Background
Breast cancer is now the most common cancer in Chinese women, with cases accounting for 12.2% of all newly diagnosed breast cancers and 9.6% of all deaths from breast cancer worldwide [
1]. There is solid evidence supporting the value of diagnosing cancer early, and Western societies have produced guidelines on early detection [
2,
3]. Indeed, breast cancer screening by population-based mammography (MAM) has been proven to reduce mortality in several randomized trials in developed Western countries [
4,
5]. However, in developing countries, a screening strategy that combines clinical breast examination (CBE) and breast ultrasonography (USG) may be a more acceptable approach [
6,
7].
There is no nationwide screening program for breast cancer in China at present [
1], although population-based studies of CBE combined with diagnostic USG, MAM, or both are currently in progress [
1,
8]. Local governments have also sponsored community-based breast cancer screening programs in several urban cities despite doubts about the efficacy of CBE for early detection in diverse Chinese populations. Interestingly, a population-based study of these breast cancer screening programs provided good performance results, with a sensitivity and specificity of 70% and 90%, respectively [
9]. Nevertheless, breast cancer screening, especially when using CBE, may not be as effective in clinical settings as it is in trial settings [
10].
In Tianjin, the fourth largest city in China in terms of urban population, initial breast cancer screening is provided at community health services without the support of MAM or USG services from 2009.CBEs are done by trained health care providers funded by the local government as part of the basic public health service package. Women determined by their primary physicians to have a lesion that is suspicious or highly suggestive of a malignancy are then referred for further diagnostic tests and treatment, but these are not part of the basic public health service package and must be paid by patients or their medical insurance.
The government and public have high expectations of community health services to protect women’s health by detecting breast cancer early. However, community health services may not always be able to fulfill this role because of their limited diagnostic capacity and the shortage of doctors compared with the considerable number of women eligible for screening [
11]. This is compounded by the fact that there has been no report on the feasibility of the current breast cancer screening program.
The primary aim of a breast screening program should be to reduce mortality from breast cancer through early detection. Unfortunately, it would take decades to confirm the effectiveness of such a screening program based on mortality indicators alone. By contrast, quality assurance allows the use of alternative performance indicators for quality control and evaluation [
12‐
14]. There is a growing need to develop approaches that reflect the relationships between performance indicators and the feasibility of a screening strategy. Such an approach should help determine those factors that should be considered most important in practice, and should help set reasonable goals for the relevant indicators.
A decision-analytic model was used in the current study to predict the feasibility of a community-based breast cancer screening strategy in China. In addition, a sensitivity analysis approach was used to identify the relevant factors that significantly influenced the feasibility of such screening in community health services and to identify the optimum control ranges of those factors.
Discussion
In this study, we evaluated the effectiveness of an ongoing, community-based, breast cancer prevention program offered by community health services in Tianjin, an urban area in China. The cost-effectiveness analysis showed that the ongoing costs of this screening program were lower than the recommended threshold of triple the GDP per capita per QALY, and that they were cost effective regardless of whether a 1- or 3-year screening interval was used. In addition, it was shown that the annual screening strategy would still be cost effective at the triple the GDP per capita per QALY threshold when the age-specific cancer incidence was as low as that in rural areas. If the anticipated reduction in breast cancer mortality is to be achieved in reality, the following targets will probably need to be met: an attendance rate of at least 50%, compliance with transfer of at least 50%, and an incidence of stage I tumor of at least 10%.
Although the effectiveness of population-based breast cancer screening is of paramount importance, cost-effectiveness analyses are also necessary and play an important role in decision-making for public health policy. A conventional community-based screening strategy for the early diagnosis of breast cancer has been available in Tianjin China since 2005, and there are no data regarding long-term outcomes. The results of the current cost-effectiveness analysis, and the probabilistic approach in particular, show that screening at a one-year interval offers cost-effective when compared with no screening. Indeed, the acceptability curve of the Monte Carlo cost-effectiveness plane suggests that, versus no screening, the probability was 80% for cost effectiveness in the parallel mode given a 2 GDPs per QALY ceiling value with the screening distribution of stage I cancer being 40%. Given that the shortage of human resources in primary care may produce a bottleneck to annual breast cancer screening [
26,
27], our data indicate that screening every three years may be a reasonable alternative. In current study comparison only was made between strategies with different frequency and with only CBE as screening test. Strategies with different screening start-age and the other combinations of screen tests were not studied in this model simulation. Age distribution in China was normal with one age peak at 45–49 years, displaying differences from USA and Chinese American with two age peaks [
28]. Chinese women present with breast cancer at an earlier age. Kwong et al. reported that 17.6% of the women were younger than 40 years old, and age distribution was significantly different from women in the SEER database [
29]. Chinese Anti-Cancer Association recommended the starting age as 40 and the local government sponsored the basic public health service package including CBE for women aged 35–69 years. The additional comparisons show that screening strategies of a later starting age at 40 were more cost-saving compared with that with a younger starting age at 35 and could be more preferable for regions with limited human resources (Please see Additional file
1: Table S1).
