In May 2011, the Dutch government decided to implement a national programme for colorectal cancer (CRC) screening using biennial faecal immunochemical test screening between ages 55 and 75. Decision modelling played an important role in informing this decision, as well as in the planning and implementation of the programme afterwards. In this overview, we illustrate the value of models in informing resource allocation in CRC screening using the role that decision modelling has played in the Dutch CRC screening programme as an example.
- COLORECTAL CANCER SCREENING
Statistics from Altmetric.com
Decision models synthesise all relevant available data and can be used to extrapolate trial findings and generate information to support optimal resource allocation in colorectal cancer (CRC) screening.
When using models to inform health policy, it is important to select a well-validated model for the analysis and closely monitor outcomes of the screening programme in comparison with model predictions.
On several occasions during the decision, planning and implementation phase of the Dutch national CRC screening programme, MISCAN-Colon model results have influenced the programme. Without modelling, other choices might have been made, possibly resulting in a less favourable balance between the benefits and harms of screening.
We believe that the MISCAN-Colon microsimulation model has contributed and will continue to contribute to the success of the Dutch national CRC screening programme.
More than 1 million people worldwide are diagnosed with colorectal cancer (CRC) each year.2 Half of these patients die from the disease, making CRC the fourth leading cause of cancer death in the world.2 Randomised controlled trials (RCTs) have shown that screening can prevent many of these deaths by detecting CRC in an earlier stage or by detecting and removing its precursor lesion: the adenoma.3 ,4
Although RCTs are the gold standard for determining the effectiveness of screening, they also have their limitations. First, RCTs are expensive and time consuming. As a result, the number of RCTs that have evaluated CRC screening is limited. Until now, only guaiac faecal occult blood test (gFOBT) and sigmoidoscopy screening have been evaluated by means of an RCT. Screening modalities such as the faecal immunochemical test (FIT) and colonoscopy, although expected to be more effective, have not. For the same reasons, so far, direct comparisons between different CRC screening modalities, as well as comparisons between different screening strategies using the same screening modality (eg, annual vs biennial gFOBT screening), have never been made. Second, RCTs usually have a limited follow-up time. As a result, they cannot be used to determine lifetime health effects and costs, which is necessary to determine the (cost-)effectiveness of screening. Third, the effectiveness of screening might differ from setting to setting. For example, a sigmoidoscopy screening trial in Norway showed a 63% attendance rate,5 while in the Netherlands an attendance rate of only 32% was observed.6 This will impact the comparative (cost-)effectiveness of sigmoidoscopy screening compared with FIT screening, for example. Finally, country-level resource demands for a certain screening programme cannot easily be inferred from an RCT. To summarise: RCTs alone do not answer the question of which screening strategy is optimal for a certain country. This might explain the large differences between the screening programmes that are currently implemented in the European Union (table 1).
Decision models provide a useful tool to extrapolate evidence from RCTs and address the question of which screening strategy is optimal given local conditions with respect to CRC risk, life expectancy, resource availability and population preferences, which is the central question in the decision phase of a CRC screening programme. This is the phase in which models have been most frequently used. However, modelling is also valuable in the phases afterwards: during the planning, implementation and evaluation of a screening programme. In this paper, we will illustrate the value of models during the whole cycle of a screening programme using the role of the MISCAN-Colon model in the Dutch CRC screening programme as an example.
MISCAN-Colon decision model
The Dutch CRC screening programme has been co-informed by MISCAN-Colon. MISCAN-Colon is a microsimulation model for CRC developed at the Department of Public Health of the Erasmus University Medical Center (Rotterdam, the Netherlands). The model's structure, underlying assumptions and calibration are described extensively in a standardised model profile (available at http://cisnet.cancer.gov/colorectal/profiles.html) and previous publications.7 ,8 In brief, MISCAN-Colon simulates the life histories of a large population of individuals from birth to death. CRC arises in this population according to the adenoma–carcinoma sequence.9 ,10 More than one adenoma can occur in an individual, and each adenoma can independently develop into CRC. Adenomas may progress in size from small (≤5 mm) to medium (6–9 mm) to large (≥10 mm), and some adenomas will eventually become malignant. Cancer can progress from a localised stage I cancer to a metastasised stage IV cancer. However, during each stage, there is a probability of the cancer being diagnosed due to symptoms. At any time during the development of the disease, the process may be interrupted because the individual dies of another cause.
