Colorectal Cancer Screening: How Health Gains and Cost-Effectiveness Vary by Ethnic Group, the Impact on Health Inequalities, and the Optimal Age Range to Screen

Indigenous and socially disadvantaged groups carry a disproportionate cancer burden in most countries, with these groups often having high incidence and poorer survival than non-indigenous and advantaged groups (1–3). There is a consistent pattern of underserving these groups by health care systems. Organized cancer screening programs offer potential to reduce inequalities in health through reductions in cancer incidence and mortality; however, generally screening programs are not designed with an equity focus and can exacerbate inequities due to poorer access to and through them for underserved populations (4). Given this, it is important to assess the impact of any proposed cancer screening programs in respect to their impact on underserved populations and inequalities in health, and to identify any modifiable factors that may help to maximize health gains (and inequality reductions) from proposed cancer screening programs. In this study, we aimed to examine the likely health gains and inequality impacts from the national rollout of a colorectal cancer screening program on the indigenous Māori population of New Zealand using a modified cost-effective approach.

Colorectal cancer screening with fecal occult blood testing (FOBTi) is effective at lowering colorectal cancer mortality (5), and has been found to be cost-effective for 50- to 74-year-olds in a number of countries, including Australia (6), England (7), and Ireland (8). New Zealand is currently considering the national rollout of a colorectal cancer screening program; however, little is known about how colorectal cancer screening affects indigenous health inequalities. Part of the difficulty in examining the impact of screening on indigenous populations results from the inconsistent collection of data on ethnicity/race (9). We are aware of only one international study that has quantified likely ethnic or racial group differences in the health gains and cost-effectiveness of colorectal cancer screening. Theuer and colleagues found that colorectal cancer screening was more cost-effective for blacks than whites, Latinos, and Asians, due to higher colorectal cancer incidence and worse survival (10).

New Zealand is a good setting for exploring the impact of colorectal cancer screening on health inequalities given the routine collection of ethnicity data in health datasets, and linkage of many of these datasets. Indigenous peoples in Australia and New Zealand have a common pattern of lower colorectal cancer incidence (although in New Zealand the gap in incidence is reducing over time; ref. 1), but worse survival once diagnosed (2, 11). This pattern raises the question as to whether colorectal cancer screening will produce less health gain due to lower incidence, or equal or greater health gain due to improving survival by incidence stage shift. Also, given that Māori have worse survival than non-Māori in the New Zealand screening naïve context, and there is evidence of greater delays to treatment for Māori compared with non-Māori (11), there is a case that survival will improve (more) for Māori than non-Māori as a result of standardized improvements in treatment following the implementation of a well-organized screening program. Complicating this pattern is the impact of lower screening coverage often achieved for indigenous populations (12). In this study, we explore the above issues through a number of equity analyses, substituting non-Māori values for Māori values of background (noncolorectal cancer) morbidity and mortality, colorectal cancer survival and incidence, screening coverage, and stage-specific survival (equal treatment scenario). We also consider the optimal age range for screening, which has not previously been examined by ethnicity or indigeneity (13).

Cost-effectiveness analyses (CEA) provide a useful framework for examining differences in the cost-effectiveness of interventions and inequalities in health gains between population groups; however, some of the assumptions in CEA methods, such as using ethnic-specific life expectancy to calculate health gains (e.g., quality-adjusted life-year; QALY), may disadvantage indigenous populations, and these need to be addressed (14). Despite the importance of CEAs in decision-making, a recent (2015) systematic review found only 19 studies worldwide on the cost-effectiveness of health-related interventions with a focus on indigenous populations, with no studies on cancer screening (15).

There are a range of current approaches to incorporate equity concerns into cost-effectiveness modeling methods in the literature (16, 17). These approaches loosely fit into two groups: (i) those that aim to include the social value of health gains between groups using methods such as equity weighting and the social welfare function; and (ii) those that attempt to incorporate equity of healthcare into the models by altering the configuration of the intervention(s) to achieve process and/or outcome equity. This article is an example of the latter. The specific objectives of this study were: (i) to estimate the differences in health gain, and cost-effectiveness of a colorectal cancer screening program of 50- to 74-year-olds by ethnic group (Māori vs. non-Māori); (ii) to explore the factors that drive differences in health gains between Māori and non-Māori, including background morbidity and mortality, colorectal cancer incidence screening coverage, and survival; (iii) to measure the impact of the intervention on absolute inequalities in quality-adjusted life expectancy (QALE) for Māori compared with non-Māori; and (iv) to determine whether the optimal (based on cost-effectiveness) screening age range differs by sex and ethnic group.

