Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Evidence‐based consumer choice: a case study in colorectal cancer screening

Evidence‐based consumer choice: a case study in colorectal cancer screening Glenn Salkeld Screening Test Evaluation Program, School of Public Health and Surgical Outcomes Research Centre, The University of Sydney, New South Wales Abstract Objectives: To elicit community preferences for colorectal cancer (CRC) screening by faecal occult blood test (FOBT) using discrete choice modeling (DCM). To provide policymakers with information that would assist them in designing the future national screening program. Michael Solomon Department of Colorectal Surgery and Surgical Outcomes Research Centre, The University of Sydney, New South Wales Leonie Short Centre for Public Health, Queensland University of Technology Methods: 301 participants in central Sydney, aged 50 to 70 years , at ‘average’ risk of CRC, participated in a face-to-face discrete choice study interview in which screening profiles were posed to derive estimates for preferences for CRC FOBT screening. Results: Three character istics were varied in our screening profiles, namely: benefit (CRC deaths prevented); potential harm (false positive induced colonoscopy); and notification policy (of test result). Ninety-four respondents (32%) did not trade off CRC deaths prevented for any reduction in harms. Twelve per cent alw ays chose no screening. The remaining 56% traded benefits and harms. These latter respondents (n=164) were willing to accept 853 (false positive induced) colonoscopies for one CRC death prevented. Mandy Ryan Health Economics Research Unit, University of Aberdeen, Scotland Jeanette E. Ward Division of Population Health, South West Sydney Area Health Service, New South Wales powerful shift is emerging in the way health care is transacted. Research suggests that consumers want more infor mation about their health care options and, for some, a more active role in making decisions about their treatment.1 The movement towards e vidence-based information and the centrality of individual patient choice and values in decision making has been described as a shift towards ‘evidence-based consumer choice’.1 At the same time, there is growing interest in the use of community preferences to inform health services planning and delivery. 2 Together, evidence-based consumer choice and recognition of community preferences in health planning pose new challenges for population health such as screening for colorectal cancer (CRC). Elsewhere, the General Medical Council in the United Kingdom (UK) has guidelines about information that should be given to people considering screening tests.3 These guidelines state that information should be provided about the purpose of screening, its uncertainties such as the risks of a f alse positive or false negative result and any significant medical, social and f inancial implications of screening for individuals and their relatives.3 To our knowledge, there are no similar guidelines in Australia. Yet recent Australian research has shown that women want to be well informed about the possibility of f alse test results when participating in mammography screening and the risks of any adv erse effects of tests.4 Women repor t receiving information on the benefits of tests and treatments signif icantly more often than they recei ved information on adv erse effects. 4 Consumer preferences ought to influence the deliver y of CRC screening. There are a number of ways to elicit consumer preferences, either revealed preferences (based on actual choices) or stated preference (based on hypothetical choices). We used stated preference to assist policymakers in the design of a future national screening program. Stated preference methods include the concept of QALYs (quality adjusted life years), Conclusions: While survival was all that mattered for just over one-third of the sample and 12% would choose no screening, the remaining individuals were prepared to trade CRC deaths prevented against other character istics. CRC screening will not receive unqualified community support, irrespective of har ms. Implications: In any future national CRC screening program, consideration of these insights about community assessment of benefits, harms, costs and other characteristics of CRC screening is warranted. ( Aust N Z J Public Health 2003; 27: 449-55) Correspondence to: Dr Glenn Salkeld, School of Public Health, University of Sydney, A27, NSW 2006. Fax: (02) 9351 7420; e-mail: glenns@health.usyd.edu.au Submitted: December 2002 Revision requested: April 2003 Accepted: June 2003 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article discrete choice modelling (DCM) or contingent valuation, based on ‘willingness to pay’ (WTP). QAL were not pursued in our Ys study because we wanted to value ‘process’ attributes as well as ‘outcomes’ attributes. We chose DCM over WTP because the latter does not provide a quantitative estimate of the relative importance of harms, benefits and information to potential screenees.2 DCM involves asking respondents to choose between different options for treatment or services based on the salient characteristics (or attributes) of that service. It has been advocated in the UK to measure patient preferences for clinical ser vices Locally, Jan et al. used DCM to elicit community preferences for hospital services in South Australia. Specifically, they asked respondents to indicate their preference for two types of hospital services based on different levels of attributes such as travel time to hospital, waiting times for elective surgery or emergency ser vices, cost and complication rates.5 By analysing respondents’ choice between two options, the authors demonstrated that the community chooses on the basis of complication rates and length of waiting times for elective surgery and in the emergency depar tment.5 In public health, DCM has been applied to primary health care services,6 varicella vaccination, 7 genetic screening,8 breast cancer screening9 and the Rotary Bowelscan (a community-based CRC testing program).10 The aim of this present study was to assess community decisions about CRC screening based on the trial evidence on harms and benef its. The community in this study was the eligible CRC screening population, men and women aged 50-70 years, living in central Sydney. We also aimed to provide policy makers with infor mation that would assist them in designing the future national screening program.11 ‘strongly agreed’ would encourage their par ticipation. The DCM approach assumes that individuals consider the information and will trade between the attributes. As this is less likely to happen as the number of attributes increases, we selected three attributes as follows: benef it (CRC deaths prevented); potential harms (the chance of a false positive test result and requiring an unnecessar y colonoscopy); and notif ication policy (specifically, either notification of any test result irrespective of its f indings or notification only of a positi ve test result). Stage 2: Allocating levels to attributes of CRC screening Next, we assigned various plausible levels to each of the three attributes (see Table 1). Levels for CRC deaths prevented and colonoscopies due to a false positive test result were based on a systematic review of trial data for CRC screening.13-15 The range of benefits represented the upper and lower 95% confidence intervals for CRC deaths prevented. Similarly the range of harms represented the upper and lower 95% conf idence intervals for false positive induced colonoscopies. As stage 1 had shown the term ‘f alse positive induced colonoscopies’ was not widely understood, we expressed this as ‘unnecessary colonoscopies’ in stage 3, explaining the ter m at the beginning of our interview. For notification policy, there were only two possible levels (namely being notified whether positive or negative or being notified only when positive). Stage 3: Experimental design Thus, our DCM consisted of two attributes each with four levels and one attribute with two levels, generating 32 possible screening profiles (combinations of benef it, potential harm and notification policy)(see Table 1). A study design was used whereby 16 pairwise choices were generated from these 32 possibilities.16 As recommended, 2 profiles were combined to ensure minimum correlation between each of the pairwise choices. As far as possible, levels for each attribute occur red an equal number of times.2 In addition to the 16 pairwise choices, an additional two choices were