Screening will alter some of the simulated life histories: some cancers will be prevented by the detection and removal of adenomas; other cancers will be detected in an earlier stage with a more favourable survival. However, screening can also result in serious complications and overdiagnosis and overtreatment of CRC (ie, the detection and treatment of cancers that would not have been diagnosed without screening). By comparing all life histories with screening with the corresponding life histories without screening, MISCAN-Colon quantifies the benefits of screening, as well as the associated harms and costs.
MISCAN-Colon was calibrated to data on the age-specific, stage-specific and localisation-specific incidence of CRC in the Netherlands11 and the age-specific prevalence and multiplicity distribution of adenomas as observed in autopsy and colonoscopy studies.12–22 Furthermore, MISCAN-Colon was calibrated to the reductions in CRC incidence and mortality observed in RCTs evaluating the effectiveness of screening with either gFOBTs or flexible sigmoidoscopy and showed good concordance with these trials results.8 ,23 ,24
The value of MISCAN-Colon in informing the Dutch CRC screening programme
The run-up to the Dutch CRC screening programme is characterised by a long history of decision-making and planning by various stakeholders. Figure 1 gives an overview of the most important milestones in this process. The process and milestones as well as the role MISCAN-Colon has played in this process are described in more detail below.
The discussion on population screening for CRC was initiated in the Netherlands by a report of the Dutch Health Council in 2001.25 This report not only recommended that feasibility studies and screening trials should be conducted, but also that a simulation model should be developed in order to make well-founded judgements about screening strategies. In 2003 and 2004, more landmark reports were published stressing the need to implement a national CRC screening programme.26–28 In response to these signals, the Netherlands Organisation for Health Research and Development and the Dutch Cancer Society joined forces and organised a consensus development meeting in February 2005 in which both public health researchers (in favour of faecal occult blood test (FOBT) screening) and clinicians (in favour of endoscopy screening) participated. During this meeting, consensus was reached to perform population screening with FOBT biennially with the specific test (FIT or gFOBT), the cut-off for referral to colonoscopy in case of an FIT and the age range for screening to be decided within 2–3 years based upon further research.29
In 2006 and 2007, screening trials were conducted in the Amsterdam, Rotterdam and Nijmegen areas to compare attendance rates and detection rates of advanced neoplasia for different FOBTs. More than 30 000 individuals aged between 50 and 74 were randomly selected from the municipal registries and randomised to receive either a gFOBT (Hemoccult II, Beckman Coulter, USA) or an FIT (OC-Sensor, Eiken, Japan). For the FIT, a cut-off for referral to colonoscopy of 50 ng/mL buffer (10 µg/g faeces) was applied, which allowed us to also calculate positivity and detection rates for higher cut-offs. All three regions showed higher attendance and detection rates for FIT compared with gFOBT screening.30 ,31 FIT detection rates were higher at lower cut-offs, but applying a low cut-off also required substantially more colonoscopies. What could not be estimated from the trials was whether the health benefit of detecting more advanced neoplasia justified the additional upfront costs of colonoscopies. To answer that question, MISCAN-Colon was developed and adjusted to reproduce the positivity and detection rates of gFOBT and FIT screening as observed in the Dutch screening trials. The model was subsequently used to predict the costs and effects of different screening strategies, varying the test and cut-off for referral to colonoscopy, as well as the age range and screening interval to determine the optimal FOBT screening strategy for the Dutch setting.
Figure 2 presents the outcomes of this analysis.32 Each symbol in the graph represents a screening strategy. The higher the symbol in the graph, the more effective the strategy, the more to the right, the more expensive. The strategies lying on the top-left, which are connected by the solid line, form the efficient frontier, that is, the economically rational subset of choices.33 Symbols lying beneath the efficient frontier represent strategies that are not as effective for the given amount of money as a point lying on the efficient frontier. These strategies are ‘dominated’ by (combinations of) other strategies. All gFOBT strategies clearly lie beneath the efficient frontier, and hence, are dominated by FIT strategies. In other words, FIT screening is more effective than gFOBT screening at lower cost. The strategies that form the efficient frontier all consist of FIT screening with a cut-off for referral to colonoscopy of 50 ng/mL (10 µg/g), indicating that screening using this low cut-off does not only result in more life-years (LYs) gained, but that the higher upfront costs of colonoscopies are also more than compensated for by higher future savings on CRC treatment.