We used a Markov model to estimate the health gains in QALYs, costs, and cost-effectiveness of a national CRC screening program with biennial FOBTi of 50- to 74-year-olds (default age group for initial analyses). The model consisted of a colorectal cancer submodel and a general part which integrated results from the submodel with general population characteristics. Two populations were modeled: a reference one without screening and an intervention one. The differences between the two determined benefits and costs.

Included in the model were all individuals ages 35 years and over in 2011, modeled through to death or age 110 years. We assumed this population to be screen-naïve. A multiple cohort approach was used, that is, each five-year age-group in 2011 was modeled separately until death or age 110 years. A cycle length of 1 year was used in the modeling. The model was implemented in Microsoft Excel, using the Ersatz add-in for uncertainty analysis (Microsoft, www.Epigear.com). Ethical approval was received from the University of Otago Ethics Committee H13/049.

As much as possible, we captured socio demographic heterogeneity by sex, age, and ethnic group (Māori and non-Māori) for colorectal cancer incidence and mortality, transition parameters, survival, background mortality and morbidity, and also in colorectal cancer screening participation. A health system perspective was used, with a 3% discount rate. Unrelated health system costs were included, that is, average expected costs to the health system by sex and age, including for years of life saved. We are able to include these unrelated health care costs owing to the very detailed nature of New Zealand data (18). This allows a much fuller approach to cost-effectiveness analysis and is in keeping with recent guidelines (i.e., including the inclusion of health care costs associated with extra lifespan as a result of the intervention; ref. 19). We note our BODE3 protocol is to include unrelated health care costs in all evaluations (20). We assumed a cost-effectiveness threshold value based upon GPD per capita, that is, NZ$45,000 per QALY (21).

The colorectal cancer submodel (Fig. 1) generated adenoma incidence, preclinical and clinical colorectal cancer incidence and prevalence, and colorectal cancer mortality rates by Dukes stage and disease prevalence for a non-screening population as compared with various screening options. The colorectal cancer submodel was developed following Whyte and colleagues (2011; ref. 22), with two differences. Unlike the Whyte model, we did not include a direct link from normal epithelium to Dukes A (see Discussion). Second, Whyte and colleagues modeled age-specific incidence of low risk adenomas for each age group separately; to avoid implausible age-specific patterns we modeled the age pattern using a Weibull distribution. Other model parameters from Whyte and colleagues (e.g., progression rates between adenoma and Dukes' stages, probability to present clinically) were used as priors and calibrated (and fitted) against New Zealand data on incidence (by age and stage, for 2011; ref. 23), and colorectal cancer mortality counts, using the solver tool in Excel. The fitted parameters were then used as priors in a Monte Carlo Markov Chain process to obtain a correlation matrix of all parameters for use in subsequent Monte Carlo analyses. The posterior parameter estimates are shown in Supplementary Table S1. Estimates of 8-year survival, by age and from New Zealand excess mortality rate modeling (24), were specified using a log normal function. Comparisons of model predicted versus observed mortality data are shown for the baseline in Supplementary Fig. S1.

In the model, people detected with low-risk adenomas resided in the state for 1 year only (with attendant costs and disability weights), returning to the healthy population the following cycle after colonoscopy. Those identified as high-risk adenoma moved to annual surveillance colonoscopies, with a dropout rate from surveillance.

The outputs from the colorectal cancer submodel (including QALYs related to colorectal cancer) acted as inputs to the general model which was used to calculate net QALYs gained by combining QALYs from the colorectal cancer intervention with those from all other diseases, net costs and the incremental cost-effectiveness ratio (ICER).

