Economic evaluations of mammography to screen for breast cancer in low- and middle-income countries: A systematic review

Background Low- and middle-income countries (LMICs) have limited resources compared to high-income countries (HICs). Therefore, it is critical that LMICs implement cost-effective strategies to reduce the burden of breast cancer. This study aimed to answer the question of whether mammography is a cost-effective breast cancer screening method in LMICs. Methods A systematic article search was conducted through Medline, Embase, Web of Science, and Econlit. Studies were included only if they conducted a full economic evaluation and focused on mammography screening in LMICs. Two reviewers screened through the title and abstract of each article and continued with full-text selection. Data were extracted and synthesized narratively. Quality assessment for each included study was conducted using the Consensus Health Economic Criteria (CHEC) extended checklist. Results This review identified 21 studies economically evaluating mammography as a breast cancer screening method in LMICs. Eighteen of these studies concluded that mammography screening was a cost-effective strategy. Most studies (71%) were conducted in upper-middle-income countries (Upper MICs). The quality of the studies varied from low to good. Important factors determining cost-effectiveness are the target age group (eg, 50-59 years), the screening interval (eg, biennial or triennial), as well as any combination with other breast cancer control strategies (eg, combination with treatment strategy for breast cancer patients). Conclusions Mammography screening appeared to be a cost-effective strategy in LMICs, particularly in Upper MICs. More studies conducted in lower-middle-income and low-income countries are needed to better understand the cost-effectiveness of mammography screening in these regions.

Economic evaluations of mammography to screen for breast cancer in low-and middle-income countries: A systematic review In this study, we systematically searched MEDLINE, EMBASE, EconLit, and Web of Science on January 30, 2020. We updated the search on June 16, 2021, and October 20, 2021. These are the four main databases recommended by van Mastrigt et al. for searching economic evaluation studies [21]. The search strategies were adapted for each database, which included the terms relating to or describing the population, intervention, and study design of our review, as follow: "breast cancer" AND "mammography" AND "screening" AND "economic evaluation" AND "LMIC". The full search strategy is available upon request. Advanced search and "exploding" a search term for retrieving all records were done with not only the exact term, but also those in the hierarchy of the controlled vocabulary using medical subject headings (MESH) in Pubmed and EMTREE in EMBASE. We did not limit the search by language or year. The search was also expanded by identifying studies from the reference lists of identified relevant studies.

Extracted information
We extracted the following characteristics from the included studies: country or region (including LMIC classification), objective, study population, intervention, and comparison. We also documented the following methodological characteristics: type of economic evaluation (eg, cost-effectiveness analysis, cost-utility analysis, other types), study design (eg, experimental, observational, model-based, etc.), perspective, time horizon, currency, currency year, and outcome measure for effectiveness, such as disability-adjusted life years (DALY), quality-adjusted life years (QALY), life-years gained (LYG) and intermediate outcome measures.
Additionally, we documented the following details: discount rates, reported incremental analysis (incremental effectiveness, incremental cost, average cost-effectiveness ratio, incremental cost-effectiveness ratio), and type of applied sensitivity analysis (one-way sensitivity analysis, probabilistic sensitivity analysis, scenario analysis). ACER is defined by the average of cost per effect, while ICER is the ratio of the change in cost to the change in effect [24]. We also displayed the main conclusion of each study.

Study evaluation
Five reviewers (AVI, ADI, QC, KCP, ACC) assessed the risk of bias for each included study using the recommended approach, the Consensus Health Economic Criteria (CHEC) extended checklist [25][26][27]. This 20-item checklist can appraise model-based or trial-based studies, with positive responses scored 1 and negative responses scored 0 [21]. The total score for each item was summed and converted to a percentage with the final scores ranging from zero to 100 (final score = [total score/20] × 100%). The total CHEC score for each study was categorized into four grades based on cut-off values: low (≤50), moderate , good (76-95), and excellent (>95). Higher scores denote higher quality. We also presented the results in a graph using the Review Manager 5.4 software. The list of the 20 individual items assessed is presented in Table 1. Table 1. List of questions assessed in CHEC-extended checklist [25] No Questions 1 Is the study population clearly described? 2 Are competing alternatives clearly described? 3 Is a well-defined research question posed in answerable form? 4 Is the economic study design appropriate to the stated objective? 5 Are the structural assumptions and the validation methods of the model properly reported? 6 Is the chosen time horizon appropriate in order to include relevant costs and consequences? 7 Is the actual perspective chosen appropriate? 8 Are all important and relevant costs for each alternative identified? 9 Are all costs measured appropriately in physical units? 10 Are costs valued appropriately? 11 Are all important and relevant outcomes for each alternative identified? 12 Are all outcomes measured appropriately? 13 Are outcomes valued appropriately? 14 Is an appropriate incremental analysis of costs and outcomes of alternatives performed? 15 Are all future costs and outcomes discounted appropriately? 16 Are all important variables, whose values are uncertain, appropriately subjected to sensitivity analysis? 17 Do the conclusions follow from the data reported? 18 Does the study discuss the generalizability of the results to other settings and patient/client groups? 19 Does the article/report indicate that there is no potential conflict of interest of study researcher(s) and funder(s)? 20 Are ethical and distributional issues discussed appropriately?