included to test for rationality. In both, program A was unambiguously better than program B (that is, it had more CRC Table 1: Characteristics and le vels used for a discrete choice study of CRC screening. Characteristic Number of CRC deaths prevented (BENEFIT) Methods We followed DCM principles and protocols as follows.2 Stage 1: Determining attributes Through informal networks of community groups, we first recr uited 54 men and women aged 50 to 70 years to participate in one of eight focus groups. To achieve a breadth of views, these focus groups consisted of two g roups of women aged 50-59 years, one of men aged 50-59 years, three of women aged 60-69 years and two groups of men aged 60-69 years. Consensus emerged from the focus groups as to the characteristics that might influence people to participate in CRC screening: namely, the organisation of and recr uitment for screening, test accuracy, performing the test, survival gains, policy regarding notif ication of test results, delivery of the test kit and the cost of the test. Following this, and as part of this representative DCM study, 791 men and women aged 50-70 years and living in central Sydney completed self-administered questionnaires. Recruitment is reported in detail elsewhere.12 Respondents were asked to rate the importance of those characteristics previously identified in the focus groups according to whether it would encourage them to participate in CRC screening.12 From these responses, we identified f ive attributes of CRC screening that at least 85% of respondents either ‘agreed’ or Levels • 3 per 10,000 screened • 8 per 10,000 screened • 14 per 10,000 screened • 20 per 10,000 screened • 100 per CRC death prevented • 300 per CRC death prevented • 600 per CRC death prevented • 800 per CRC death prevented Number of unnecessary colonoscopies (HARM) Negative test result (NOTIFICA TION) • Notified of either a positive or negative test result • Notified only of a positive test result and not notified of a negative test result AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH VOL. 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings deaths prevented and fewer colonoscopies than the alternative). Notification polic y was identical in each. Failing one rationality test may be due to random er ror.2 Failing both rationality tests suggests irrational decision making.2 Therefore individuals who failed both tests of rationality were dropped from the analysis. Stage 4: Data analysis Based on the assumption that respondents would choose the option that provided them with the highest level of satisfaction (or utility), the following benefit equation (model 1) was estimated for consistent respondents who chose a screening option: ∆B = ß1BENEFIT + ß2HARM + ß3NOTIFICATION where ∆B is the change in benef it (utility) in moving from screening program B to program A and the explanatory variables Figure 1: Outline of interview schedule. Respondents first were given an outline of the purpose of the interview. Inter viewees were then provided with the following information: Background • A descr iption of what happens with respect to the diagnosis of CRC in the absence of screening. • A descr iption of CRC screening, what it involves and an example of a FOB test kit. • A statement on the oppor tunity cost of a national CRC screening prog ram, specifically ‘A screening program like this would be expensive. It would cost the Commonwealth Government tens of millions of dollars each year. That is money that has to come from somewhere – either from increased taxation or reductions in other health programs or both.’ CRC screening • Asked to imagine that they are invited by their local doctor (GP) to ha ve this test free of charge. • A descr iption of two identical CRC biennial screening programs for a population of 10,000 men and women aged between 50-69 years. • A statement that the programs w ere identical except for the three char acteristics for which standardised descriptions were provided: benefits (number of CRC deaths prevented); harms (number of colonoscopies due to a false positive FOB test result); and notification policy (whether or not they would be notified of a negative test result). • A descr iption on what is involved in having a colonoscopy and a statement on the rate of complications for that scenar io ar ising from colonoscopies performed on the screened population. ‘Warm up’ choice question Could y ou please compare the two programs and tell me whether you would prefer Program A, Program B or whether you would prefer not to have the screening test. EXAMPLE ONLY: Per 10,000 men and women screened over a 10-year period. Example scenario Program A Program B Number of CRC deaths prevented 20 3 Number of unnecessary colonoscopies 2,000 2,400 Notification of negative test result Yes Yes Which w ould you prefer? Program A, Program B or no screening? Stage 4: Collecting the data Interview schedule Our face-to-face interview commenced with a detailed explanation of what happens in the absence of screening, the nature and purpose of CRC screening, the benef its and potential har ms, a description of colonoscopy as a diagnostic test, the possible complications of colonoscopy and, f inally, the nature and purpose of the DCM questionnaire (see Figure 1). Respondents were told that screening would be organised through their GP and would be free of charge. Hence, these two organisational characteristics, cost and screening by GPs, were fixed in the DCM study. Respondents then were taken through a warm-up exercise. In the ensuing 18 pair-wise choices, respondents were asked to nominate their preference. For e xample, Figure 2 shows how respondents were asked to choose between the additional six CRC deaths prevented in program B against 6,000 fewer false positive-induced colonoscopies in program A. Potential har ms and benefits were presented as absolute frequencies to help individuals make more accurate judgements.17 To help us interpret individuals’ choices we included in the interview schedule two closed questions in which respondents were asked whether they made all of their choices based on one attribute only and, if so, whether that attribute was benefits, harms or notification policy. Respondents were also asked whether or not they developed a decision-heuristic (a rule of thumb ) to help them cope with the potentially difficult task of trading between two risk-based attributes from a large set of 18 discrete choice questions. Further details on these questions are presented in Analysis. At the end of the interview, we asked respondents to rate their level of agreement with the statement ‘I am satisfied with my decisions overall’ on a f ive-point Liker t scale (‘strongly agree’ to ‘strongly disagree’). We also asked them to rate how ‘easy’ or ‘difficult’ they found making the choices, using another five-point Likert scale (‘v ery easy’ to ‘ver y difficult’). Subjects Respondents for the DCM consisted of 414 respondents to the earlier rating exercise who had consented to par ticipate in future Figure 2 Could you please compare the two programs and tell me whether y ou would prefer Program A, Program B or whether you would prefer not to have the screening test. Per 10,000 men and women screened over a 10-year per iod Example scenario Program A Pr ogram B Number of CRC deaths prevented 8 14 Number of unnecessary colonoscopies 2,400 8,400 Notification of negative test result Yes Yes Which w ould you prefer? Program A, Program B or no screening? stages of the study. Of these, five w ere excluded because the y had been previously diagnosed with CRC and, therefore, would be ineligible for screening. Thus, 409 (209 male and 200 female) participants were eligible for interview. Of these, 301 (138 men and 163 women) completed it and 108 did not consent to be interviewed. Elsewhere,13 we had collected socio-demographic information, including two terms to assess whether or not the respondent had a family member with CRC and/or knew someone with cancer. 