Table 2 gives an overview of the FIT strategies on the efficient frontier. The strategy with the lowest costs per LY gained was 3-yearly screening between ages 60 and 69; next came lowering the start age to 55. The age range recommended by the Council of Europe (50–75) was not among the cost-effective options. As said, each of these strategies is an economically rational choice. Which strategy to choose depends on the willingness-to-pay for an LY gained. Generally in the Netherlands, a threshold between €20 000 and €40 000 per quality-adjusted life-year gained is used for preventive interventions. All efficient strategies resulted in costs per LY gained well beneath that threshold, making the most intensive strategy (ie, annual screening between ages 45 and 80) still an appropriate choice in the Dutch setting.
This MISCAN-Colon analysis was an important component of the 2009 Health Council advice on CRC screening.34 The Health Council advised the Minister of Health to implement biennial FIT screening between ages 55 and 75 using a cut-off for referral to colonoscopy of 75 ng/mL (15 µg/g). Based on the outcomes of the cost-effectiveness analysis, a different age range was chosen than recommended by the European Council (ie, 55–75 instead of 50–75) and a different cut-off was chosen than recommended by the FIT manufacturer (ie, 75 ng/mL (15 µg/g) instead of 100 ng/mL (20 µg/g)). The Health Council recognised that applying a lower cut-off for referral to colonoscopy was more cost-effective and that the willingness-to-pay threshold allowed for more intensive FIT screening. However, their choice also reflects the anticipated lack of colonoscopy capacity in the Netherlands to implement such a colonoscopy-intensive programme.34
In January 2010, the Minister of Health responded to the Health Council advice. He acknowledged the value of a nationwide CRC screening programme, but felt forced to postpone a final decision on its implementation.35 The financial climate at that time put the government in a situation of radical cost reductions, so there was no budget for a national CRC screening programme. In addition, the minister considered the anticipated shortage of colonoscopy capacity36 an important bottleneck that needed to be resolved before a national CRC screening programme could be implemented and emphasised the need for a system for quality assurance. He therefore commissioned the National Institute for Public Health and the Environment to investigate the feasibility of a national CRC screening programme in the Netherlands. The purpose of this feasibility study was to ascertain the prerequisites for a CRC screening programme and to determine how such a programme could be introduced successfully. The study should identify potential problems with implementation and suggest how to deal with them, including issues of capacity, communication, quality assurance, flexibility in the light of new technological developments, link with further diagnostics and care, as well as monitoring and evaluation.37
To investigate the issue of capacity, the National Institute for Public Health and the Environment requested Erasmus MC to predict the resource requirements for a national CRC screening programme using MISCAN-Colon.38 The model was used to simulate the Dutch population from 2013 up to 2042 under the implementation of a national CRC screening programme as proposed by the Health Council, including a phased roll-out from 2013 to 2018 (figure 3).
Assuming attendance, positivity and detection rates for FIT screening with a cut-off of 75 ng/mL (15 µg/g), the model predicted the annual numbers of FIT analyses, colonoscopies, histopathological examinations, surgical procedures, as well as the numbers of CRC deaths prevented, and the costs of screening from the anticipated start of the programme in 2013 until 2042 when resource requirements and screening benefits were expected to stabilise (figure 4).37 ,38 A comparison of required and available endoscopy capacity showed that in 2016, 2017 and 2018 there would be a shortage of endoscopy capacity (figure 5). The professional groups proposed that increasing colonoscopy efficiency and increasing the intake to training programmes could overcome the expected shortage in these years.