Search results
The stepwise selections of articles according to our selection criteria are presented in Figure 1.   [28][29][30][31][32]34,35,37,47], two studies from Africa [43,48], one study from both Africa and Asia (undetermined which country) [42], five studies from South America [39][40][41]44,45], two studies from North America [36,46], one study from both South and North America (Mexico and Costa Rica) [38], and one study from Europe (Turkey) [33].   Reviewed studies were based on either mathematical models (n = 18) or empirical data (n = 3). Most studies (67%) combined costs and effects in cost-effectiveness analyses (CEA), while the other studies were cost-utility analyses (CUA). Table 2 presents the intervention evaluated by each study. A total of thirteen studies were evaluating mammography screening alone, while eight studies evaluated mammography screening in combination with some other breast cancer control. From those eight studies, five studies combined mammography screening with treatment strategies for breast cancer patients stages I-IV; two studies combined mammography screening with risk factor assessment and ultrasound or clinical breast examination (CBE). Table 3 presents further study characteristics. Most studies used a lifetime horizon or something approximating lifetime (n = 18); other studies used 10 years (n = 1), one year (n = 1) or were unclear about the time horizon used (n = 1). Furthermore, only four studies were conducted using a societal perspective. In contrast, the other studies used a health care system perspective (n = 12), or a third-party payer perspective (n = 1). The perspectives in the remaining four studies were not clearly stated. In addition, most studies applied discounting for both costs and effects (n = 14), whereas two studies applied discounting to the costs only, three studies did not apply discounting at all, two studies provided unclear information, none applied differential discounting. Table 3 shows that most of the included studies applied one-way sensitivity analysis, with several of them also applying probabilistic sensitivity analysis or scenario analysis. However, only three studies explicitly validated the model they used for their analysis. Table 4 presents detailed results of ICER and ACER for each study.    pre-requisites to achieving cost-effectiveness. More specifically, Nguyen et al. [31] stated that mammography screening would be cost-effective if the program was conducted in women aged 50-54 years or 55-59 years.

Study findings
On the other hand, the program was not cost-effective for women aged 45-49 years or 60-64 years. Haghighat et al. [35] stated that the first round of mammography screening is cost-effective while the second and third rounds are not, due to lower detection rates.
Among eight studies which evaluated mammography screening in combination with other strategies for breast cancer control, seven studies concluded that mammography screening is cost-effective. However, two of these seven studies concluded that mammography alone is not a cost-effective strategy. Furthermore, one study considered mammography screening in combination with other strategies for breast cancer control as not cost-effective.
To optimally assess cost-effectiveness, screening frequency and population age need to be considered. Studies included in this review evaluated multiple age groups, but mostly (80%) from the age between 40 and 70 years. They compared a range of screening frequencies (from annually to triennially). The studies by Ginsberg et al.  [39] revealed that biennial screen-filmed mammography is more economically viable than annual screening (both strategies were cost-effective). Wang et al. [28] found that biennial mammography (45-70 years) is more cost-effective than triennial mammography (45-70 years). Ulloa-Perez et al. [36] reported that mammography screening approaches cost-saving only if the screening interval is biennial or triennial (40-70 years).
In terms of the type of mammography device, two studies made a comparison between conventional (screenfilmed) mammography and digital mammography [39,44]. Both studies concluded that conventional mammography is more cost-effective compared to digital mammography. This was mainly due to the higher cost of digital mammography. However, Souza et al. [39] also revealed that digital mammography could be cost-effective if implemented in the <50 years old (the age group where digital mammography has the highest benefit) in combination with conventional mammography in the 50-69 years age group. ICER -incremental cost-effectiveness ratio, ACER -average cost-effectiveness ratio, CBE -clinical breast examination, MMG -mammography, US -ultrasonography, BSMP -Bahçeşehir Mammography Screening Project, TNBCRP -Turkish National Breast Cancer Registry Program, GDP -gross domestic product *The results presented by the author in the narrative and table were switched. However, reading through the paper, we finally use the number being given in the narrative as it was making more sense. Coverage rate was also evaluated by studies included in this review. Wang et al. [28] evaluated coverage of 100%, 80%, or 60%; Ginsberg et al. [42] assessed coverage of 50%, 80%, or 90%; Gonzales Marino et al. [45] used coverage of larger than 50%; Valencia-Mendoza et al. [46] evaluated coverage of 25% and 50%. The results were varied as presented in Table 4. Figure 2 and Table 5 summarize the results of risk of bias assessment of included studies, as indicated by the percentage score. Each item was scored 1 (green light) if the study met the requirement or 0 (red light) if it did not meet or only partly met the requirements. Moreover, we noted "not applicable" (N/A) if the item assessed was not relevant for that particular study and it was scored as 1. The quality of all studies ranged from 40% to 95%. 11 studies were categorized as good quality [28,30,33,38,[39][40][41][42][43]47,48], seven as moderate quality [29,31,32,34,35,46], and three as low quality [36,44,45].