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article Table 2: Socio-demographic and CRC risk characteristics of the 301 respondents, by gender. Men n=138 n % Age group 50-60 61-70 Highest level of education Didn’t complete secondary Completed secondary T rade or technical qual. University or college Employment status Employed Not employed Marital status Married Not married Household income <$20,000 pa $20,000 to $35,000 pa $35,001 to $50,000 pa >$50,000 pa Country of birth Australia Outside Australia Health status (self report) Excellent Ver y good Good Fair Poor Perceived individual risk of CRC (compared with someone of own A lot less Somewhat less The same Somewhat more A lot more 90 (65) 48 (35) 23 (18) 29 (21) 35 (26) 48 (35) 99 (73) 37 (27) 106 (78) 30 (22) 12 (9) 16 (12) 42 (32) 60 (46) 98 (71) 40 (29) 28 (20) 45 (33) 46 (33) 17 (12) 2 (1) age) 14 (10) 20 (15) 73 (55) 22 (16) 5 (4) Women n=163 n % 78 (48) 85 (52) 24 (15) 70 (44) 55 (12) 47 (29) 72 (45) 88 (55) 105 (65) 157 (35) 30 (21) 34 (23) 32 (22) 49 (34) 119 (73) 44 (27) 34 (21) 49 (30) 58 (36) 18 (11) 4 (2) Total n=301 n % 168 (56) 133 (44) 47 (16) 99 (34) 54 (18) 95 (32) 171 (57) 125 (43) 211 (71) 87 (29) 42 (15) 50 (18) 74 (27) 109 (40) 217 (72) 84 (28) 62 (21) 94 (31) 104 (35) 35 (12) 6 (2) are the differences in the attributes of the two screening programs, A and B, as shown in Table 1. There is no constant term in the equation because respondents were told that all aspects of the two screening programs were identical other than those specif ied in the questionnaire. A probit model was used to analyse the discrete choice data. In such a model, there is potential for correlation between observations from each respondent and errors in the model not being independent.18 Hence, a random effects probit model was used to test for the presence of intra-observer correlations using LIMDEP econometric software.19 Model 1 – Regression analysis based on trading and potential non-trading responses The base case regression analysis (model 1) included all respondents who passed the two tests of rationality and who did not choose the no screening option for each of the 16 pairwise questions. From this model we were able to determine how important each of the three characteristics was to an individual’s choice of a CRC screening program. The coefficient for each characteristic (ßi) shows the relative impor tance of a marginal change in that characteristic. The willingness of individuals to trade the harms and benefits is the rate at which they give up one unit of a characteristic for an increase in another.20 One unit refers to one e xtra colonoscopy or one CRC death prevented. For example, ß 1 / ß2 enumerates how many colonoscopies an individual is willing to trade to prevent one CRC death. (A multinomial model was also used to test for the effect of no screening responses on overall benefit scores.) Interaction model In addition to the main effects regression models, an interac- 14 (9) 31 (20) 74 (46) 31 (20) 8 (5) 28 (10) 51 (17) 147 (50) 53 (18) 13 (5) Likelihood of being screened for CRC within the next six months Definitely will be screened 11 (8) Probably will be screened 23 (17) Probably will not be screened 44 (33) Definitely will not be screened 5 (4) Can’t say 34 (26) Need more info on screening 16 (12) Degree of difficulty in making choices Extremely easy 38 (28) Easy 58 (42) Neither easy nor difficult 26 (19) Difficult 11 (8) Ver y difficult 5 (3) Satisfied with choices Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree 58 (42) 71 (51) 8 (6) 1 (1) 0 tion model was generated to test whether there were any significant interactions between the main attributes and the effect of knowing someone with CRC or respondents’ characteristics on preferences. We hypothesised that respondents who had a f amily member or friend with CRC would be more likely to choose the screening program with the highest level of benef its, regardless of the harms or notif ication policy. Non-significant variables were excluded from the interaction model. Estimating benefits of alternative screening programs To assess the level of an individual’ satisf action with CRC s screening, a ‘benefit score’ was derived from the multinomial regression model for alternative combinations of benefits and har ms. This was done by multiplying the level of benefit and har m by 10 (6) 16 (10) 65 (41) 5 (3) 47 (30) 16 (10) 39 (24) 55 (34) 39 (24) 21 (13) 9 (5) 54 (33.5) 85 (53) 13 (8) 9 (6) 1 (0.5) 21 (7) 39 (13) 109 (37) 10 (3) 81 (28) 32 (12) 77 (26) 113 (38) 65 (21) 32 (10) 14 (5) 112 (37) 156 (52) 21 (7) 10 (3) 1 (1) the regression coef ficient for that attribute and summing these to derive a total weighted benef it score.21 For the purposes of informing policy, we generated a comparator screening program and two alternatives (A and B) to demonstrate how our DCM results could assist policymakers to choose the optimal screening program among those who are prepared to be screened for CRC (that is, those people who are ‘screening-friendly’). Model 2 – Analysis based on trading responses only A second ‘reduced regression model’ (model 2) also was estimated in which we excluded data from ‘potential non-trading’ VOL. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings respondents. This allowed a direct comparison of the f inal regression results with and without the potential non-trading responses. In DCM studies, deriving an estimate of benef it depends on the respondents trading between attributes. The standard test for this ‘trading condition’ is to assess whether an individual always chooses according to the best level of a given attribute. For the two ordinal attrib utes where ‘best’ could be defined, we tested whether an individual always chose the option with the highest le vel of benef its or the lowest number of unnecessar y colonoscopies. It is difficult to distinguish high valuers from nontraders, however. 2 Non-traders would always choose the ‘best’ level for one attribute only, no matter what is on offer.2 High valuers are those people who, if pushed, still trade. They may trade at more extreme levels of harms and benef its. Respondents who stated at the end of the interview that they based their choice on one attribute only or who said that they developed a decisionheuristic were classif ied as non-traders. specification was appropriate. The McFadden’ R2 statistic was s 25% indicating sound goodness-of-f it for both models.16 Model 1 used responses from 258 participants (see Table 3). Co-efficients for each characteristic were significant, indicating that respondents regarded them as being important in their preferences for CRC screening. Each extra CRC death pre vented increased the benef it score by 0.15 (per 10,000 screened). Each extra colonoscopy reduced the benefit score by 0.00013 (per 10,000 screened). Having a notification policy whereby participants would be notified of any test result irrespective of results increased the benef it score by 0.45. The willingness of individuals to trade the benefits and potential harms was examined by dividing the regression co-efficient for each characteristic. In model 1 (n=258), respondents were willing to accept up to 1,179 colonoscopies (95% CI 1,127-1,185) per 10,000 men and women screened biennially over 10 years for one CRC death prevented. Interaction model The only signif icant interaction between choice and respondent characteristics was for subjects who had a family member with CRC and CRC deaths prevented (but not for unnecessary colonoscopies or notif ication). For this group, each extra CRC Results Table 2 summarises the characteristics of the 301 respondents who completed the interview (response rate 74%). The mean duration of the discrete choice interview was 38 minutes (range 1565 minutes). No respondents chose program B for the warm-up question. We excluded data from nine respondents (3%) who failed the two consistenc y questions posed within the 18 screening profiles. death prevented increased utility by an additional 0.034 (per 10,000 screened) compared with those respondents who did not have a f amily member with CRC (p<0.001). Estimating benefits of alternative screening programs In Table 4, total weighted benef it scores are shown for the two alternative CRC screening programs, using the results of our regression analysis. Notif ication policy was held constant. Program Model 1 – Regression analysis based on trading and potential non-trading responses Within the f inal sample of 292, 34 respondents (12%) chose ‘no screening’ for ever y one of the 18 choice questions, no matter how man y CRC deaths were pre vented or unnecessar y colonoscopies incurred. Only one respondent variable was signif icantly associated with this persistent preference for ‘no screening’. Specifically, 50% of the ‘no screening’ respondents knew someone with cancer compared with 69% for those respondents who chose to be screened at least once (χ 2 = 4.44, 1df, p=0.03). For model 1, p was significant, indicating that a random ef fects A is characterised by a more sensitive FOB test than the comparator, producing more test positives (requiring colonoscopy) b ut yielding more benef its (CRC deaths prevented). As the benef it score was greater than zero (0.729), our results suggest the community would support program A. In the second example, program B is characterised by a more sensitive FOB test than the comparator, producing more test positives than program A but fewer CRC deaths prevented. However, the benefit score is Table 3: Random effects probit regression results, with and without ‘potential non-traders’. Explanatory variables Model 1 (including potential non-traders) (n=258) ß coeff Benefit Harm Notification ρ Log likelihood McFadden’s R 2 Chi-square Count R-squareda 0.150 –0.00013 0.45 0.012 –2,321 0.25 1,579 75% ( p<0.001) Model 2 (e xcluding potential non-traders) (n=164) (std error) (0.005) (0.000006) (0.02) (0.006) <0.001 <0.001 <0.001 <0.05 ß coeff 0.119 –0.00014 0.536 0.025 –1528 0.22 870 69% (std error) (0.0067) (0.000007) (0.026) (0.008) <0.001 <0.001 <0.001 <0.01 ( p<0.01) Note: (a) The number of correct predictions/number of obser vations. 