Realizing the Promise of Personalized Colorectal Cancer Screening in Practice

Current recommendations for age of initiation of colorectal cancer (CRC) screening and selection of screening strategies such as colonoscopy and fecal immunochemical testing take a one-size-fits-all approach, except for those with a family history of CRC (1,2). Even with risk stratification using risk calculators or family history, it is unclear whether screening is similarly effective across the risk spectrum for both low- and high-risk groups.

In their recent article in the Journal, Wang et al. (3) address these evidence gaps using data from the Nurses’ Health Study and Health Professionals Follow-up Study cohorts. Specifically, they derived a risk score based on 8 CRC risk factors—family history, aspirin, height, body mass index, smoking history, physical activity, alcohol use, and diet—and measured cumulative CRC incidence and mortality across risk groups to 1) determine whether increased risk scores were associated with CRC risk, 2) establish whether exposure to screening colonoscopy reduced risk for incident and fatal CRC across risk scores, and 3) estimate risk-adapted starting ages of screening, defined as when individuals with a given CRC risk score reached the CRC cumulative incidence of the general population at commonly recommended ages to initiate screening. Higher risk scores were associated with increasing risk, with a 2.5-fold increased risk for the highest compared with the lowest score group. Screening colonoscopy was associated with decreased incident and fatal CRC risk across all groups. Notably, absolute risk reductions associated with screening colonoscopy increased with increasing risk score, with absolute CRC incidence reduction being 0.34% for the highest and 0.15% for the lowest risk group. The CRC incidence threshold associated with guideline-recommended screening initiation at age 45 or 50 years was reached at younger ages for the higher risk groups and at an older age for the lower risk groups. For example, the highest and lowest risk groups reached the general population age 45 CRC incidence threshold by ages 38 and 51 years, respectively. The main limitation of the study is the focus on a mostly non-Hispanic White, health professional population, possibly affecting generalizability (3).

This thought-provoking and novel study confirms that screening colonoscopy reduces CRC risk across a range of risk profiles and suggests the benefit of screening colonoscopy may be greater among those with higher CRC risk. The study adds to a large and growing body of work demonstrating demographic, lifestyle, and genetic factors can be used to identify individuals at low vs high risk for CRC (4). Indeed, 1 review of studies from 2000 to 2014 found 32 models having acceptable to good discrimination for CRC risk prediction (4). Further, family history and genomic-based prediction models have demonstrated near sufficient consistency to be used for usual practice risk assessment (5,6). CRC risk prediction models such as the Wang et al. (3) model may be leveraged for at least 3 purposes: identifying the optimal age to initiate screening; screening test selection, with preference towards colonoscopy for those at highest risk; and tailoring risk-reducing interventions such as diet and lifestyle modification. As such, the current challenge is not a lack of good, personalized CRC risk models or a lack of actionable uses for risk stratification, but realizing the promise of these models in practice. Indeed, in the era of big data health care, sophisticated tools for developing risk prediction models such as machine learning, as well as genomic and other ′omic predictors, our ability to predict risk will likely continue to outpace our ability to implement novel models. Unless we address the challenges of implementation, our toolbox of prediction models will continue to go largely unused.