Based on the outcomes of the feasibility study, a new Minister of Health decided in May 2011 to implement a national CRC screening programme in accordance with the Health Council advice.1 After 2 years of preparation of programme infrastructure, quality assurance protocols and communication materials, the national programme was first piloted in the Rotterdam-Rijnmond area in the fall of 2013 and then gradually rolled out nationally starting in January 2014. Based on the outcome of a public tender, the FOB-Gold (Sentinel, Italy) was chosen as the preferred test. The cut-off for referral to colonoscopy was set at 88 ng/mL (15 µg/g), which corresponds with the 75 ng/mL (15 µg/g) of the OC-Sensor as recommended by the Dutch Health Council. Because of the slight delay in the start of the programme (January 2014 instead of September 2013), not only individuals aged 63, 65, 67 and 75 but also individuals aged 76 were invited in 2014, thereby assuring that these individuals, originally scheduled for screening in 2013, still got a chance to participate in screening at least once.
Because the information technology (IT)-system especially developed for the CRC screening programme allowed for continuous monitoring of the programme, attendance, positivity and detection rates could be tracked real time. The programme was an immediate success. Attendance to the programme was higher than expected (68% vs 60%),39 as was the detection rate of advanced adenomas/CRC (4.0% vs 2.7%).34 ,40 However, the positivity rate was also considerably higher than expected (13.4% vs 6.4%) and the observed positive predictive value for detecting an advanced adenoma/CRC was substantially lower than expected (30.0% vs 42.5%).37 ,40 Consequently, colonoscopy capacity became an important bottle neck and waiting times for diagnostic colonoscopy increased.
Several steps were taken to address this problem. In a first step, positivity and detection rates at several cut-offs as observed in the national programme were compared with those observed in the Rotterdam screening trial. This comparison showed that applying the same cut-off level resulted in a higher positivity and a higher detection rate in the national programme than in the trial (figure 6A, B). The correlation between the positivity rate and the detection rate, however, was strikingly similar (figure 6C). Based on a MISCAN-Colon analysis in which we corrected for the difference in the age distribution between the individuals screened within the Rotterdam screening trial and the national programme, who were substantially older, it was concluded that the test characteristics corresponding to a cut-off level of 75 ng/mL (15 µg/g) as observed in the trial could be reproduced by elevating the cut-off level in the national programme to 275 ng/mL (47 µg/g).
In a second step, MISCAN-Colon was used to quantify the impact of the modified roll-out, the higher-than-expected attendance and the higher-than-expected positivity rate on the anticipated colonoscopy demand for 2014 and to determine which measure could best be taken to reduce this demand. Also, screening 76 year olds in 2014 was found to increase the anticipated colonoscopy demand for 2014 from 28 000 to 33 000 (figure 7). The higher attendance rate further increased this demand to 38 000 colonoscopies. However, the higher positivity rate resulted in the largest increase in colonoscopy demand for 2014: up to 64 000 colonoscopies. By 2030, the higher attendance and positivity rate would result in a doubling of colonoscopy demand compared with what was anticipated based on the feasibility study.
To reduce colonoscopy demand for 2014, two measures could be taken: screening could be postponed until 2016 in one or more of the age groups scheduled for screening in 2014 or the cut-off for referral to colonoscopy could be elevated in all age groups. To determine which of these measures was best suited to reduce colonoscopy demand for 2014, MISCAN-Colon was used to estimate the associated loss in benefit from screening and the reductions in colonoscopy demand for 2014. The best measure to reduce colonoscopy demand for 2014 was defined as the measure that resulted in the largest reduction in colonoscopy demand per CRC death not prevented.
Postponing screening in 75-year-old and 76-year-old individuals, which implies not screening them at all, reduced colonoscopy demand by 21 and 22 per CRC death not prevented, respectively (table 3). Postponing screening in 63-year-old, 65-year-old and 67-year-old individuals was somewhat more efficient, reducing colonoscopy demand by 54, 60 and 53 per CRC death not prevented, respectively. However, temporarily elevating the cut-off for referral in all age groups to 275 ng/mL (47 µg/g) reduced colonoscopy demand by 68 per CRC death not prevented and was most efficient. The National Institute for Public Health and the Environment therefore decided to increase the cut-off for referral to colonoscopy to 275 ng/mL (47 µg/g), starting on 23 July 2014.40 MISCAN-Colon predicted that applying this higher cut-off will result in a similar number of CRC deaths prevented as anticipated based on the 2010–2011 feasibility study (figure 8).
