Stated Preference for Cancer Screening: A Systematic Review of the Literature, 1990–2013

Introduction Stated-preference methods provide a systematic approach to quantitatively assess the relative preferences for features of cancer screening tests. We reviewed stated-preference studies for breast, cervical, and colorectal cancer screening to identify the types of attributes included, the use of questions to assess uptake, and whether gaps exist in these areas. The goal of our review is to inform research on the design and promotion of public health programs to increase cancer screening. Methods Using the PubMed and EconLit databases, we identified studies published in English from January 1990 through July 2013 that measured preferences for breast, cervical, and colorectal cancer screening test attributes using conjoint analysis or a discrete-choice experiment. We extracted data on study characteristics and results. We categorized studies by whether attributes evaluated included screening test, health care delivery characteristics, or both. Results Twenty-two studies met the search criteria. Colorectal cancer was the most commonly studied cancer of the 3. Fifteen studies examined only screening test attributes (efficacy, process, test characteristics, and cost). Two studies included only health care delivery attributes (information provided, staff characteristics, waiting time, and distance to facility). Five studies examined both screening test and health care delivery attributes. Overall, cancer screening test attributes had a significant effect on a patient’s selection of a cancer screening test, and health care delivery attributes had mixed effects on choice. Conclusion A growing number of studies examine preferences for cancer screening tests. These studies consistently find that screening test attributes, such as efficacy, process, and cost, are significant determinants of choice. Fewer studies have examined the effect of health care delivery attributes on choice, and the results from these studies are mixed. There is a need for additional studies on the barriers to cancer screening uptake, including health care delivery attributes, and the effect of education materials on preferences.


Introduction
Stated-preference methods provide a systematic approach to quantitatively assess the relative preferences for features of cancer screening tests. We reviewed stated-preference studies for breast, cervical, and colorectal cancer screening to identify the types of attributes included, the use of questions to assess uptake, and whether gaps exist in these areas. The goal of our review is to inform research on the design and promotion of public health programs to increase cancer screening.

Methods
Using the PubMed and EconLit databases, we identified studies published in English from January 1990 through July 2013 that measured preferences for breast, cervical, and colorectal cancer screening test attributes using conjoint analysis or a discretechoice experiment. We extracted data on study characteristics and results. We categorized studies by whether attributes evaluated included screening test, health care delivery characteristics, or both.

Results
Twenty-two studies met the search criteria. Colorectal cancer was the most commonly studied cancer of the 3. Fifteen studies examined only screening test attributes (efficacy, process, test characteristics, and cost). Two studies included only health care delivery attributes (information provided, staff characteristics, waiting time, and distance to facility). Five studies examined both screening test and health care delivery attributes. Overall, cancer screening test attributes had a significant effect on a patient's selection of a cancer screening test, and health care delivery attributes had mixed effects on choice.
Introduction Screening for certain cancers may increase the identification of early-stage disease and likelihood of successful treatment and survival (1). Screening for breast, cervical, and colorectal cancer is recommended by the US Preventive Services Task Force (USP-STF) (2). Recent analysis of the 2013 National Health Interview Survey indicates that the percentages of the population screened for breast, cervical, and colorectal cancer were 72.6%, 80.7%, and 58.2%, respectively (3), below the Healthy People 2020 recommended targets of 81.1%, 93.0%, and 70.5%, respectively (4).
Research that leads to an understanding of how patients value the attributes of health care interventions is critical to the design, development, and implementation of effective programs. Incorporating patient values in the decision-making process may result in operational policies and programs that enhance the effectiveness of health care interventions by improving the uptake of and adherence to recommended preventive health care services (5).
Stated-preference (SP) methods systematically assess the relative preferences for screening tests or the features of screening tests using questions that present hypothetical trade-offs. Furthermore, SP studies can incorporate questions to assess the factors that affect reported likelihood of uptake for cancer screening (5). Previous reviews of SP studies indicate that people have identifiable preferences for the features of cancer screening tests (6)(7)(8).
This article reviews SP studies of preferences for cancer screening tests for breast, cervical, and colorectal cancer recommended by the USPSTF that were collected using conjoint analysis (CA) and discrete-choice experiments (DCEs). CA and DCEs describe tests (or other goods) using a set of attributes (features) with varying levels and allow estimation of relative preferences for different attributes. The goal of the review was to assess the types of cancer screening test attributes researchers have considered, differentiating between attributes of the screening tests themselves and attributes that capture other elements of the patient experience. We also reviewed the use of questions to determine reported likelihood of uptake. Understanding how test attributes affect reported likelihood of uptake may help improve public health programs to increase cancer screening.