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article negative (-0.072). Hence the extra benefit would not be sufficient to compensate for the additional colonoscopies. In this case, the ‘screening-friendly’ community would prefer the less sensitive test (A). Moving from the comparator screening program to test (A) would result in a net gain to society whereas switching to test (B) would result in a net loss (where the harms outweigh the benef its). Discussion The introduction of CRC screening in Australia raises some important and immediate challenges with respect to e videncebased consumer choice. Our results provide useful insights regarding community preferences with cur rent technology and also a robust method for assessing changes in community preferences as new tests become available or other attributes of organised CRC screening are introduced. Our discrete choice results can be used to inform the selection of FOB tests in CRC screening that yield var ying ratios of harms to benefits. Policymakers should attempt to reach the highest benefit score (as illustrated in Table 4) within an available budget. Thus, the population should be offered CRC screening with attributes that assure the highest benefit score.20 In this study, cost was held constant across choices. If it was allowed to var y, willingness to pay (WTP), a monetary measure of benefit, could have been estimated for the different attrib utes, as well as for different ways of providing the service. 22 Including such an attribute in a DCM is useful at the policy level, since it allows benef its to be directly compared with costs within the framework of a cost-benefit analysis frame work. However, the inclusion of a monetary attribute raises important issues. Given the precedent of mammographic screening, Australians will have the expectation that CRC screening will be free of charge at the point of consumption. This raises the question of what is the appropriate payment vehicle w hen including a monetary attribute.23 For example, Jan used changes in the Medicare levy as a proxy for cost5 and Ryan et al.22 used travelling costs as a pro xy. Another study20 introduced a time attribute, waiting time, and then used its value to impute an indirect WTP Future work should . explore including a price proxy within a DCM evaluating different CRC screening tests. An important assumption w hen applying the DCM approach is that subjects are willing to trade between attributes. We found Model 2 – Regression analysis based on trading responses only Of the 258 respondents who were consistent in their responses to the screening options, 94 were identif ied as potential non-traders (PNT). That is, they always chose according to the ‘best’ level of a particular attrib ute. Of the 94 potential non-traders, 48 stated that they did not base ‘all’ of their choices on one sole attribute. Further, 43 out of the 48 stated they had developed a decision heuristic ‘rule of thumb’ when making their choices. This suggests that nearly all of the 94 PNTs were not trading. Among PNT, 65% of women said that they based their choice on one factor only compared with 35% of men (χ 2=5.6, 1df, p=0.017). There were no statistically significant differences in the sociodemographic characteristics of ‘traders’ compared with PNTs. Excluding these 94 ‘potential non-traders’, 164 respondents (‘traders’) were analysed for model 2 (see Table 3). In model 2 (n=164), respondents were willing to accept up to 853 colonoscopies (95% CI 838-860) per 10,000 men and women screened biennially over 10 years for one CRC death prevented. This range of values, namely from 853 to 1,179, indicates the community’s preference for the ratio of har m to benefits among those who are prepared to be screened for CRC (that is, those people who are ‘screening-friendly’). Fifteen per cent of respondents found making the choices difficult, 21% found it neither easy nor difficult yet 89% of all respondents were satisfied with the choices they made (see Table 2). Table 4: Estimated total weighted benefit scores for four CRC screening programs. Attribute Example 1 Benefit Harm Notification Total weighted benefit score 0.153 –0.000126 0.45 14 5,000 1 8 3,500 1 Coefficient Hypothetical CRC screening program (A) Comparator CRC screening program Difference in characteristic level 6 1,500 0 Benefit score (difference x coefficient) 0.918a –0.189a 0 0.729 Attribute Example 2 Benefit Harm Notification Total weighted benefit score Coefficient Hypothetical CRC screening program (B) 10 6,500 1 Comparator CRC screening program 8 3,500 1 Difference in characteristic level 2 3,000 0 Benefit score (difference x coefficient) 0.306a –0.378a 0 –0.072 0.153 –0.000126 0.45 Note: (a) Benefit or harm per 10,000 men and women screened (biennial FOB) over 10 years. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH VOL. 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings that 56% of respondents were willing to trade CRC deaths prevented for the process attribute of being notif ied of a test result, whether positive or negative. This finding supports the Pilot Bowel Cancer Implementation policy of notifying all screenees of their test result. Further, having a family member with CRC increases the marginal benef it derived from screening. A preference for no screening may be related to characteristics of the CRC screening program and/or characteristics of the respondents. Future work in applying DCM to screening should explore the usefulness of nested logit models in examining the reasons for nonparticipation in screening, alongside further qualitative work. One-third of the sample were not willing to trade. Of immediate interest to policy makers is that g roup of 46 people who stated they based their choice on benefits onl y. For this group, the downsides or harms of screening are irrelevant to their choices. They will choose the screening option that offers the g reatest number of lives saved. Women are signif icantly more likely than men to base their choice on benefits only. Identifying one-third of the total sample as potential non-traders is not unusual for DCMs in health care. Scott notes that other DCM studies have found between 30% and 71% of respondents do not appear to trade.24 Regarding the degree of difficulty in making the choices, that is, thinking about the value of changes within characteristics of CRC screening (thinking ‘at the margin’), we would not expect individuals to f ind it easy. While 15% of respondents found making the choices difficult, 89% were satisf ied with their choices. They were confident with their response. Further work needs to be conducted within DCM studies on whether non-trading behaviour reflects a ver y strong preference for a particular attribute (and that indi viduals are not prepared to trade it off) or whether the attribute levels within the choices on offer are insufficient to induce trading. Findings will be relevant to policymakers seeking to assess the extra benef its and costs of proposed changes to an existing health service. Acknowledgements We thank Sandra Easter and Tracey Bruce for outstanding data collection and Jane Young for her comments on an earlier draft of this paper. This project was funded by a National Health and Medical Research Council project grant. This project was approved by the Human Ethics Committee of The University of Sydney. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Australian and New Zealand Journal of Public Health Wiley

Evidence‐based consumer choice: a case study in colorectal cancer screening

Loading next page...
