Personalized colorectal cancer screening: study protocol of a mixed-methods study on the effectiveness of tailored intervals based on prior f-Hb concentration in a fit-based colorectal cancer screening program (PERFECT-FIT)

In 2014, a national population-based colorectal cancer (CRC) screening program was implemented in the Netherlands. Biennial fecal immunochemical testing (FIT) for hemoglobin is used at a cut-off of 47 µg (µg) hemoglobin (Hb)/g (gram) feces. The CRC screening program successfully started, with high participation rates and yield of screening resulting in a decrease in overall and advanced-stage CRC incidence [1‐3]. Now that the program has reached a steady state, there is potential to further optimize the program.

Every year, about 2.2 million people are invited to participate in the Dutch CRC screening program. The participation rate is about 72% [4]. About 4.5% of participants has a positive FIT, meaning they have a fecal hemoglobin (f-Hb) concentration above the pre-set FIT cut-off [4]. Of these participants, about 85% undergo a colonoscopy, with around 40% of these people having a relevant finding (6% CRC and 36% advanced adenoma (AA)) [4]. This implies that about 98% of participants in CRC screening do not experience any benefit from screening; 95.5% of participants because they have a negative FIT and 2.7% because they have a positive FIT without relevant findings at colonoscopy.

Ideally, screening should be offered primarily to those who would benefit most, that is, those who are at high risk of the disease. Personalized screening has been discussed for a long time (about 25 years) [5]. To date, however, such an approach has not taken off, due to the limited predictive power of a number of known risk factors [6]. A risk model that combined genetic information with lifestyle factors, family history and sex had a discriminatory power of 63% for predicting CRC risk [10].

There is increasing evidence that f-Hb concentration is a good predictor of future diagnosis of advanced neoplasia (AN) (Table 1). Models incorporating f-Hb concentrations could reach a discriminatory power of about 80% [6‐10]. The major advantage of this predictive factor is that the f-Hb concentration is automatically obtained within FIT-based CRC screening programs and thus is readily available information. The likelihood that the integration of tailored invitation intervals based on prior f-Hb concentration after negative FIT lowers the participation rate is therefore smaller than if another (not automatically obtained) risk factor would be used to personalize CRC screening. Sex and age are also automatically registered, but their predictive value is much lower than the f-Hb concentration (odds ratios for AN: 1.6 (male sex) and 0.9–1.1 (increasing age) vs. 2.5–21.8 (increasing f-Hb concentrations), respectively [7]). In addition, a strong association was observed between the measured f-Hb concentration in participants with a negative FIT and the risk of developing interval CRC in the Dutch CRC screening program [11]. Interval CRC is defined as CRC diagnosed after a negative FIT and before invitation to the next screening round [12]. Participants in the category with an f-Hb concentration just below the FIT cut-off (15–46.9 µg Hb/g feces) are 13 times more likely to develop an interval CRC compared to participants with an unmeasurable f-Hb concentration (0 µg Hb/g feces) [personal communication].

Almost half of all interval CRCs occur in a small group of participants (3.5%) with an f-Hb concentration between 15 and 46.9 µg Hb/g feces [13]. Two-thirds of these cancers occur in the second year after screening [13]. This means that one-third of interval CRCs could potentially have been prevented by inviting only 3.5% of participants to screening one year earlier. Based on more recent data, we expect around 85% of participants to have an f-Hb concentration of 0 µg Hb/g feces and thus to be at lowest risk of developing an interval CRC. If the interval between invitations for this group would be extended by one year, this would represent a 40% reduction in the screening burden for the population as a whole.

Now that the FIT-based CRC screening program has been fully rolled out in the Netherlands, has high participation rates and shows favorable results, there is potential for further optimization of the CRC screening program. We designed a mixed-methods study consisting of: (1) a parallel group, two-arm, superiority randomized controlled trial (RCT), (2) focus group studies, and (3) decision modelling. The aim of this mixed-methods study is to identify the yield and (cost-) effectiveness of personalized CRC screening, whether it could be feasible within population-based CRC screening programs, and whether the population is able to understand and accept it.

At three time points during the study, a focus group study will be conducted.

We will use the well-established MIcrosimulation SCreening ANalysis for CRC (MISCAN-Colon) model [19, 20] to estimate harms, benefits, resources and costs of uniform screening with a biennial interval and compare that with those of personalized screening intervals of 1, 2 or 3 years based on prior f-Hb concentrations.

Outcome of the modelling study is the long-term (cost-) effectiveness of personalized screening by using prior f-Hb concentrations. Long-term outcomes include CRC incidence, CRC-related mortality, (quality-adjusted [QA]) life-years [LYs] gained, false-positive tests, colonoscopy complications, and costs, which will be compared for personalized screening versus uniform screening in the Dutch population.