Stated-preference techniques
Researchers have developed several approaches consistent with economic theory to measure preferences for market and nonmarket goods, interventions, and policies (5). Revealed-preference methods use information from actual behavior or purchases to infer individuals' preferences; SP methods use surveys or experimental methods with hypothetical scenarios to elicit preferences. There are varied SP methods, including contingent value, timetrade-off, standard gamble, and other variations. The Medical Device Innovation Consortium has more information on SP methods in health care research (9). This review focuses on CA and DCE studies, a type of SP study where the good or policy is defined by a set of attributes with varying levels (for a general discussion, see Hensher et al [10]). These surveys allow researchers to identify and quantify the relative effect of the changes in different attributes on choices. Good practice suggests that the number of attributes should be limited depending on the nature of the attributes and that researchers should make decisions about the attributes to include and exclude (5). Researchers use their research question and findings from previous studies and pretesting to select attributes that respondents find relevant. To examine reported likelihood of uptake and attributes that influence reported uptake, researchers can include a fixed alternative in the choice question, usually a reference test representing the standard of care or the option of not getting a test, or a follow-up question asking if the respondent would get the hypothetical test they selected in the choice question. CA and DCE approaches have been used for decades in the fields of marketing, transportation, environmental policy, and health care.

Data sources and literature review strategies
Studies eligible for this systematic review met the following criteria: was a CA or DCE study; examined patient preferences for breast, colorectal, or cervical cancer screening recommended by the USPSTF; had the full-text article available in English; and was published from January 1990 through July 2013. We excluded studies that examined cancer treatment, cancer therapy, pharmaceuticals, healthy behaviors, or cancer prevention strategies not recommended by the USPSTF. We also excluded studies that included only physicians in their sample (Table 1). We abstracted the following data items from the selected studies: author(s), year, sample size and population, cancer type, purpose of the study, attributes studied, significant attributes (defined as categorical attributes in which at least 2 levels were significantly different from each other or a continuous attribute with a significant coefficient [P ≤ .10]), whether the design included a no-test option, and predicted uptake as reported in the articles.
The review focused on the types of attributes included in the studies. To provide more focus for the review, the studies were categorized as studies that focused on screening test attributes only, health care delivery attributes only, or a combination of both. The categories were defined as follows: •Screening test attributes: attributes of the tests independent of the patient's characteristics. These included efficacy (sensitivity, expected reduction in cancer rates or mortality, specificity), test features (type of test, preparation before the test, length of test, pain during test, complication risk), recommended frequency, where the test was administered, how soon results were available, whether a follow-up test was needed to address abnormal findings, and cost.
•Health care delivery attributes: attributes related to the patient experience in the health care setting in which the screening was offered that are unrelated to the attributes of the test. These included attributes such as information provided to patients, how information was delivered, characteristics of the doctor and health care staff, waiting time for appointments, and distance to facility.
Studies were qualitatively assessed to identify common results.