 
/lp/wiley/evidence-based-consumer-choice-a-case-study-in-colorectal-cancer-1IyEa0zOzK

References (30)

Publisher
Wiley
Copyright
Copyright © 2003 Wiley Subscription Services, Inc., A Wiley Company
ISSN
1326-0200
eISSN
1753-6405
DOI
10.1111/j.1467-842X.2003.tb00425.x
Publisher site
See Article on Publisher Site

Abstract

Glenn Salkeld Screening Test Evaluation Program, School of Public Health and Surgical Outcomes Research Centre, The University of Sydney, New South Wales Abstract Objectives: To elicit community preferences for colorectal cancer (CRC) screening by faecal occult blood test (FOBT) using discrete choice modeling (DCM). To provide policymakers with information that would assist them in designing the future national screening program. Michael Solomon Department of Colorectal Surgery and Surgical Outcomes Research Centre, The University of Sydney, New South Wales Leonie Short Centre for Public Health, Queensland University of Technology Methods: 301 participants in central Sydney, aged 50 to 70 years , at ‘average’ risk of CRC, participated in a face-to-face discrete choice study interview in which screening profiles were posed to derive estimates for preferences for CRC FOBT screening. Results: Three character istics were varied in our screening profiles, namely: benefit (CRC deaths prevented); potential harm (false positive induced colonoscopy); and notification policy (of test result). Ninety-four respondents (32%) did not trade off CRC deaths prevented for any reduction in harms. Twelve per cent alw ays chose no screening. The remaining 56% traded benefits and harms. These latter respondents (n=164) were willing to accept 853 (false positive induced) colonoscopies for one CRC death prevented. Mandy Ryan Health Economics Research Unit, University of Aberdeen, Scotland Jeanette E. Ward Division of Population Health, South West Sydney Area Health Service, New South Wales powerful shift is emerging in the way health care is transacted. Research suggests that consumers want more infor mation about their health care options and, for some, a more active role in making decisions about their treatment.1 The movement towards e vidence-based information and the centrality of individual patient choice and values in decision making has been described as a shift towards ‘evidence-based consumer choice’.1 At the same time, there is growing interest in the use of community preferences to inform health services planning and delivery. 2 Together, evidence-based consumer choice and recognition of community preferences in health planning pose new challenges for population health such as screening for colorectal cancer (CRC). Elsewhere, the General Medical Council in the United Kingdom (UK) has guidelines about information that should be given to people considering screening tests.3 These guidelines state that information should be provided about the purpose of screening, its uncertainties such as the risks of a f alse positive or false negative result and any significant medical, social and f inancial implications of screening for individuals and their relatives.3 To our knowledge, there are no similar guidelines in Australia. Yet recent Australian research has shown that women want to be well informed about the possibility of f alse test results when participating in mammography screening and the risks of any adv erse effects of tests.4 Women repor t receiving information on the benefits of tests and treatments signif icantly more often than they recei ved information on adv erse effects. 4 Consumer preferences ought to influence the deliver y of CRC screening. There are a number of ways to elicit consumer preferences, either revealed preferences (based on actual choices) or stated preference (based on hypothetical choices). We used stated preference to assist policymakers in the design of a future national screening program. Stated preference methods include the concept of QALYs (quality adjusted life years), Conclusions: While survival was all that mattered for just over one-third of the sample and 12% would choose no screening, the remaining individuals were prepared to trade CRC deaths prevented against other character istics. CRC screening will not receive unqualified community support, irrespective of har ms. Implications: In any future national CRC screening program, consideration of these insights about community assessment of benefits, harms, costs and other characteristics of CRC screening is warranted. ( Aust N Z J Public Health 2003; 27: 449-55) Correspondence to: Dr Glenn Salkeld, School of Public Health, University of Sydney, A27, NSW 2006. Fax: (02) 9351 7420; e-mail: glenns@health.usyd.edu.au Submitted: December 2002 Revision requested: April 2003 Accepted: June 2003 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article discrete choice modelling (DCM) or contingent valuation, based on ‘willingness to pay’ (WTP). QAL were not pursued in our Ys study because we wanted to value ‘process’ attributes as well as ‘outcomes’ attributes. We chose DCM over WTP because the latter does not provide a quantitative estimate of the relative importance of harms, benefits and information to potential screenees.2 DCM involves asking respondents to choose between different options for treatment or services based on the salient characteristics (or attributes) of that service. It has been advocated in the UK to measure patient preferences for clinical ser vices Locally, Jan et al. used DCM to elicit community preferences for hospital services in South Australia. Specifically, they asked respondents to indicate their preference for two types of hospital services based on different levels of attributes such as travel time to hospital, waiting times for elective surgery or emergency ser vices, cost and complication rates.5 By analysing respondents’ choice between two options, the authors demonstrated that the community chooses on the basis of complication rates and length of waiting times for elective surgery and in the emergency depar tment.5 In public health, DCM has been applied to primary health care services,6 varicella vaccination, 7 genetic screening,8 breast cancer screening9 and the Rotary Bowelscan (a community-based CRC testing program).10 The aim of this present study was to assess community decisions about CRC screening based on the trial evidence on harms and benef its. The community in this study was the eligible CRC screening population, men and women aged 50-70 years, living in central Sydney. We also aimed to provide policy makers with infor mation that would assist them in designing the future national screening program.11 ‘strongly agreed’ would encourage their par ticipation. The DCM approach assumes that individuals consider the information and will trade between the attributes. As this is less likely to happen as the number of attributes increases, we selected three attributes as follows: benef it (CRC deaths prevented); potential harms (the chance of a false positive test result and requiring an unnecessar y colonoscopy); and notif ication policy (specifically, either notification of any test result irrespective of its f indings or notification only of a positi ve test result). Stage 2: Allocating levels to attributes of CRC screening Next, we assigned various plausible levels to each of the three attributes (see Table 1). Levels for CRC deaths prevented and colonoscopies due to a false positive test result were based on a systematic review of trial data for CRC screening.13-15 The range of benefits represented the upper and lower 95% confidence intervals for CRC deaths prevented. Similarly the range of harms represented the upper and lower 95% conf idence intervals for false positive induced colonoscopies. As stage 1 had shown the term ‘f alse positive induced colonoscopies’ was not widely understood, we expressed this as ‘unnecessary colonoscopies’ in stage 3, explaining the ter m at the beginning of our interview. For notification policy, there were only two possible levels (namely being notified whether positive or negative or being notified only when positive). Stage 3: Experimental design Thus, our DCM consisted of two attributes each with four levels and one attribute with two levels, generating 32 possible screening profiles (combinations of benef it, potential harm and notification policy)(see Table 1). A study design was used whereby 16 pairwise choices were generated from these 32 possibilities.16 As recommended, 2 profiles were combined to ensure minimum correlation between each of the pairwise choices. As far as possible, levels for each attribute occur red an equal