MISCAN-colon was developed by the Department of Public Health of Erasmus MC to evaluate the cost-effectiveness of different CRC screening policies, and it has been used to inform CRC screening policy in the Netherlands, the United States, Canada, and Australia [20‐23]. In brief, the MISCAN-Colon model simulates the life histories of a large population of individuals from birth to death and has a natural history component that tracks the progression of underlying colorectal disease in the absence of screening. As each simulated individual ages, there is a chance that one or more adenomas may develop depending on age, sex, race and individual risk. Adenomas can progress from small (1–5 mm) to medium (6–9 mm) to large (≥ 10 mm) size, and some may eventually become malignant. A preclinical cancer (i.e., not detected) has a chance of progressing through different stages and may be detected by symptoms at any stage. With screening, adenomas and preclinical cancers may be detected depending on the sensitivity of the screening test for that lesion and, for endoscopic tests, whether the lesion is within reach of the endoscope.

The aim of this study is to evaluate the yield, feasibility, acceptability and (cost-) effectiveness of personalized CRC screening through tailored invitation intervals based on prior f-Hb concentrations. This personalized approach could contribute to a better balance between the harms and benefits of CRC screening on both an individual and population level.

Introducing tailored invitation intervals results in both direct and indirect consequences of personalized CRC screening. Direct consequences are the detection of precancerous lesions or CRC at an earlier stage, as well as reduction of the number of interval CRCs in individuals at higher risk for CRC, by offering specific individuals a shorter invitation interval. In the long-term, this could contribute to a lower burden of CRC-related morbidity and mortality. By inviting participants with an f-Hb concentration just below the cut-off (> 15–46.9 µg Hb/g feces) at a shorter interval, it is expected that, compared to uniform CRC screening, slightly more people will test false positive compared to true positives. Still, the balance of benefits and harms in the high-risk group is expected to be at least as favorable as that of individuals in the low-risk groups. In these low-risk groups, less intensive screening intervals ensures lower burden of screening. There will potentially be an increase in the incidence of interval CRCs in this group because participants will be invited to CRC screening one year later. However, our hypothesis is that the reduction in screening burden clearly outweighs the potential small increase in incidence of interval CRCs. Altogether, it is expected that the balance between harms and benefits of personalized CRC screening will be more favorable compared to uniform CRC screening.

Indirect consequences of implementing personalized CRC screening include ethical and communication challenges [24]. When introducing personalized CRC screening to individuals, there could be confusion between screened individuals living in the same household if they are invited after different time intervals. Another disadvantage could be that those individuals who receive a longer invitation interval will experience stress from the longer waiting time, because of the increased risk of interval CRC. Therefore, providing clear and explicit information on the different invitation intervals based on an individual’s risk is of great importance. The focus group studies will provide invaluable information on perspectives on and acceptability of personalized CRC screening that can be used when personalized CRC screening is potentially introduced at a population level.

It is inevitable that the direct and indirect consequences of personalized CRC screening versus uniform CRC screening will need to be assessed, should personalized screening eventually be implemented at the population level. Possible benefits of a personalized screening approach (i.e., increase in detection of AN, decrease in false-positives, overtreatment, etc.) should be monitored closely, as well as predicted long-term outcomes (i.e., CRC incidence, CRC-related mortality, QALY’s gained, cost-effectiveness). If successful, this study will not only provide evidence for personalized CRC screening, but will also be an important benchmark for quality assurance in future implementation of personalized CRC screening, similar as previous pilot studies preceding the implementation of the Dutch CRC screening program have been for the current uniform program [14, 25‐29].

Some limitations or our study should be mentioned. The design of our study is fixed and based on the current test (FIT; FOB-Gold; Sentinel Diagnostics, Milan, Italy), cut-off (47 μg Hb/g feces) and age range (individuals aged 55–75) used in the Dutch CRC screening program. Nevertheless, even if the CRC screening program would be modified in terms of test, cut-off or age range, we expect that the results of our study are still relevant: the effect of the risk factor f-Hb holds for all ages, and the literature shows that it also holds for other cut-offs and FIT brands [7, 11, 13‐17, 30‐32]. Furthermore, even if the decision should be made to use another test instead of FIT, the study is still informative on the acceptability of risk-based screening in general. Obviously, there will always be organizational and political aspects that need to be considered when planning the real-time implementation of personalized CRC screening [24]. Nevertheless, by embedding this study in the current and ongoing CRC screening program in the Netherlands, it is hoped and expected that (most of) these challenges can be overcome.

We expect there are many future directions in personalized CRC screening; more information will become available on outcomes of multiple screening rounds and on well-known risk factors such as age and sex. Furthermore, in the future other risk factors might also be collected by default within the IT infrastructure, such as lifestyle and genetic (i.e., single nucleotide polymorphisms) factors [24]. If we can implement these risk factors in (advanced) prediction models, the risk prediction for personalized CRC screening can be even further improved, for example through better identification and categorization of the risk groups. If this study demonstrates that personalized CRC screening is successful, such a development would only make risk-based screening more favorable than uniform screening.

In conclusion, the aim of this study is to identify the yield and (cost-) effectiveness of personalized CRC screening, whether it could be feasible within population-based programs, and whether the population is able to understand and accept it. This knowledge may be of guidance for health policy makers and may provide evidence for implementing personalized CRC screening in the Netherlands and/or other countries.