Personalizing Colorectal Cancer Screening: A Systematic Review of Models to Predict Risk of Colorectal Neoplasia

Clinical Gastroenterology and Hepatology 2014;-:-–-

Personalizing Colorectal Cancer Screening: A Systematic Review of Models to Predict Risk of Colorectal Neoplasia Gene K. Ma* and Uri Ladabaum*,‡ ‡

Division of Gastroenterology and Hepatology, *Department of Medicine, Stanford University School of Medicine, Stanford, California BACKGROUND & AIMS:

A valid risk prediction model for colorectal neoplasia would allow patients to be screened for colorectal cancer (CRC) on the basis of risk. We performed a systematic review of studies reporting risk prediction models for colorectal neoplasia.

METHODS:

We conducted a systematic search of MEDLINE, Scopus, and Cochrane Library databases from January 1990 through March 2013 and of references in identified studies. Case-control, cohort, and cross-sectional studies that developed or attempted to validate a model to predict risk of colorectal neoplasia were included. Two reviewers independently extracted data and assessed model quality. Model quality was considered to be good for studies that included external validation, fair for studies that included internal validation, and poor for studies with neither.

RESULTS:

Nine studies developed a new prediction model, and 2 tested existing models. The models varied with regard to population, predictors, risk tiers, outcomes (CRC or advanced neoplasia), and range of predicted risk. Several included age, sex, smoking, a measure of obesity, and/or family history of CRC among the predictors. Quality was good for 6 models, fair for 2 models, and poor for 1 model. The tier with the largest population fraction (low, intermediate, or high risk) depended on the model. For most models that defined risk tiers, the risk difference between the highest and lowest tier ranged from 2-fold to 4-fold. Two models reached the 0.70 threshold for the C statistic, typically considered to indicate good discriminatory power.

CONCLUSIONS:

Most current colorectal neoplasia risk prediction models have relatively weak discriminatory power and have not demonstrated generalizability. It remains to be determined how risk prediction models could inform CRC screening strategies.

Keywords: Colon Cancer Screening; Risk Stratification; Systematic Review; Early Detection.

olorectal cancer (CRC) is the second leading cause of cancer-related death in the United States.1 CRC screening decreases CRC incidence and mortality through prevention and early detection of CRC.2–6 In addition to the demonstrated benefit of screening on CRC-related mortality, multiple health economic models suggest that screening is cost-effective.7–11 Currently, an undifferentiated screening approach to all “average risk” persons starting at age 50 years is recommended by the U.S. Preventive Services Task Force and most major professional societies.12–14 Despite the proven benefit, a large fraction of eligible persons do not participate in CRC screening programs.15 Persons at higher risk of CRC are most likely to adhere to CRC screening.16–18 Multiple demographic and clinical risk factors for CRC have been identified, such as age, gender, smoking, diet, obesity, physical activity, ethnicity, and family history of CRC.19–25 It is conceivable that depending on the presence or absence of risk factors, the population

C

considered at average risk could be stratified across a spectrum of predicted risk. Several models that use demographic and clinical risk factors have been developed to predict risk of colorectal neoplasia.26–34 A colorectal neoplasia risk prediction model with acceptable discriminatory power could be used to tailor the CRC screening schedules of persons currently considered at average risk. This could improve clinical outcomes and decrease resource utilization and costs. Our aim was to review existing risk prediction models for colorectal neoplasia. We performed a systematic review of the literature with attention to study quality,

Abbreviations used in this paper: CI, confidence interval; CRC, colorectal cancer. © 2014 by the AGA Institute 1542-3565/$36.00 http://dx.doi.org/10.1016/j.cgh.2014.01.042

2

Ma and Ladabaum

population, model characteristics, and the risk stratification tools’ discriminatory power.

Methods A systematic review adhering to the recommendations of the PRISMA statement35 was performed to identify studies reporting the development or attempted validation of a tool or model to predict the risk of colorectal neoplasia on the basis of demographic or clinical risk factors.

Data Source and Searches Two investigators independently conducted targeted literature searches by using MEDLINE, Scopus (which searches Embase but does not make use of Emtree), and the Cochrane Library Database from January 1990 to March 2013. The search strategies were developed with the help of technical experts from Stanford University School of Medicine Lane Library. The search strategy for MEDLINE was the following: (Colorectal neoplasm* OR colorectal cancer OR colonic neoplasm* OR colon cancer OR rectal neoplasm* OR rectal cancer OR anus cancer OR anus neoplasm*) AND (personal OR individual) AND risk AND (score* OR scoring OR index OR stratif* OR measur* OR tool* OR index OR status) AND screen*. The search strategy for Scopus was the following: TITLE-ABSKEY(Colorectal neoplasm* OR colorectal cancer OR colonic neoplasm* OR colon cancer OR rectal neoplasm* OR rectal cancer OR anus cancer OR anus neoplasm*) AND (personal OR individual) AND risk AND (score* OR scoring OR index OR stratif* OR measur* OR tool* OR index OR status) AND screen*. The Cochrane Library Database is organized by topic, and all articles listed under the topic “Colorectal Cancer” were reviewed by the authors. The reference lists of studies selected for review were searched manually to identify additional potentially relevant studies.