number of times.2 In addition to the 16 pairwise choices, an additional two choices were included to test for rationality. In both, program A was unambiguously better than program B (that is, it had more CRC Table 1: Characteristics and le vels used for a discrete choice study of CRC screening. Characteristic Number of CRC deaths prevented (BENEFIT) Methods We followed DCM principles and protocols as follows.2 Stage 1: Determining attributes Through informal networks of community groups, we first recr uited 54 men and women aged 50 to 70 years to participate in one of eight focus groups. To achieve a breadth of views, these focus groups consisted of two g roups of women aged 50-59 years, one of men aged 50-59 years, three of women aged 60-69 years and two groups of men aged 60-69 years. Consensus emerged from the focus groups as to the characteristics that might influence people to participate in CRC screening: namely, the organisation of and recr uitment for screening, test accuracy, performing the test, survival gains, policy regarding notif ication of test results, delivery of the test kit and the cost of the test. Following this, and as part of this representative DCM study, 791 men and women aged 50-70 years and living in central Sydney completed self-administered questionnaires. Recruitment is reported in detail elsewhere.12 Respondents were asked to rate the importance of those characteristics previously identified in the focus groups according to whether it would encourage them to participate in CRC screening.12 From these responses, we identified f ive attributes of CRC screening that at least 85% of respondents either ‘agreed’ or Levels • 3 per 10,000 screened • 8 per 10,000 screened • 14 per 10,000 screened • 20 per 10,000 screened • 100 per CRC death prevented • 300 per CRC death prevented • 600 per CRC death prevented • 800 per CRC death prevented Number of unnecessary colonoscopies (HARM) Negative test result (NOTIFICA TION) • Notified of either a positive or negative test result • Notified only of a positive test result and not notified of a negative test result AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH VOL. 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings deaths prevented and fewer colonoscopies than the alternative). Notification polic y was identical in each. Failing one rationality test may be due to random er ror.2 Failing both rationality tests suggests irrational decision making.2 Therefore individuals who failed both tests of rationality were dropped from the analysis. Stage 4: Data analysis Based on the assumption that respondents would choose the option that provided them with the highest level of satisfaction (or utility), the following benefit equation (model 1) was estimated for consistent respondents who chose a screening option: ∆B = ß1BENEFIT + ß2HARM + ß3NOTIFICATION where ∆B is the change in benef it (utility) in moving from screening program B to program A and the explanatory variables Figure 1: Outline of interview schedule. Respondents first were given an outline of the purpose of the interview. Inter viewees were then provided with the following information: Background • A descr iption of what happens with respect to the diagnosis of CRC in the absence of screening. • A descr iption of CRC screening, what it involves and an example of a FOB test kit. • A statement on the oppor tunity cost of a national CRC screening prog ram, specifically ‘A screening program like this would be expensive. It would cost the Commonwealth Government tens of millions of dollars each year. That is money that has to come from somewhere – either from increased taxation or reductions in other health programs or both.’ CRC screening • Asked to imagine that they are invited by their local doctor (GP) to ha ve this test free of charge. • A descr iption of two identical CRC biennial screening programs for a population of 10,000 men and women aged between 50-69 years. • A statement that the programs w ere identical except for the three char acteristics for which standardised descriptions were provided: benefits (number of CRC deaths prevented); harms (number of colonoscopies due to a false positive FOB test result); and notification policy (whether or not they would be notified of a negative test result). • A descr iption on what is involved in having a colonoscopy and a statement on the rate of complications for that scenar io ar ising from colonoscopies performed on the screened population. ‘Warm up’ choice question Could y ou please compare the two programs and tell me whether you would prefer Program A, Program B or whether you would prefer not to have the screening test. EXAMPLE ONLY: Per 10,000 men and women screened over a 10-year period. Example scenario Program A Program B Number of CRC deaths prevented 20 3 Number of unnecessary colonoscopies 2,000 2,400 Notification of negative test result Yes Yes Which w ould you prefer? Program A, Program B or no screening? Stage 4: Collecting the data Interview schedule Our face-to-face interview commenced with a detailed explanation of what happens in the absence of screening, the nature and purpose of CRC screening, the benef its and potential har ms, a description of colonoscopy as a diagnostic test, the possible complications of colonoscopy and, f inally, the nature and purpose of the DCM questionnaire (see Figure 1). Respondents were told that screening would be organised through their GP and would be free of charge. Hence, these two organisational characteristics, cost and screening by GPs, were fixed in the DCM study. Respondents then were taken through a warm-up exercise. In the ensuing 18 pair-wise choices, respondents were asked to nominate their preference. For e xample, Figure 2 shows how respondents were asked to choose between the additional six CRC deaths prevented in program B against 6,000 fewer false positive-induced colonoscopies in program A. Potential har ms and benefits were presented as absolute frequencies to help individuals make more accurate judgements.17 To help us interpret individuals’ choices we included in the interview schedule two closed questions in which respondents were asked whether they made all of their choices based on one attribute only and, if so, whether that attribute was benefits, harms or notification policy. Respondents were also asked whether or not they developed a decision-heuristic (a rule of thumb ) to help them cope with the potentially difficult task of trading between two risk-based attributes from a large set of 18 discrete choice questions. Further details on these questions are presented in Analysis. At the end of the interview, we asked respondents to rate their level of agreement with the statement ‘I am satisfied with my decisions overall’ on a f ive-point Liker t scale (‘strongly agree’ to ‘strongly disagree’). We also asked them to rate how ‘easy’ or ‘difficult’ they found making the choices, using another five-point Likert scale (‘v ery easy’ to ‘ver y difficult’). Subjects Respondents for the DCM consisted of 414 respondents to the earlier rating exercise who had consented to par ticipate in future Figure 2 Could you please compare the two programs and tell me whether y ou would prefer Program A, Program B or whether you would prefer not to have the screening test. Per 10,000 men and women screened over a 10-year per iod Example scenario Program A Pr ogram B Number of CRC deaths prevented 8 14 Number of unnecessary colonoscopies 2,400 8,400 Notification of negative test result Yes Yes Which w ould you prefer? Program A, Program B or no screening? stages of the study. Of these, five w ere excluded because the y had been previously diagnosed with CRC and, therefore, would be ineligible for screening. Thus, 409 (209 male and 200 female) participants were eligible for interview. Of these, 301 (138 men and 163 women) completed it and 108 did not consent to be interviewed. Elsewhere,13 we had collected socio-demographic information, including two terms to assess whether or not the respondent had a family member with CRC and/or knew someone with cancer. 