Risk of bias assessment
Studies generally (70%) scored 0 on item number 5 as they mainly lacked stating assumptions made in their model or in model validation. Furthermore, most studies (75%) scored 0 on item number 20 since they did not sufficiently discuss ethical and distributional issues. Half of the studies scored 0 on items number 7, number 8, and number 18. Those studies did not justify not using a societal perspective, not accounting for all important and relevant cost (eg, follow-up or palliative cost), or not discussing the generalizability of their results to other settings and patient/client groups.  Table 4 shows the conclusions made by each study and the study quality. Among 18 studies which concluded that mammography screening is cost-effective, with or without combination with other strategies, study quality varied from low to good. In detail, nine studies have good quality, seven studies have moderate quality, and two studies have low quality. On the other hand, from three studies that concluded that mammography screening is not cost-effective, two have good quality and one has moderate quality.

DISCUSSION
This review identified 21 studies on the economic evaluation of mammography as a breast cancer screening method in LMICs. 19 studies were from MICs (Upper or Lower) and two studies from both MICs and LICs. By conducting this systematic review in 2021, we found 17 additional studies examining the cost-effectiveness of mammography screening than the 2013 systematic review by Zelle et al. [49] about economic evaluation of breast cancer control in LMICs. Our systematic review found that mammography screening appeared to be a cost-effective strategy, particularly in Upper MICs.

Study findings and implications
18 out of 21 studies included in our review concluded that mammography screening was a cost-effective strategy, some of which did not assess mammography as a single screening strategy compared to no screening. Instead, 12 studies compared mammography screening only to no strategy or other strategy; two studies compared the combination of mammography, risk-based assessment, and ultrasound/CBE to no strategy or other strategy; four studies compared mammography screening plus treatment of stage I-IV to other strategy.
The majority of those 18 studies concluding that mammography screening was a cost-effective strategy was conducted in Upper MICs (71%). This conclusion appears to be in line with the WHO guideline, which states that an organized population-based mammography screening is recommended for implementation in well-resourced settings or limited resource settings with relatively strong health systems [1].
The majority (76%) of the included studies in our review specified the screening interval. For most of those studies the screening interval was two years, which was in general found to be cost-effective. When a comparison was made between biennial screening and annual or triennial screening, a biennial interval was found to be more cost-effective than the other screening interval strategies. This is in line with the conclusion of the review of systematic reviews by Mandrik et al. [18] as well as the WHO recommendation of a screening interval of two years among women aged 50-69 years in well-resourced settings or limited resource settings with relatively strong health systems [1]. This recommendation was based on modelling studies and further analysis of trials showed that screening every two years seems to provide the best trade-off between benefits (mortality reduction) and harms (overdiagnosis or overtreatment).
It is also essential to discuss the issue of the appropriate age group for screening. In several HICs (eg, United States, Sweden, Japan) mammography screening begins at the age of 40 years [50,51]. However, there is clear uncertainty about the magnitude of overdiagnosis among both younger and older women. Mammography screening at age 40 years reduced mortality more than at age 50, but it consumed more resources and resulted in more false-positives [1,49]. This also occurs in screening among women over 69 years of age that will generate some mortality reduction, but will also substantially increase overdiagnosis [1].
Generally, the studies included in our review examined mammography screening among women aged 40-70 years. Five studies explored mammography in the range of 50-70 years old and concluded that mammography screening is cost-effective. In contrast, 15 studies assessed mammography screening at a younger starting age (ie, 25, 35, 40, 45 years old). These studies came up with inconclusive results on its cost-effectiveness. Although the target age group is a major issue in organizing a breast cancer screening program, each country has its specific incidence and unique screening design which cannot be easily adopted from different settings [52].
Two of the studies included in our review also examined women by a risk-based approach. The screening strategy evaluated by Ribeiro et al. [41] combined mammography screening, CBE, and risk factor assessment. Sun et al. [30] used 'Your Disease Risk' which calculates individual cancer risk, after which high-risk women underwent mammography and ultrasound screening. The economic analysis by Ribeiro et al [41] and Sun et al. [30] revealed that screening with a risk-based approach is cost-effective. Some researchers argued that riskbased screening might not be optimal, considering that 80% of women with newly diagnosed breast cancers have no known major risk factors [50,53]. However, compared to one-size-fits-all screening strategy that may induce unnecessary inclusion of a large population with lower risk, screening individuals tailored to higher risk of developing breast cancer could be a more economically viable alternative [52].
The most recent study in China (Upper MIC) revealed that biennial mammography screening compared to no-screening cost US$25 261 per life years gained, which was considered cost-effective [28]. Furthermore, the budget impact analysis showed that screening would result in a net cost of US$38.1 million for a city with one million citizens for ten years. On the other hand, a study from sub-Saharan African (Lower MIC) and Southeast Asia countries (LIC) showed that biennial mammography screening in combination with treatment of all cancer stages cost Int$2000-6000 per DALY averted and can also be considered cost-effective [48]. To our knowledge, budget impact analysis studies on mammography screening in these regions are still unavailable. Thus, it is difficult to estimate the affordability of mammography screening program in Lower MIC and LIC.