Inputs included proportions of the alive population (screened and unscreened) by sex, ethnic group, and age (single year) in each of the following states: low- and high-risk adenoma, false positive and true positive test results, preclinical and clinical Dukes stage colorectal cancer, and proportions dying each year of colorectal cancer (Table 1). We included 3 months of lead time in the model for those receiving the colorectal cancer screening intervention, by shifting the excess mortality curves for all stages of colorectal cancer to the right by 0.25 year. Our results were insensitive to the amount of lead time.

Our primary analyses included background (i.e., noncolorectal cancer) mortality rates and expected morbidity loss in the main life table. Morbidity loss was estimated by sex, ethnic group, and age, using the total years of life lived with disability (YLD) from a New Zealand Burden of Disease Study (25, 26), divided by the population count in the same group. Health gain could only occur in the envelope of “good health” that reduced with increasing age, and was less for Māori. For example, a Māori woman ages 60 to 64 years has an expected YLD of 0.288, meaning a year of life gained in this population group has a maximum utility value of 0.712.

Intervention costs for screening were counted in each year. Excess costs (i.e., excess to that expected for an “average” New Zealand citizen of the same sex and age (18)) and cost offsets varied for being in the first year of colorectal cancer diagnosis, remission or last 6 months of life if dying of colorectal cancer (Supplementary Fig. S2).

Model parameters were determined by a combination of literature review, data from the New Zealand Bowel Screening Programme Pilot (27), parameters used in previous models published in peer-reviewed literature, and analyses of New Zealand linked health datasets. For each parameter, we determined an “expected value” and distribution for use in probabilistic uncertainty analysis (Table 1).

The colorectal cancer mortality reductions estimated by our model under various screening scenarios were similar to those reported in randomized controlled trials (RCT) and other colorectal cancer screening simulation models (e.g., CISNET and SimCRC; Supplementary Table S2; ref. 28).

We carried out a series of equity analyses: (i) we replaced Māori data with non-Māori values for baseline mortality and morbidity to address the issue that lower life expectancy of indigenous population results in less opportunity for life years gained with screening; (ii) we used non-Māori colorectal cancer incidence and incidence trends for Māori to explore their contribution to differences in ethnic-specific health gains and ICERs; (iii) we assessed the impact of achieving equal screening coverage for Māori; and (iv) we assessed the impact of improvements in treatment for Māori following the implementation of a screening program, given evidence of greater delays to treatment for Māori compared with non-Māori leading to poorer survival (11). For this last scenario, we replaced Māori excess mortality from colorectal cancer with the non-Māori rate, for Māori who are screened. This scenario ignores any improvements to non-Māori survival from treatment improvements, and likely overestimates the improvements for Māori by ignoring the influence of co-morbidities on treatment and survival. The equity analyses were modeled with expected values (rather than Monte Carlo analyses) to give stability in the estimates and to allow the measurement of the change in QALYs between scenarios.

We measured the impact of the colorectal cancer screening intervention on absolute inequalities in QALE by determining the absolute difference in QALE for Māori compared to non-Māori by age, sex, and ethnic group, before and after the intervention. We subtracted the baseline difference (non-Māori minus Māori) in QALE from the intervention difference in QALE and converted to a number of healthy days (by multiplying by 365). To facilitate more specific comparisons of social group impacts, 50- to 54-year-old and 70- to 74-year-old age cohorts were run separately and for selection of the optimal age range to screen, the 40- to 44-year-old age cohort was used. Tornado plots were constructed to examine the impact of uncertainty in each input parameter on uncertainty in the output ICER, by rerunning the model with the 2.5th and 97.5th percentile values of the input parameters (Supplementary Fig. S3).

The modeled colorectal cancer screening intervention for 50- to 74-year-olds resulted in health benefits for all population groups. However, the size of the health gains and the cost-effectiveness of the intervention were greater for non-Māori compared with Māori and for men compared with women (Table 2). Non-Māori health gains were substantially greater than Māori: approximately two times greater for 50- to 54-year-olds (0.058 QALYs per person for non-Māori, c.f. 0.031 for Māori), and five times greater for 70- to 74-year-olds (0.003 c.f. 0.016; Table 2). Gains were greater for men than women aged 50- to 54-year-olds (0.060 c.f. 0.049 QALYs gained) but similar for men and women aged 70- to 74-year-olds (0.015 c.f. 0.014 respectively; Table 2).