Several approaches may be considered for implementation of prediction models for CRC risk stratification, each with unique challenges (Table 1). At the health system and provider levels, integration into electronic health record software (sometimes referred to as electronic health-care predictive analytics [e-HPA] applications) enables use of the output for counseling and referral (7). Use of e-HPA applications in the outpatient setting can be challenged by feasibility issues, such as limited acceptability from clinicians, inadequate infrastructure to implement and maintain tools, and lack of incentivization for using and interpreting risk prediction models for screening and prevention (8-11). Incentives and reimbursement strategies will likely be required to drive implementation and use of e-HPA applications for predicting and managing CRC risk. At the patient level, an electronic health (defined as use of emerging information and communication technology like the internet to improve health) (12) strategy can be envisioned where online risk calculator use is promoted via online advertisements or other approaches. At issue here is that public health entities (such as the National Cancer Institute) (13) hosting calculators may lack sufficient resources to drive traffic to their websites, and that, to be maximally effective, these calculators need to be able to link individuals “activated” by their risk assessments to health-care providers. Developing partnerships with industry providers of screening tests or health-care providers might help to drive traffic to electronic health resources and allow for a warm handoff back to these providers for individuals ready to follow through on their risk-based recommendations (14). Another patient and population-facing opportunity would be using the growing consumer interest in mobile health (mHealth) technologies, defined as personalized, interactive services that provide access to medical advice and information through a mobile device such as a smartphone (12). An mHealth application assessing cancer risk, including CRC, based on demographics, family history, lifestyle, and perhaps genetic data, could greatly enhance an individual’s insight regarding CRC risk and facilitate personalized screening and preventive interventions. High-risk patients might be empowered to request earlier screening initiation via colonoscopy, and lower risk individuals may be enabled to discuss a later screening initiation age or feel comfortable selecting from a wider array of CRC screening options. Again, public health entities may lack resources to develop and maintain mHealth apps and therefore may need to develop industry partnerships (such as with companies providing direct-to-consumer genetic tests) for mHealth apps to be viable for promoting CRC risk prediction and prevention. These example strategies, and their associated challenges, underscore a need for more investment in a research agenda that not only includes predictive model development, but also studying optimal implementation of these models in usual practice. Notably, to avoid inequities associated with access to technology, the research agenda must include individuals and communities with a diversity of health and technology literacy as well as backgrounds (12).

Table 1.

Challenges and potential solutions for realizing the promise of personalized CRC screening with prediction models in practice^a

Strategy	Application(s)	Challenges	Potential solutions
e-HPA	Integrate prediction model into EHR Leverage prediction model to promote risk-based interventions such as early-age CRC screening, optimal CRC screening test selection, and diet and lifestyle management	Limited time for health-care team to use and act on model High-level informatics support required for implementation and maintenance	Develop incentives and reimbursement models to promote e-HPA use Develop implementation strategies early in lifecycle of predictive model strategies
eHealth	Post online risk calculators and drive traffic to websites through advertisements	Public health entities lack resources to drive traffic, such as through advertisements Linkage of “activated” individuals at website to health-care providers	Promote partnerships between industry providers of screening tests, genetic tests or health systems to support advertisements and links to care
mHealth	mHealth smartphone app for assessing CRC risk based on demographics, family history, lifestyle, and genetic data	Public health entities lack resources to develop and maintain mHealth apps Linkage of activated patients to care	Promote partnerships between industry providers of screening tests, genetic tests, mHealth innovators, or health systems to support mHealth app development, and links to care

^aapp = application; CRC = colorectal cancer; eHealth = electronic health (ie, use of emerging information and communication technology such as the internet to improve health); e-HPA = electronic healthcare predictive analytic applications; EHR = electronic health record; mHealth = mobile health (ie, personalized, interactive services that provide access to medical advice and information through a mobile platform such as a smartphone).

The Wang et al. (3) study provides an excellent example of the power of prediction models for stratifying CRC risk, and an urgent reminder of the work that must be done to develop implementation strategies to realize their promise for ensuring on-time, risk-appropriate CRC screening and prevention in practice.

Funding

This work was supported by grant numbers 1R37CA 222866–01 (Principal Investigator: Gupta), 1UG3CA233314-01A1 and 4UH3CA233314-02 (Martínez, Gupta, Casteñeda, Multiple Principal Investigators), and 5F32CA23960-03 (Principal Investigator: Demb) from the National Cancer Institute/National Institutes of Health.

Notes

Role of the funder: The funders had no role in the writing of this editorial or the decision to submit it for publication.

Disclosures: SG has been a paid consultant for Freenome Holdings, Inc, Guardant Health, Inc, and Cellmax Life. All three of these companies are developing blood-based colorectal cancer screening tests. JD has no disclosures.

Author contributions: Writing, original draft—JD, SG; writing, editing and revisions—JD, SG.

Data Availability

Not applicable.