Prediction Models for Advanced Neoplasia: Risky Business

Abstract

I hold that it is an excellent thing for a physician to practice forecasting.Hippocrates1Jones W.H.S. Potter P. Withington E.T. et al.Hippocrates. Heinemann Putnam, Portsmouth, NH1923Google Scholar Physicians have long valued the ability to predict the future. Hippocrates suggested that physicians be judged, in part, on the basis of this ability. As our clinical armamentarium has grown, some would argue that the physician's ability to “see the future” is even more valuable. Specifically, knowledge of risk for future disease allows us the opportunity to reduce that risk by interrupting its natural course. Colorectal cancer (CRC) is an obvious target for such preventive intervention. The disease is common and potentially deadly; thus, affecting the course of even a small percentage of patients destined for CRC may translate into significant benefit. Furthermore, CRC takes years to develop and allows physicians the opportunity to identify early neoplasia (in the form of polyps) and remove it entirely before it can put a patient at risk of death. Not surprisingly, because of its disease characteristics, clinicians have taken steps to interrupt that natural history of CRC in a number of ways. Recommendations and interventions to screen for the disease are the most obvious evidence of efforts to reduce disease burden.2Levin B. Lieberman D.A. McFarland B. et al.Screening and surveillance for the early detection of colorectal cancer and adenomatous polyps, 2008: a joint guideline from the American Cancer Society, the US Multi-Society Task Force on Colorectal Cancer, and the American College of Radiology.Gastroenterology. 2008; 134: 1570-1595Abstract Full Text Full Text PDF PubMed Scopus (1673) Google Scholar, 3U.S. Preventive Services Task ForceScreening for colorectal cancer: U.S. Preventive Services Task Force recommendation statement.Ann Intern Med. 2008; 149: 627-637Crossref PubMed Scopus (1217) Google Scholar Another approach to limit the impact of CRC has been the development and use of risk prediction models for this cancer. The clinical importance of these predictive models for cancer has been summarized in a recent review.4Win A.K. Macinnis R.J. Hopper J.L. et al.Risk prediction models for colorectal cancer: a review.Cancer Epidemiol Biomarkers Prev. 2012; 21: 398-410Crossref PubMed Scopus (72) Google Scholar The currently used models are largely based on a panel of recognized risk factors for CRC including obesity and dietary and drug exposure such as aspirin and hormonal replacement therapy. Use of a cancer prediction model can reduce disease burden by identifying those most likely to develop CRC and encouraging individuals to modify their behavior to decrease risk. In addition, knowledge of their risk for CRC (particularly if it is interpreted as high) might encourage individuals to get screened for the disease. For those identified with a strong family history as a key component of their predictive model, genetic testing might be in order. To date, at least 4 models predicting cancer have been evaluated,5Colditz G.A. Atwood K.A. Emmons K. et al.Harvard report on cancer prevention volume 4: Harvard Cancer Risk Index—Risk Index Working Group, Harvard Center for Cancer Prevention.Cancer Causes Control. 2000; 11: 477-488Crossref PubMed Scopus (255) Google Scholar, 6Freedman A.N. Slattery M.L. Ballard-Barbash R. et al.Colorectal cancer risk prediction tool for white men and women without known susceptibility.J Clin Oncol. 2009; 27: 686-693Crossref PubMed Scopus (171) Google Scholar, 7Imperiale T.F. Wagner D.R. Lin C.Y. et al.Using risk for advanced proximal colonic neoplasia to tailor endoscopic screening for colorectal cancer.Ann Intern Med. 2003; 139: 959-965Crossref PubMed Scopus (149) Google Scholar, 8Ma E. Sasazuki S. Iwasaki M. et al.10-Year risk of colorectal cancer: development and validation of a prediction model in middle-aged Japanese men.Cancer Epidemiol. 2010; 34: 534-541Crossref PubMed Scopus (46) Google Scholar and generally speaking, they have shown reasonable ability to discriminate between those likely and not likely to develop CRC. However, in truth, the vast majority of screening (at least in the United States) is not driven by stratification on such tools but rather the simple fact that an individual has reached age 50 and is “due” for their colonoscopy. More recently, a new line of predictive models has been developed. The models do not predict subsequent risk for cancer but whether an individual is likely to harbor advanced neoplasia, which is defined as large polyps (≥1 cm) or those with advanced histology. In this issue, Lee et al9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar report on a large (n = 3144) cross-sectional evaluation they performed to determine whether those at high risk for coronary heart disease measured by the Framingham Risk Score were at high risk for advanced neoplasia. Participants were part of a health checkup program based in Seoul, Korea, and they underwent evaluation including a telephone interview, physical examination, and laboratory assays as well as a colonoscopy. When comparing those at highest risk for coronary heart disease (≥20%) with