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article Table 2: Socio-demographic and CRC risk characteristics of the 301 respondents, by gender. Men n=138 n % Age group 50-60 61-70 Highest level of education Didn’t complete secondary Completed secondary T rade or technical qual. University or college Employment status Employed Not employed Marital status Married Not married Household income <$20,000 pa $20,000 to $35,000 pa $35,001 to $50,000 pa >$50,000 pa Country of birth Australia Outside Australia Health status (self report) Excellent Ver y good Good Fair Poor Perceived individual risk of CRC (compared with someone of own A lot less Somewhat less The same Somewhat more A lot more 90 (65) 48 (35) 23 (18) 29 (21) 35 (26) 48 (35) 99 (73) 37 (27) 106 (78) 30 (22) 12 (9) 16 (12) 42 (32) 60 (46) 98 (71) 40 (29) 28 (20) 45 (33) 46 (33) 17 (12) 2 (1) age) 14 (10) 20 (15) 73 (55) 22 (16) 5 (4) Women n=163 n % 78 (48) 85 (52) 24 (15) 70 (44) 55 (12) 47 (29) 72 (45) 88 (55) 105 (65) 157 (35) 30 (21) 34 (23) 32 (22) 49 (34) 119 (73) 44 (27) 34 (21) 49 (30) 58 (36) 18 (11) 4 (2) Total n=301 n % 168 (56) 133 (44) 47 (16) 99 (34) 54 (18) 95 (32) 171 (57) 125 (43) 211 (71) 87 (29) 42 (15) 50 (18) 74 (27) 109 (40) 217 (72) 84 (28) 62 (21) 94 (31) 104 (35) 35 (12) 6 (2) are the differences in the attributes of the two screening programs, A and B, as shown in Table 1. There is no constant term in the equation because respondents were told that all aspects of the two screening programs were identical other than those specif ied in the questionnaire. A probit model was used to analyse the discrete choice data. In such a model, there is potential for correlation between observations from each respondent and errors in the model not being independent.18 Hence, a random effects probit model was used to test for the presence of intra-observer correlations using LIMDEP econometric software.19 Model 1 – Regression analysis based on trading and potential non-trading responses The base case regression analysis (model 1) included all respondents who passed the two tests of rationality and who did not choose the no screening option for each of the 16 pairwise questions. From this model we were able to determine how important each of the three characteristics was to an individual’s choice of a CRC screening program. The coefficient for each characteristic (ßi) shows the relative impor tance of a marginal change in that characteristic. The willingness of individuals to trade the harms and benefits is the rate at which they give up one unit of a characteristic for an increase in another.20 One unit refers to one e xtra colonoscopy or one CRC death prevented. For example, ß 1 / ß2 enumerates how many colonoscopies an individual is willing to trade to prevent one CRC death. (A multinomial model was also used to test for the effect of no screening responses on overall benefit scores.) Interaction model In addition to the main effects regression models, an interac- 14 (9) 31 (20) 74 (46) 31 (20) 8 (5) 28 (10) 51 (17) 147 (50) 53 (18) 13 (5) Likelihood of being screened for CRC within the next six months Definitely will be screened 11 (8) Probably will be screened 23 (17) Probably will not be screened 44 (33) Definitely will not be screened 5 (4) Can’t say 34 (26) Need more info on screening 16 (12) Degree of difficulty in making choices Extremely easy 38 (28) Easy 58 (42) Neither easy nor difficult 26 (19) Difficult 11 (8) Ver y difficult 5 (3) Satisfied with choices Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree 58 (42) 71 (51) 8 (6) 1 (1) 0 tion model was generated to test whether there were any significant interactions between the main attributes and the effect of knowing someone with CRC or respondents’ characteristics on preferences. We hypothesised that respondents who had a f amily member or friend with CRC would be more likely to choose the screening program with the highest level of benef its, regardless of the harms or notif ication policy. Non-significant variables were excluded from the interaction model. Estimating benefits of alternative screening programs To assess the level of an individual’ satisf action with CRC s screening, a ‘benefit score’ was derived from the multinomial regression model for alternative combinations of benefits and har ms. This was done by multiplying the level of benefit and har m by 10 (6) 16 (10) 65 (41) 5 (3) 47 (30) 16 (10) 39 (24) 55 (34) 39 (24) 21 (13) 9 (5) 54 (33.5) 85 (53) 13 (8) 9 (6) 1 (0.5) 21 (7) 39 (13) 109 (37) 10 (3) 81 (28) 32 (12) 77 (26) 113 (38) 65 (21) 32 (10) 14 (5) 112 (37) 156 (52) 21 (7) 10 (3) 1 (1) the regression coef ficient for that attribute and summing these to derive a total weighted benef it score.21 For the purposes of informing policy, we generated a comparator screening program and two alternatives (A and B) to demonstrate how our DCM results could assist policymakers to choose the optimal screening program among those who are prepared to be screened for CRC (that is, those people who are ‘screening-friendly’). Model 2 – Analysis based on trading responses only A second ‘reduced regression model’ (model 2) also was estimated in which we excluded data from ‘potential non-trading’ VOL. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings respondents. This allowed a direct comparison of the f inal regression results with and without the potential non-trading responses. In DCM studies, deriving an estimate of benef it depends on the respondents trading between attributes. The standard test for this ‘trading condition’ is to assess whether an individual always chooses according to the best level of a given attribute. For the two ordinal attrib utes where ‘best’ could be defined, we tested whether an individual always chose the option with the highest le vel of benef its or the lowest number of unnecessar y colonoscopies. It is difficult to distinguish high valuers from nontraders, however. 2 Non-traders would always choose the ‘best’ level for one attribute only, no matter what is on offer.2 High valuers are those people who, if pushed, still trade. They may trade at more extreme levels of harms and benef its. Respondents who stated at the end of the interview that they based their choice on one attribute only or who said that they developed a decisionheuristic were classif ied as non-traders. specification was appropriate. The McFadden’ R2 statistic was s 25% indicating sound goodness-of-f it for both models.16 Model 1 used responses from 258 participants (see Table 3). Co-efficients for each characteristic were significant, indicating that respondents regarded them as being important in their preferences for CRC screening. Each extra CRC death pre vented increased the benef it score by 0.15 (per 10,000 screened). Each extra colonoscopy reduced the benefit score by 0.00013 (per 10,000 screened). Having a notification policy whereby participants would be notified of any test result irrespective of results increased the benef it score by 0.45. The willingness of individuals to trade the benefits and potential harms was examined by dividing the regression co-efficient for each characteristic. In model 1 (n=258), respondents were willing to accept up to 1,179 colonoscopies (95% CI 1,127-1,185) per 10,000 men and women screened biennially over 10 years for one CRC death prevented. Interaction model The only signif icant interaction between choice and respondent characteristics was for subjects who had a family member with CRC and CRC deaths prevented (but not for unnecessary colonoscopies or notif ication). For this group, each extra CRC Results Table 2 summarises the characteristics of the 301 respondents who completed the interview (response rate 74%). The mean duration of the discrete choice interview was 38 minutes (range 1565 minutes). No respondents chose program B for the warm-up question. We excluded data from nine respondents (3%) who failed the two consistenc y questions posed within the 18 screening profiles. death prevented increased utility by an additional 0.034 (per 10,000 screened) compared with those respondents who did not have a f amily member with CRC (p<0.001). Estimating benefits of alternative screening programs In Table 4, total weighted benef it scores are shown for the two alternative CRC screening programs, using the results of our regression analysis. Notif ication policy was held constant. Program Model 1 – Regression analysis based on trading and potential non-trading responses Within the f inal sample of 292, 34 respondents (12%) chose ‘no screening’ for ever y one of the 18 choice questions, no matter how man y CRC deaths were pre vented or unnecessar y colonoscopies incurred. Only one respondent variable was signif icantly associated with this persistent preference for ‘no screening’. Specifically, 50% of the ‘no screening’ respondents knew someone with cancer compared with 69% for those respondents who chose to be screened at least once (χ 2 = 4.44, 1df, p=0.03). For model 1, p was significant, indicating that a random ef fects A is characterised by a more sensitive FOB test than the comparator, producing more test positives (requiring colonoscopy) b ut yielding more benef its (CRC deaths prevented). As the benef it score was greater than zero (0.729), our results suggest the community would support program A. In the second example, program B is characterised by a more sensitive FOB test than the comparator, producing more test positives than program A but fewer CRC deaths prevented. However, the benefit score is Table 3: Random effects probit regression results, with and without ‘potential non-traders’. Explanatory variables Model 1 (including potential non-traders) (n=258) ß coeff Benefit Harm Notification ρ Log likelihood McFadden’s R 2 Chi-square Count R-squareda 0.150 –0.00013 0.45 0.012 –2,321 0.25 1,579 75% ( p<0.001) Model 2 (e xcluding potential non-traders) (n=164) (std error) (0.005) (0.000006) (0.02) (0.006) <0.001 <0.001 <0.001 <0.05 ß coeff 0.119 –0.00014 0.536 0.025 –1528 0.22 870 69% (std error) (0.0067) (0.000007) (0.026) (0.008) <0.001 <0.001 <0.001 <0.01 ( p<0.01) Note: (a) The number of correct predictions/number of obser vations. 