The modeled colorectal cancer screening intervention was highly cost-effective for all sex and ethnic groups, given a willingness to pay threshold based on GDP per capita (e.g., NZ$45,000 per QALY) (Fig. 2).

Substituting non-Māori background mortality for Māori background mortality resulted in a 29% increase in Māori QALY gains compared with the default model (9,140 c.f. 7,060 QALYs; Table 3).

We used non-Māori colorectal cancer incidence and trends to explore the contribution of these factors to differences in health gains achieved for Māori compared with non-Māori. The scenario of using non-Māori incidence increased QALY gains, by 23% for Māori (8,730 c.f. 7,060 QALYs; Table 3). In contrast, changing the incidence trend from one that was flat (0% change per year for Māori in default model) to one of reducing background incidence (2.5% reduction per year for under 65 years and 1.5% for 65+ years) resulted in a 17% reduction in QALY gains for Māori (Table 3). The combination of using non-Māori incidence and applying reducing incidence trends, all but cancelled each other out, with a 2% net increase in measured QALY gains compared with the default analysis. Using values of non-Māori background morbidity had a smaller gain in QALYs for Māori, with a 3% increase relative to the default model. In all of the above scenarios, there was little change to the costs (up to a 6% difference from the default costs), and, therefore, changes to the ICERs were primarily driven by changes in the health gains.

We additionally undertook analyses to explore the changes to QALY gains for Māori that might result from plausible changes to the intervention itself. Increasing Māori expected screening coverage to the level for non-Māori (i.e., from 45.4% up to 58.3%) and using the same percentage population that never uptake screening (31.6%), increased Māori QALY gains by 29% compared with the default model. Improving stage-specific survival for Māori to the level of non-Māori only resulted in a 1% increase in Māori QALYs (results not shown).

The colorectal cancer intervention increases absolute inequalities in QALE for Māori compared with non-Māori. Absolute inequalities increase more for men than women, and the greatest gains are seen in the age groups that immediately begin screening in our model, ages 50 to 74 years (Table 4). Non-Māori men ages 60 to 64 years gain an additional 25.6 (12.5–40.3) healthy days compared with Māori of the same age while non-Māori women aged 30 to 34 years gain and additional 7.2 (−3.9 to 17.8) healthy days (Table 4).

The optimal starting five-year age group was considered from a cost-effectiveness perspective. For both ethnic groupings was 65- to 69-year-olds (Supplementary Tables S3 and S4), then 65 to 74, 60 to 74, 60 to 79, and 55 to 79 (for 10-, 15-, 20-, and 25-year age ranges, respectively).

Population cancer screening programs offer the potential to both improve population health and address inequalities in health; however, national screening programs consistently underserve indigenous populations and, therefore, risk increasing indigenous health and healthcare inequalities. From our study, national population colorectal cancer screening in New Zealand using biennial FOBTi is estimated to be cost-effective (similar to studies in other countries; refs. 6, 8, 29), and would result in health gains for both Māori and non-Māori. However, due to the lower incidence of colorectal cancer, higher background mortality and likely lower screening coverage, this program is likely to increase inequalities in QALE for Māori compared with non-Māori. This finding is consistent with findings from Australia's national colorectal cancer screening program which has achieved lower coverage for the indigenous population of Australia (which also has a lower incidence of colorectal cancer) and has increased inequalities in cancer outcomes for indigenous Australians (12).

Achieving equal screening coverage for indigenous populations is possible, while rare (30). Multiple parallel strategies would be required across the screening pathway to achieve equal bowel cancer screening coverage and healthcare access for Māori (including active Māori engagement in all aspects of the program's development and implementation, active recruitment and follow-up of eligible Māori participants, and close monitoring of treatment access and quality) (31). In our model, we estimate that achieving equal coverage for Māori and non-Māori would increase the modeled health gains for Māori by 29%—not enough so that Māori gains were equivalent to those for non-Māori. Colorectal cancer is one of the only a few cancers for which Māori in New Zealand have lower incidence, and, therefore, even with equal screening coverage the modeled health gains from colorectal cancer screening for Māori are less. It is theoretically possible to achieve equal health gains by increasing screening coverage for Māori to a level high enough to offset the lower colorectal cancer incidence and higher background mortality. However, these resources may be better used elsewhere to reduce indigenous health inequalities, perhaps, on diseases or interventions that offer even greater potential for inequalities reductions, for example, tobacco control.