those at lowest risk for that outcome (<10), they found that the high risk group was also significantly more likely to have advanced neoplasia on colonoscopy (adjusted odds ratio, 3.31; 95% confidence interval, 1.94–5.65). In absolute terms, 18% had advanced neoplasia in the highest risk group for coronary heart disease, and only 5% had advanced neoplasia in the lowest coronary heart disease risk group. The article is important, in part, because it contributes to the literature confirming an association between coronary heart disease and colorectal neoplasia. In some ways, the association is not surprising because of the shared risk factors operating in similar directions.10Neugut A.I. Rosenberg D.J. Ahsan H. et al.Association between coronary heart disease and cancers of the breast, prostate, and colon.Cancer Epidemiol Biomarkers Prev. 1998; 7: 869-873PubMed Google Scholar Factors associated with increased risk for both diseases include obesity and cigarette smoking, whereas factors such as physical activity and aspirin intake work in the opposite direction associated with reduced risk. Prior work has shown fairly convincingly that the presence of coronary heart disease as measured by angioplasty was associated with the presence of colorectal neoplasia.11Chan A.O. Jim M.H. Lam K.F. et al.Prevalence of colorectal neoplasm among patients with newly diagnosed coronary artery disease.JAMA. 2007; 298: 1412-1419Crossref PubMed Scopus (114) Google Scholar Lee et al9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar have extended that by showing that even at an earlier time point (ie, before overt heart disease is present), the diseases appear related. At the very least, clinicians can advise their patients that a “heart healthy” lifestyle likely has benefit outside the myocardium in the walls of the colon. The article also adds to the literature of risk prediction models for advanced colorectal neoplasia. In the last 2 years, 3 other groups have reported on predictive models for advanced neoplasia.12Cai Q.C. Yu E.D. Xiao Y. et al.Derivation and validation of a prediction rule for estimating advanced colorectal neoplasm risk in average-risk Chinese.Am J Epidemiol. 2012; 175: 584-593Crossref PubMed Scopus (72) Google Scholar, 13Schroy 3rd, P.C. Coe A.M. Mylvaganam S.R. et al.The Your Disease Risk Index for colorectal cancer is an inaccurate risk stratification tool for advanced colorectal neoplasia at screening colonoscopy.Cancer Prev Res. 2012; 5: 1044-1052Crossref PubMed Scopus (16) Google Scholar, 14Yeoh K.G. Ho K.Y. Chiu H.M. et al.The Asia-Pacific Colorectal Screening score: a validated tool that stratifies risk for colorectal advanced neoplasia in asymptomatic Asian subjects.Gut. 2011; 60: 1236-1241Crossref PubMed Scopus (199) Google Scholar The 2 groups that had success with their prediction models12Cai Q.C. Yu E.D. Xiao Y. et al.Derivation and validation of a prediction rule for estimating advanced colorectal neoplasm risk in average-risk Chinese.Am J Epidemiol. 2012; 175: 584-593Crossref PubMed Scopus (72) Google Scholar, 14Yeoh K.G. Ho K.Y. Chiu H.M. et al.The Asia-Pacific Colorectal Screening score: a validated tool that stratifies risk for colorectal advanced neoplasia in asymptomatic Asian subjects.Gut. 2011; 60: 1236-1241Crossref PubMed Scopus (199) Google Scholar did not use a previously developed tool like the Framingham Risk Index. Instead, they used regression modeling to identify a set of factors that appeared to predict risk of advanced neoplasia and then validated them in a separate data set. This approach is more inherently attractive because it allows the data on the outcome of interest (in this case advanced neoplasia) to identify the variables that are most important. For example, the model by Lee et al9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar does not consider family history of CRC because the Framingham index logically does not include that as a predictor of coronary heart disease. In fact, in the article by Lee et al, those with a family history of CRC were excluded from the study. If they had included such patients, it is likely that the model would not have worked as effectively as it did. Because they excluded those with family history of CRC in their analysis, the tool could not be used to predict advanced neoplasia prevalence in those populations. Therefore, the article by Lee et al9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar joins a group of studies showing some success in predicting the presence of advanced neoplasia among average-risk adults eligible for CRC screening. If I had to make a prediction, I would guess that over time, these models will become even more refined and effective at predicting this measure. For example, the addition of biologic information on risk (eg, genes) to these epidemiologic risk factors will further enhance accuracy. Although I have little doubt that we will be able to predict advanced neoplasia fairly effectively, I have concerns that information may not be as helpful as it seems. My concerns center largely on the value of predicting a surrogate outcome like advanced neoplasia. The goal of CRC screening is to prevent death from this disease, with the extension of the overall life span. Although the exact time for a lesion to progress from advanced neoplasia to cancer is unknown, it likely takes years. In the seminal work by Stryker et al15Stryker S.J. Wolff B.G. Culp C.E. et al.Natural history of untreated colonic polyps.Gastroenterology. 