It will only be many years after the introduction of a breast screening program that any potential reductions in breast cancer mortality can be expected [
30]. In the meantime, it is important that interim outcome measures are monitored to determine whether or not the program is performing satisfactorily. One suitable performance indicator that determines the outcome of a screening program is the attendance rate. The recommended targets for attendance were fulfilled in two previous national breast cancer screening projects in the urban areas of Chengdu (49.0%) and Mianyang (52.1%) in 2008 [
31]. However, a cross-sectional study in the same area showed a much lower attendance rate (31.9%). An equally disappointing participation rate (21.7%) was reported by the 2010 China Chronic Disease and Risk Factor Surveillance System, which included data from both urban and rural areas [
32]. Comprehensive and prioritized strategies are therefore needed to improve breast cancer screening participation and ensure its cost-effectiveness.
Women with positive CBE results were advised to undergo a combination of USG and MAM for diagnosis, even though these were not covered by the community breast cancer screening package. Compliance with these further investigations also affected the efficacy of the screening strategy. It was identified, that to ensure a probability of 95% for cost effectiveness of the annual screening strategy at the 3 GDP level of WTP, the quality assurance target would need to be 50%. In another breast cancer screening program with a similar design in Qibao Shanghai, about 30% of participants who should have undergone imaging did not [
33]. It is important to educate, train, and motivate referring clinicians in community health services of the importance of their role in enabling women to make informed decisions [
34,
35].
Finally, CBE was introduced as the initial method of screening for breast cancer in our community-based strategy [
36,
37]. Apart from the attendance and transfer compliance rates, the performance of CBE should also be considered relevant to the overall efficacy of the program [
36], especially given that measures of screening accuracy are particularly important interim indicators of effectiveness [
4]. In our study, we considered the rate of detection of earlier stage tumors to be a suitable indicator of screening efficacy, rather than the overall rate of cancer detection or the screening sensitivity. The annual screening strategy remained cost effective when detection rate of stage I tumor was assumed even lower (10%/20%) and the total proportion of other stages was set at 90%/80% with each proportion for stage II to IV simulated randomly change. Similar screening that relied on initial CBE in Shanghai indicated that a detection proportion of 29% for stage I tumors would be acceptable at a 3 GDP level of WTP [
33]. The performance of CBE, however, can only be assured when the involved clinical staff are sufficiently well trained and have appropriate knowledge of the principles of breast cancer diagnosis, management, and screening.
As a major limitation of the current study, it should be noted that any Markov decision model should be validated using external empirical data. However, the screening program still requires long-term follow up to provide this empirical data. To mitigate this, we took care to calibrate the analysis to fit local empirical observations, and most of the parameters assigned to the Markov cycle tree were derived from previous screening program [
9]. The Monte Carlo simulation was also done with 1000 runs to select values at random from appropriate probabilistic distributions of model parameters. Three performance indicators for quality assurance were identified in the current study using Monte Carlo simulation, and these might be useful as initial measures of program quality. Moving forward, continuous follow up of the target population is needed over an extended period of time to facilitate a long-term evaluation of its effectiveness [
38]. Over-diagnosis was always an important issue when a screening strategy was discussed. However, only over-diagnosis of DCIS was evaluated in current study. The screen-detected breast cancer cases had probability dying of other cause before being clinical occurrences that could be evaluated only when sojourn time was available. Arrospide, A et al. reported that 4% of screen-detected cancers were over-diagnosed [
39]. Another study presented that one breast cancer death prevented for about every three over-diagnosed cases identified and treated [
40]. When a screening program was discussed, both benefits and harms should be taken into accounted. It should also be noted that the time horizon was set at 50 years in current study and for women elder than 85, the model is not able to predict whether death will be due to BC or other causes that might influence the effectiveness of screening, but it will not cause significant changes. In addition, Markov cohort model was used to compare breast cancer screening strategies with different screening intervals. One important limitation should be noted was Markov cohort model couldn’t track individual patient histories which resulted in the fail to specify the benefit of screening over no screening in individual view. Another useful technique discrete event simulation modeling can track individual patient histories, such that each individual in the economic model can carry a large amount of information which can affect their future treatment options, risk of events and prognosis over time. However, the utilization of discrete event simulation required more data. Due to the lack of age distribution of preclinical phase onset and its mean duration, Markov cohort model was chosen over discrete event simulation. It should be mentioned that Markov cohort model might resulted better estimates for the decision making of the health care resource allocation [
41,
42].