Quality of evidence
We found 18 studies that concluded mammography screening is cost-effective, with or without combination with other strategies, with studies' quality varying from low to good. Three studies that concluded that mammography screening is not cost-effective were of good and moderate quality. This variation complicates the grading of the strength of evidence and advocating a recommendation. However, if we ignore the item in CHEC which would not influence the results of ICER (ie, discussion on ethics, distributional, and generalizability, conflict of interest), the majority of studies (80%) that concluded mammography screening is cost-effective would be rated as good, while the quality rating of studies which conclude that mammography screening is not cost-effective remain the same.
Nearly all studies with good quality were model-based. It is generally advised to the use of modelling studies in economic analysis. Modelling approaches enable the researcher to be flexible in the inclusion of important methodological characteristics such as adequate time horizon and more appropriate to evaluate a broad array of interventions across different groups [49]. However, most studies (70%) got a zero score concerning model validation in their research. Model validation is important in showing if the model sufficiently represents the system under assessment. The lack of model validation may have influenced the ICER results and therefore affect the extent to which results can serve as a solid basis for decision making [54].

Strengths and limitations
One strength of our review was that it captured a substantial amount of full economic evaluation studies (CEA or CUA) in LMICs, as we used a comprehensive search strategy and included studies without any language or year's limitation. This review included 17 additional studies conducted in LMICs compared to another published systematic review on breast cancer control [49]. However, most studies were conducted in Upper MICs, and only few studies were from Lower MICs or LICs.
We did not assess publication bias in this review. This may lead to an overestimation on the cost-effectiveness of mammography. To minimize the publication bias, we also searched at Embase database, which covered grey literature such as conference proceedings [55].
Furthermore, there are some concerns raised by Welch et al. [56] worth mentioning here although they may not apply in the same way to LMICs. First, mammography screening may cause overdiagnosis (mostly of small tumours) and therefore overestimation of the benefit of screening. Second, reduction in fatality rates over time may be a result of improved treatment options. Both these issues could lead to overly optimistic assessments of cost-effectiveness. Although a model-based cost-effectiveness analysis would ideally adjust for these things, there may still be confounding in the background, and so results should be interpreted with caution.
We used the CHEC-extended checklist, which can appraise model-based or trial-based or observational-based cost-effectiveness studies. However, this checklist is quite generic to accommodate studies of various methodologies. Some of the CHEC items only apply to model-based analyses and it is complicated to rate the risk of bias of an observational-based study according to these same standards. Therefore risk of bias scores are difficult to compare across studies and should be interpreted with caution.
Drawing an overall conclusion for a systematic review of economic evaluation has always been a difficult task. A quantitative meta-analysis of ICERs, costs, or health benefits is hardly feasible due to heterogeneity across study designs, methods, country policies, health care settings, and many further practical challenges [57]. Therefore, the synthesis of results commonly takes a narrative approach, as we applied in this review. Without pooling the results, narrative synthesis could identify outcome patterns relating to the direction of an effect [58]. In addition, determining the transferability of the findings of this review to other countries remains a challenge.

CONCLUSIONS
To conclude, this systematic review found that mammography screening was a cost-effective strategy in LMICs, particularly in Upper MICs. However, there is still limited evidence in the Lower MICs and LICs and the quality of studies varied widely. Thus, more studies conducted in Lower MICs and LICs are needed to better understand the cost-efectiveness of mammography screening in these regions.