2003 VOL. 27 NO . 4 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Salkeld et al. Article negative (-0.072). Hence the extra benefit would not be sufficient to compensate for the additional colonoscopies. In this case, the ‘screening-friendly’ community would prefer the less sensitive test (A). Moving from the comparator screening program to test (A) would result in a net gain to society whereas switching to test (B) would result in a net loss (where the harms outweigh the benef its). Discussion The introduction of CRC screening in Australia raises some important and immediate challenges with respect to e videncebased consumer choice. Our results provide useful insights regarding community preferences with cur rent technology and also a robust method for assessing changes in community preferences as new tests become available or other attributes of organised CRC screening are introduced. Our discrete choice results can be used to inform the selection of FOB tests in CRC screening that yield var ying ratios of harms to benefits. Policymakers should attempt to reach the highest benefit score (as illustrated in Table 4) within an available budget. Thus, the population should be offered CRC screening with attributes that assure the highest benefit score.20 In this study, cost was held constant across choices. If it was allowed to var y, willingness to pay (WTP), a monetary measure of benefit, could have been estimated for the different attrib utes, as well as for different ways of providing the service. 22 Including such an attribute in a DCM is useful at the policy level, since it allows benef its to be directly compared with costs within the framework of a cost-benefit analysis frame work. However, the inclusion of a monetary attribute raises important issues. Given the precedent of mammographic screening, Australians will have the expectation that CRC screening will be free of charge at the point of consumption. This raises the question of what is the appropriate payment vehicle w hen including a monetary attribute.23 For example, Jan used changes in the Medicare levy as a proxy for cost5 and Ryan et al.22 used travelling costs as a pro xy. Another study20 introduced a time attribute, waiting time, and then used its value to impute an indirect WTP Future work should . explore including a price proxy within a DCM evaluating different CRC screening tests. An important assumption w hen applying the DCM approach is that subjects are willing to trade between attributes. We found Model 2 – Regression analysis based on trading responses only Of the 258 respondents who were consistent in their responses to the screening options, 94 were identif ied as potential non-traders (PNT). That is, they always chose according to the ‘best’ level of a particular attrib ute. Of the 94 potential non-traders, 48 stated that they did not base ‘all’ of their choices on one sole attribute. Further, 43 out of the 48 stated they had developed a decision heuristic ‘rule of thumb’ when making their choices. This suggests that nearly all of the 94 PNTs were not trading. Among PNT, 65% of women said that they based their choice on one factor only compared with 35% of men (χ 2=5.6, 1df, p=0.017). There were no statistically significant differences in the sociodemographic characteristics of ‘traders’ compared with PNTs. Excluding these 94 ‘potential non-traders’, 164 respondents (‘traders’) were analysed for model 2 (see Table 3). In model 2 (n=164), respondents were willing to accept up to 853 colonoscopies (95% CI 838-860) per 10,000 men and women screened biennially over 10 years for one CRC death prevented. This range of values, namely from 853 to 1,179, indicates the community’s preference for the ratio of har m to benefits among those who are prepared to be screened for CRC (that is, those people who are ‘screening-friendly’). Fifteen per cent of respondents found making the choices difficult, 21% found it neither easy nor difficult yet 89% of all respondents were satisfied with the choices they made (see Table 2). Table 4: Estimated total weighted benefit scores for four CRC screening programs. Attribute Example 1 Benefit Harm Notification Total weighted benefit score 0.153 –0.000126 0.45 14 5,000 1 8 3,500 1 Coefficient Hypothetical CRC screening program (A) Comparator CRC screening program Difference in characteristic level 6 1,500 0 Benefit score (difference x coefficient) 0.918a –0.189a 0 0.729 Attribute Example 2 Benefit Harm Notification Total weighted benefit score Coefficient Hypothetical CRC screening program (B) 10 6,500 1 Comparator CRC screening program 8 3,500 1 Difference in characteristic level 2 3,000 0 Benefit score (difference x coefficient) 0.306a –0.378a 0 –0.072 0.153 –0.000126 0.45 Note: (a) Benefit or harm per 10,000 men and women screened (biennial FOB) over 10 years. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH VOL. 27 NO. 4 Methods and Concepts Evidence-based choice in colorectal cancer screenings that 56% of respondents were willing to trade CRC deaths prevented for the process attribute of being notif ied of a test result, whether positive or negative. This finding supports the Pilot Bowel Cancer Implementation policy of notifying all screenees of their test result. Further, having a family member with CRC increases the marginal benef it derived from screening. A preference for no screening may be related to characteristics of the CRC screening program and/or characteristics of the respondents. Future work in applying DCM to screening should explore the usefulness of nested logit models in examining the reasons for nonparticipation in screening, alongside further qualitative work. One-third of the sample were not willing to trade. Of immediate interest to policy makers is that g roup of 46 people who stated they based their choice on benefits onl y. For this group, the downsides or harms of screening are irrelevant to their choices. They will choose the screening option that offers the g reatest number of lives saved. Women are signif icantly more likely than men to base their choice on benefits only. Identifying one-third of the total sample as potential non-traders is not unusual for DCMs in health care. Scott notes that other DCM studies have found between 30% and 71% of respondents do not appear to trade.24 Regarding the degree of difficulty in making the choices, that is, thinking about the value of changes within characteristics of CRC screening (thinking ‘at the margin’), we would not expect individuals to f ind it easy. While 15% of respondents found making the choices difficult, 89% were satisf ied with their choices. They were confident with their response. Further work needs to be conducted within DCM studies on whether non-trading behaviour reflects a ver y strong preference for a particular attribute (and that indi viduals are not prepared to trade it off) or whether the attribute levels within the choices on offer are insufficient to induce trading. Findings will be relevant to policymakers seeking to assess the extra benef its and costs of proposed changes to an existing health service. Acknowledgements We thank Sandra Easter and Tracey Bruce for outstanding data collection and Jane Young for her comments on an earlier draft of this paper. This project was funded by a National Health and Medical Research Council project grant. This project was approved by the Human Ethics Committee of The University of Sydney.

Journal

Australian and New Zealand Journal of Public HealthWiley

Published: Aug 1, 2003

There are no references for this article.