The increased cut-off will be sustained in 2015, but in that same year the MISCAN-Colon model will again be used to compare the increased cut-off with other measures to reduce colonoscopy requirements for the longer term: lengthening the screening interval and narrowing the age range. In addition, when data from repeat screenings become available, the model will be updated to reflect observed positivity and detection rates for repeat screenings and will again be used to predict long-term resource requirements and benefits of the Dutch CRC screening programme. In case of substantial changes in anticipated resource requirements and benefits, the impact of changes to the programme may need to be evaluated anew.
In an established programme that has been running for several years and in which a steady state has been reached, the value of modelling may be less apparent than in the decision, planning and implementation phase of a screening programme. However, modelling also has its value in a well-established programme. In the first place, modelling can be used for evaluation of the screening programme. Is the programme working as expected? What are the expected changes in the long-term impact of the programme based on differences in anticipated and observed programme indicators? Model predictions can be used as a benchmark for observed CRC incidence and mortality to determine whether the programme is having the anticipated benefit. An important example of such work has been done using the MISCAN-breast model. Because screening resulted in a substantial increase in the incidence of breast cancer in women in the screen-eligible age range, there was considerable debate about the amount of overdiagnosis from mammography screening. Using the model, we demonstrated that this increase in incidence could be anticipated and that it is almost completely compensated for by a sharp decrease in breast cancer incidence at older age.41
Second, in every programme, even the well-conducted programmes, there is room for improvement. There might be regional variation in performance indicators (eg, attendance, delay in diagnostic follow-up) and modelling can be used to estimate the impact of this regional variation on long-term outcomes of the screening programme. This way the impact of reducing regional variation can be determined for each indicator and interventions can be prioritised.
Finally, medicine, in general, but CRC screening, in particular, is a continuously developing field, and therefore, a moving target. New technologies for screening may become viable such as computed tomographic colonography, stool DNA testing and serum testing.42–44 In addition, the call for precision medicine, and in that light risk-stratified screening, is increasing.45–47 It is important to continuously follow these developments and determine their potential benefits and harms for an existing CRC screening programme. For example, for CRC screening, it is well known that test characteristics of FIT differ by gender and age48 ,49 and using differential cut-offs for men versus women and older versus younger people has therefore been proposed. Modelling can help determine whether these calls are justified and what the potential benefit of gender-specific and age-specific FIT screening is.
This overview shows that decision modelling played an important role in the decision, planning and implementation phase of the Dutch CRC screening programme, and we believe it will continue to do so in the coming years as it has done for other programmes. On several occasions, model results have influenced the programme: in the decision phase, FIT screening was chosen over gFOBT screening, a higher age to start screening was chosen than that recommended by the Council of Europe and a lower cut-off for referral to colonoscopy was chosen than that recommended by the test's manufacturer. To remediate the higher-than-anticipated colonoscopy demand during the implementation phase, the cut-off for referral to colonoscopy was temporarily elevated. If modelling would not have been available or used, these choices might not have been made and the benefits and harms of the screening programme could have turned out less favourable than they will now.
Validation of model results
This overview describes how modelling has influenced and changed the Dutch CRC screening programme. However, the use of a model does not necessarily imply that the right decisions are made. Policymakers considering using a model to inform their (CRC) screening programmes should be aware of the considerable variation in quality of available models. An important strength of the MISCAN-Colon model is that it has been extensively validated against available evidence from RCTs and other sources, and, where necessary, adapted to accurately predict the impact of screening on CRC incidence and mortality. The good fit with trial results builds confidence in model extrapolations to the Dutch population and other settings. However, it remains very important to closely monitor the outcomes of the screening programme and compare them with model predictions. Important outcomes to consider include detection rates during repeat screening rounds, interval cancer rates and CRC mortality. For example, to validate the decision to increase the cut-off for a positive FIT in the programme, it is important to monitor the interval cancer rate after a negative screening in the programme. This rate should not exceed the rate predicted by the model.