1987; 93: 1009-1013Crossref PubMed Scopus (782) Google Scholar that reported the natural history of large polyps (≥1 cm) followed by barium enema, the cumulative risk of a cancer diagnosis at 5 years was 2.5%. I would suggest that before we consider applying these models clinically, a number of important questions need to be answered. First, what risk of advanced neoplasia should prompt some type of action? In the article by Lee et al,9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar the authors suggested that those with a cardiac risk (on the Framingham measure) of 10% or more might be targeted for intervention. The absolute risk of advanced neoplasia in that group ranged from 9% to 18%. Does it make sense to stratify screening differently in those with a 7% risk for advanced neoplasia vs 10%? A second related but equally important question that needs answering is who interprets the significance of the risk. Is the goal to provide this type of information to the patient so they can decide whether it meets their action threshold, or will clinicians be applying the tool and directing care from there? If the information is being funneled through the patient, then significant work will be needed to help them understand the meaning of advanced neoplasia and its implications. Of course, the most important question is exactly what we will do differently when we know the absolute risk of advanced neoplasia before looking for it. The obvious answer is that we will triage “high risk” individuals to the gold standard test (ie, colonoscopy) and divert the low risk population to stool testing. In fact, the article by Lee et al9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar concludes that might be the next step if their work is confirmed by others. Does this approach really make good sense? Yes, the one-time application of colonoscopy is more sensitive for the detection of advanced neoplasia relative to the one-time application of a stool test such as the fecal immunochemical test.16Quintero E. Castells A. Bujanda L. et al.Colonoscopy versus fecal immunochemical testing in colorectal-cancer screening.N Engl J Med. 2012; 366: 697-706Crossref PubMed Scopus (631) Google Scholar However, colonoscopy is certainly not perfectly sensitive for the detection of cancer, never mind advanced neoplasia, and if colonoscopy does not detect the lesion, follow-up may range from 3 to 10 years on the basis of current guidelines. Because it likely takes years to progress from advanced neoplasia to cancer, there may be little harm to serially applying a fecal immunochemical test that may be initially negative but subsequently positive when the lesion is larger (ie, less likely to be missed by colonoscopy) but likely still in a treatable phase (ie, screen detected). Some would argue that there is simply no downside in applying the tool and pursing colonoscopy earlier in the higher risk group. However, we do not know this. Assuming that clinicians take the role of applying the tool and directing care, many individuals will be told that they are at “low risk” (eg, “only a risk of advanced neoplasia of 7%”). Will compliance with any screening fall when an individual is informed that they are in a lower risk group and do not need colonoscopy? Determining the answers to these questions are the logical next steps in efforts to make the use of these predictive models a reality. Thus, putting aside our long-ingrained tradition valuing prediction that dates back to Hippocrates, I would argue that at least for the present, we should stick with what we know. We tell patients that CRC is common and lethal and that screening has been shown to significantly reduce death from this disease. Options for screening include a panel of tests including endoscopy and stool testing, and the individual should select the test that they are most likely to comply with. At the present time, using prediction models outside of research studies cannot be recommended. Through further research with these tools, including the one developed by Lee et al,9Lee J.Y. Hong S.N. Kim J.H. et al.Risk for coronary heart disease increases risk for colorectal neoplasm.Clin Gastroenterol Hepatol. 2013; 11: 695-702Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar we can answer important questions about their application including their actual impact on screening compliance and outcomes that matter. In the absence of those types of studies, the value of triaging individuals on surrogate markers such as advanced neoplasia is at best unpredictable. Risk for Coronary Heart Disease Increases Risk for Colorectal NeoplasmClinical Gastroenterology and HepatologyVol. 11Issue 6PreviewColorectal neoplasms and coronary artery disease have similar risk factors. Patients with established coronary artery disease have a high prevalence of colorectal neoplasms. However, little is known about the risk of colorectal neoplasms among individuals at risk for coronary artery disease. Full-Text PDF