Results
Of the 22 studies, 15 included only screening test attributes, 2 included only health care delivery attributes, and 5 were a mixture of the 2. Tables 2 and 3 summarize the study characteristics and results.
DCE and CA studies can be set up as a forced choice, where respondents pick between tests, or they can include a no-test option where the respondent can select "no test" instead of the hypothetical options posed in the choice question. Two-thirds of the studies included a no-test option. In addition, 1 study included a separate question asking about preferences for specific unlabeled tests assigned with the characteristics of existing tests and included the option of no test (23). Four studies provided predictions of uptake for tests with specific characteristics. Gyrd-Hansen (15) found that predicted uptake for a hypothetical test screening people aged 50 to 69 years every second year with features drawn from the literature and a program in Denmark (80%-88%) was similar to estimates of actual uptake (88%). Hol et al (18) (25) estimated uptake using risk reductions taken from the clinical literature to be 75% for biennial FOBT screening, 80% for 5-yearly flexible sigmoidoscopy screening, and 71% for 10yearly colonoscopy screening for this sample from the Netherlands.
Another feature that distinguished the studies was whether the screening test was identified by the process or name of the procedure. This feature was most relevant for colorectal cancer screening, in which available tests range from stool samples to colonoscopies. De Bekker-Grob et al (14) compared an unlabeled design with a labeled design. Howard et al (20) used a labeled design. Four studies included an attribute that identified the type of colorectal cancer screening test by name or through the process (17,18,21,22). The rest of the studies described the tests through attributes related to efficacy and process without mentioning the type of test.
All studies included some kind of efficacy attribute. Forty percent defined efficacy as the accuracy of the test (the probability that the test found cancer or precancerous growths); the rest presented the reduction in risk of cancer mortality. The efficacy attributes were significant in every study. Forty-seven percent of the studies also included specificity (the risk of false negatives) as an attribute, which was significant in every study except one (16).
Test experience attributes included preparation before the test, discomfort during the test, waiting time for results, whether a followup visit was needed if results were abnormal, complication risk, duration of screening procedure, recommended test frequency, out-of-pocket cost, and type of facility where the test was conducted. The attributes that were always significant were preparation before the test (included in 47% of the studies), discomfort or pain during the test (included in 40% of the studies), waiting time for the results (included in 27% of the studies), complication risk (included in 27% of the studies), cost (included in 67% of the studies), and the type of facility where the test was preformed (included in 13% of the studies). Waiting time to get test results was not significant in 1 of the 4 times it was included (12), location of test in 1 of 2 times (14), test frequency in 2 of 11 times (15,16), and whether a follow-up test was needed to confirm abnormal results in 1 of 4 times (22).
The primary purpose of most studies was to examine preferences for screening test features; however, 3 of the studies investigated questions about DCE or CA methods. De Bekker-Grob et al (14) looked at the effect of a labeled versus unlabeled design. Pignone et al (23) compared choice-based CAs with rating and ranking.
Howard and Salkeld (19) examined the effect of attribute framing (whether sensitivity and specificity were presented as cancers found or cancers missed).

Studies with only health care delivery attributes
Only 2 studies, which looked at preferences for genetic counseling, included exclusively what we termed health care delivery attributes (27,28). Griffith et al (27) looked at preferences for genetic testing among women with a low, moderate, or high risk of breast cancer. Peacock et al (28) examined preferences for the type of information received during counseling for women at high risk of carrying the BRCA1 or BRCA2 genetic mutations, which are associated with a higher risk for breast and ovarian cancer.
The attributes in Griffith et al (27) were related to the appointment and were all significant, except whether the screening test was available only for high-risk women (versus the entire population), which was not significant to high-risk women, and the length of the appointment, which was not significant to low-risk women. The attributes in Peacock et al (28) included 4 topics that could be discussed during counseling; all were significant. Four studies included sensitivity of the screening test, reduction in cancer mortality, or both, and 4 included the chance of a false-negative (specificity). These attributes were significant in all the studies, except specificity, defined as rate of unnecessary colonoscopy in Nayaradou et al (31). Cost was included in 3 of the studies and was consistently significant (29,31,32).