As a part of implementing a national screening program, there is likely to be greater standardization of care and monitoring of the screening pathway (including treatment). This improvement in colorectal cancer care will offer benefits to all colorectal cancer cases (screened and unscreened), but potentially even greater benefits for indigenous populations who experience less access to and quality of healthcare compared with non-indigenous populations. Providing “equal treatment” is important in reducing inequalities in healthcare, but in our model, achieves relatively minor health gains (only 1% increase in modeled health gains for Māori).

In this study, we used a cost-utility approach to consider a range of equity scenarios. By standardizing background levels of morbidity and mortality, we acknowledge that existing inequalities in health are unfair but modifiable, and their inclusion in disease and economic models further works to further disadvantage these groups (14).

Our model was a macro-simulation model. Therefore, we did not incorporate the possible impacts of heterogeneity in adenoma growth rates that may affect cost-effectiveness. Nevertheless, our model performed similarly to other simulation studies (Supplementary Table S2).

Unlike Whyte and colleagues (2011; ref. 22) and Sharp and colleagues (2012; ref. 8), we did not include a transition probability directly from normal epithelium to colorectal cancer (i.e., Dukes A). Whyte suggests that this pathway is uncommon (22). Such cancers will benefit less from screening (as there are no adenomas to be detected), meaning we may have modestly over-estimated the health gains from screening (i.e., our model is calibrated to incidence and mortality data, meaning we assume all cancers have an “early detectable phase”, whereas in reality they do not). To the best of our knowledge, any such bias is likely constant across sex, age, and ethnic groups, meaning our social group comparisons are unlikely to be biased for this reason.

Within our model, we use the costs published from the NZ pilot study (27), and applied them evenly across population groups. This spread the cost of any additional activities that were undertaken to increase screening for specific groups across the entire population, compared with explicitly modeling an increase in costs for underscreened people (which would see the marginal cost-effectiveness for these groups worsen). It was beyond the scope of this article, and data quality, to undertake such a marginal analysis of increased recruitment costs per screened person.

One possible limitation of our Markov modeling is allowing for lead time bias. We used a lead time of three months for all stages of disease, consistent across age and ethnic groups. Our sensitivity analyses, of varying lead times made inconsequential differences to QALYs gained and net costs. However, the adjustment for lead time in macrosimulation models is relatively crude, and, therefore, there may be some residual lead-time bias in our model, which would overestimate modeled colorectal cancer screening effectiveness.

There are numerous strengths to our modeling and data inputs. We used detailed New Zealand epidemiologic and cost data to both specify parameters (e.g., the transition probabilities) and as input parameters themselves. Key inputs including of background morbidity and mortality, and colorectal cancer incidence and survival varied by sex, ethnic group, and age, allowing us to examine social group differences in modeled health gains and ICERs.

Although a national colorectal cancer screening program in New Zealand is likely to be cost-effective and improve total population health, this will likely come with the unfortunate consequence of increasing inequalities in health. Based upon our model, the most effective way of limiting the inequalities likely to be generated by this program is to give attention to improving screening coverage.

No potential conflicts of interest were disclosed.

Conception and design: M. McLeod, D. Sarfati, N. Wilson, T. Blakely

Development of methodology: M. McLeod, M. Boyd, J. Barendregt, T. Blakely

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Kvizhinadze, M. Boyd, J. Barendregt, T. Blakely

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. McLeod, G. Kvizhinadze, M. Boyd, J. Barendregt, T. Blakely

Writing, review, and/or revision of the manuscript: M. McLeod, G. Kvizhinadze, M. Boyd, J. Barendregt, D. Sarfati, N. Wilson, T. Blakely

Study supervision: T. Blakely

We thank the New Zealand Ministry of Health for its development of Health Tracker, which allows for high-quality health system costing data in New Zealand.

The BODE3 Programme receives funding support from the Health Research Council of New Zealand (10/248). The funding body has no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.