a limited number of studies taking individual adherence behavior into consideration. Brailsford, Harper, and Sykes (2012) used a three-phase discrete event simulation to model breast cancer and screening policies incorporating women's behavioral factors in their model. They assigned behavioral attributes to each simulated woman to control her compliance with the prescribed mammograms in their model. They compared a limited number of screening policies, including the current UK policy, in terms of the number of screen detected cancers, and life yeas saved. In another study, Ayer, Alagoz, Stout, and Burnside (2015) analyzed the role of behavioral heterogeneity in women's adherence on optimal mammography screening recommendations.

In this research, incorporating women's adherence behavior to mammography recommendations, a mathematical framework is proposed to evaluate and compare various screening policies in terms of QALYs and the lifetime mortality risk of breast cancer. Our study is different from the two studies discussed above in the approach of incorporating patient adherence to the screening tests. In contrast to earlier studies, we allow uncertainty in a patient's adherence probabilities in the model. Adherence is also a function of the length of the interval between two subsequent screenings. Moreover, assuming a patient's adherence behavior is related to and can be estimated from her perception/planning toward mammography, not only can we evaluate and compare screening policies for a specific patient knowing her personal characteristics, but we can identify efficient policies for the general population as well. In addition, we incorporate in our model the possibility of interval cancer detection, and the increased risk of developing breast cancer due to exposure to X-ray radiation during a mammography test.

The remainder of this paper is organized as follows. In Section 2, the proposed model is presented. In Section 3, model inputs and parameters estimation, and computational results are presented. Finally, Section 4 summarizes the findings and discusses future work.