Discussion
Overall, the studies suggest that respondents valued improvements in attributes related to the characteristics of cancer screening tests, including sensitivity, process, and cost. The significance of the health care delivery attributes was uneven across studies, especially in studies combining test and health care delivery attributes. More than half of the studies included only screening test attributes. Thirteen included some type of opt-out option, but only 4 calculated predicted uptake for specific tests. With a larger sample of 22 studies, we confirmed the findings in the earlier reviewsthat patients had preferences over multiple attributes and that sensitivity was an important feature. This review included articles published through July 2013. Since this review was completed, several additional CA studies, not included in this review, have been published, including 8 more on colorectal cancer screening and 1 on breast cancer screening (34-42). Three of these more recent studies included health care delivery attributes such as travel time to breast screening appointment and the sex of staff members conducting breast screening (35,39,41). As with the 2 previous re-views (6,7), we found that most of the studies were administered to the general population at average risk of cancer; however, there are now more studies of populations at high risk of cancer or with screening histories. Several of the new studies focused on specific populations including older adults and Hispanics (34,35,39), and 1 study was conducted in Japan (41).
There are many ways in which these results from SP studies can aid in the design of future research and be applied to public health programs designed to increase screening. For example, in the United States, physicians may be more likely to recommend colonoscopy than other tests (43,44); however, the DCE and CA studies suggest that preparation, discomfort, and cost are important to patients and that some patients may prefer a stool test. In countries where stool tests are the standard of care, offering colonoscopies could improve uptake among people who have strong preferences for high sensitivity.
Health care delivery variables were sometimes nonsignificant. In SP surveys, process variables such as waiting time for an appointment may be nonsignificant relative to variables such as sensitivity, but these process factors may be important in determining whether people get screened. If an acceptable test exists, then process factors related to making appointments, getting the test, and getting the results may have a big influence on uptake for that test. Our understanding of preferences and uptake could be improved by additional research on the best way to include attributes associated with health care delivery. Health behavior theory, which has been used to develop and evaluate public health interventions (45), could provide a useful structure to develop attributes or other supporting questions related to attitude, environmental, or social factors influencing uptake (see Tsunematsu et al [41] for an example).
The hypothetical nature of SP surveys makes it challenging to accurately predict uptake. Nonetheless, adding a no-test option and providing estimates of uptake for specific tests when appropriate will provide more information on preferences and predicted uptake.
The issue of labeled versus unlabeled designs can affect predictions of uptake. De Bekker-Grob et al (14) found that choices differed based on whether labels were included. They concluded that respondents were less attentive to the attributes when labels were provided and that labeled designs may be more appropriate for respondents who were familiar with the labels and for studies interested in predicting uptake. It is unknown whether including test names as attributes is similar to using a labeled design. We focused on patient preferences; however, studies have been done with physicians or comparing patients and physicians (12,22,29,46). Studies on physician preferences are important, because patients often rely on their physicians for advice (7). If patients and physicians value attributes differently, patient-preference surveys provide an opportunity for physicians and patients to identify differences in perspective, which could improve communication and shared decision making.

PREVENTING CHRONIC DISEASE
CA and DCE surveys could also be used more extensively to test the effect of messages on preferences and willingness of different populations, including underserved populations, to be screened. The results could help shape strategies for public health communication, especially because studies have found that the type of information provided can affect preferences for screening tests (7,38).
Our review has limitations. We reported attribute significance; however, the significance or lack of significance of attributes should be viewed as conditional on the set of attributes included and the range of levels. An attribute may be more or less important depending on the other attributes included in the survey. In general, best practice suggests that researchers include attributes that are important to respondents, implying that most attributes should be significant. However, even with careful pretesting, changes in attributes that are important in isolation may not be important when included in a wider set of attributes. The surveys differed in objectives and format, limiting our ability to compare findings across studies. Furthermore, few studies were conducted in the same country, which limits the generalizability of findings, because differences in national health policies vary widely among countries. For example, although many studies focused on colorectal cancer screening, only 3 were conducted in the United States.
A growing number of studies examine preferences for cancer screening tests. These studies consistently find that screening test attributes such as efficacy, process, and cost are significant determinants of choice. Fewer studies have examined the effect of health care delivery attributes on choice, and the results from these studies are mixed. Going forward, there is a need for studies on the barriers to cancer screening uptake, the impact of education materials on preferences, and the role of preference studies in patient and physician communication. Patient-preference studies may become more important as patient-centered care gains more prominence.