An important example of how model-induced changes to a screening programme were validated by subsequent monitoring of the programme comes from the Dutch cervical cancer screening programme. This programme originally offered women 3-yearly Pap smear testing between ages 35 and 53: a total of seven smears. Evaluation using the MISCAN-Cervix model indicated that spreading those seven smears over a wider age range increased the benefits of screening without increasing its costs.50 Based on this analysis, the cervical cancer screening programme was changed to offer 5-yearly screening between ages 30 and 60. Evaluation of this change several years later showed that the 9-year incidence of cervical cancer after a negative primary smear did not increase.51 This example clearly illustrates how modelling resulted in the right decision to change an existing screening programme.
Conditions for decision modelling
Decision modelling in the Dutch CRC screening programme could only be applied because several critical conditions were met. First of all, the availability of local data on adherence and yield of FIT screening from the Dutch screening trials was essential to reliably estimate the required capacity and long-term impact of FIT screening in the Netherlands. Second, involvement of monitoring and evaluation experts of the Department of Public Health in the development of quality indicators ensured that all indicators relevant for decision modelling were consistently collected in the screening programme. Third, the IT-system developed for the CRC screening programme allowed real-life tracking, and thus, continuous monitoring of all relevant data from the screening programme. These data timely revealed the higher-than-anticipated adherence to and referral rate of FIT screening and allowed for further diagnosis of the problem followed by the model analysis described above. However, perhaps the most important factor was the good collaboration between the Department of Public Health, the National Institute for Public Health and the Environment, and the Dutch Ministry of Health and the willingness of the decision-makers involved to consider model results.
Other examples of applications of decision modelling in screening
The Dutch CRC screening programme is not the only screening programme that has applied modelling to inform the design, planning and implementation of screening. Modelling has also been used to inform the Irish, Canadian and Australian CRC cancer screening programmes.52–54 For Ireland, modelling showed that FIT-based screening would be very effective, but that colonoscopy demand could not be met instantly. A staggered age-based roll-out was therefore suggested to gain time to increase colonoscopy capacity to meet the programme demand.54 In Canada, modelling was used to inform the National Committee on Colorectal Cancer Screening on the mortality reduction, cost-effectiveness and resource requirements of biennial gFOBT screening.53 The expansion of screening ages in the Australian CRC screening programme has been accelerated to occur in the coming 5 years instead of the previously proposed 17 years after a model analysis showed that this would increase the number of CRC deaths prevented in the upcoming 40 years by almost 30%.52 Models were also used to inform the US Preventive Services Task Force (USPSTF) recommendations for lung, breast and CRC screening55–57 and the Centers for Medicare and Medicaid Services coverage decisions for FIT, stool DNA and CT colonography screening.58 ,59 Co-informed by modelling outcomes, the USPSTF no longer recommends routine screening for breast cancer before age 50 and after age 74, nor CRC screening after age 75 in those with an adequate screening history. Interestingly, most examples relate to the use of modelling in the decision phase of screening programmes. The potential of modelling in the planning, implementation and established programme phase is currently underused.
In this overview, we have shown that modelling has been very useful in the decision, planning and implementation phase of the Dutch CRC screening programme. In the absence of a decision model, decisions concerning the programme would have to be made based on expert opinion and implicit assumptions. Decision models synthesise all relevant available data and can be used to extrapolate trial findings and generate information to support optimal resource allocation in CRC screening. When using models to inform health policy, it is important to select a well-validated model for the analysis and closely monitor outcomes of the screening programme in comparison with model predictions. We believe that the MISCAN-Colon microsimulation model has contributed and will continue to contribute to the success of the Dutch CRC screening programme.
Twitter Follow Frank van Hees at @frankvanhees
Contributors FvH and IL-V were responsible for drafting the manuscript. All authors contributed to the conception and design of the study; the generation, collection, assembly, analysis and/or interpretation of data; and the revision of the manuscript. All authors approved of the final version of the manuscript.
Funding This work was made possible by the Dutch National Institute for Public Health and the Environment (contract number: 3910055067) and the Health Programme of the European Union (contract number: 2013 2203). The development of MISCAN-Colon was made possible by the Netherlands Organization for Health Research and Development (contract numbers: 62200022 and 63300022) and the National Cancer Institute as part of the Cancer Intervention and Surveillance Modeling Network (contract number: U01CA152959).
Competing interests None declared.
Provenance and peer review Commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.