Study Selection The 2 investigators independently reviewed the titles and, when indicated, abstracts of identified articles against a priori inclusion and exclusion criteria. The inclusion criteria were case-control, cohort, and cross-sectional studies that developed or tested a model to predict risk of colorectal neoplasia (cancer or advanced neoplasia) that was based on demographic or clinical risk factors in a general population considered at average risk. We restricted inclusion to English-language articles. Exclusion criteria included studies published as abstracts only, because full assessment is not possible in such cases, and studies reporting on risk stratification or risk prediction in the absence of a discrete risk score model. Full-text review by both investigators was performed for all articles that were selected for review by at least 1 investigator.

Clinical Gastroenterology and Hepatology Vol.

-,

No.

-

Data Extraction and Quality Assessment The 2 investigators independently abstracted prespecified elements into standardized data extraction tables (Supplementary Figure 1) documenting study design and methods, patient population, and characteristics of the risk model created. Discrepancies were resolved through discussion between the 2 investigators. Additional discussion with a third party was planned to resolve discrepancies if necessary, but this was not needed. Although universally accepted criteria for study quality in the development of risk prediction models are not available, the investigators independently applied predefined criteria to assess the model quality as good, fair, or poor on the basis of previously proposed principles.36 Quality in studies reporting risk scores with external validation in an independent validation cohort was considered good. Quality in studies reporting risk scores with internal validation such as bootstrapping, crossvalidation, or random split-sample methods was considered fair. Quality in studies without internal or external validation was considered poor. As with the data extraction items, discrepancies were resolved through discussion between the 2 investigators, and additional discussion with a third party was not needed.

Data Synthesis and Analysis A quantitative summary of the studies in the form of a meta-analysis was planned in the event that the data allowed this. The identified risk scores were too heterogeneous to allow a meaningful quantitative summary. Therefore, we used a qualitative, narrative synthesis method.35 For the narrative synthesis, we focused on outcome predicted, risk factors included, tiers of risk, sizes of subpopulations within tiers, and range or predicted risk.

Results Risk Prediction Models Of the 4420 articles identified through literature searches and review of references cited in the selected studies, 17 met inclusion criteria after title and abstract review (Figure 1). After examination of the full texts, 6 studies were excluded because a risk model was not developed or tested (n ¼ 2),37,38 the focus was not on CRC screening in the general population (n ¼ 2),39,40 or there was insufficient detail to evaluate the proposed model (n ¼ 2).41,42 Of the 11 studies that were ultimately included in the systematic review, 9 developed a new risk assessment model for either CRC or advanced neoplasia (which includes CRC or advanced adenomas [large adenoma and adenoma with villous features or

-

2014

Colorectal Neoplasia Risk Prediction Models

3

Outcomes Predicted, Risk Factors, and Risk Tiers

Figure 1. Flow diagram summarizing study identification and selection.

high-grade dysplasia]),26–34 and 2 reported testing of previously developed models (Table 1).43,44 Most models addressed the entire colorectum; one focused on proximal advanced neoplasia.26

Model Development, Validation, and Quality The statistical methods used to develop the 9 risk assessment models were logistic regression in six,26–28,30,33,34 Cox regression in one,32 Poisson regression in one,31 and recursive partitioning analysis in one.29 Two models were both internally validated and externally validated,26,34 four were externally validated with independent validation cohorts,29,30,32,33 one was internally validated by bootstrap resampling,28 one used an internal validation cohort,27 and one was not validated.31 On the basis of the quality criteria reflecting validation methods, 6 models were considered of good quality, 2 models were considered of fair quality, and 1 model was considered of poor quality (Table 1).

The outcome of interest was CRC specifically in 5 models and advanced neoplasia in 4 models (Table 2). The model of Imperiale et al26 was designed to predict the risk of proximal advanced neoplasia specifically, including as a predictor the findings at sigmoidoscopy. The risk prediction models differed in terms of risk factors included and how risk was assessed (Table 2). The number of risk factors included in each risk model ranged from a low of three26 to a high of eleven.30 The risk factors most frequently represented were age and smoking, which were each included in 6 of 9 models.28,30–34 Several models included age, gender, smoking, a measure of obesity, and/or family history of CRC among the predictors (Table 2). Six of the 9 risk models initially evaluated more risk factors than were ultimately included.28,30–34 Four models defined 3 tiers of risk,26,28,31,33 two models defined 2 tiers of risk,27,34 and the remaining 3 did not define any tiers but instead predicted risk in a continuum or focused on identifying high-risk persons.29,30,32 For most models that defined tiers, the difference in risk for the highest vs lowest risk tier ranged from 2-fold to 4-fold (Table 2).

Relative Sizes of Subpopulations Within Tiers of Predicted Risk The proportions of subjects who were included in the different tiers of predicted risk varied by model (Table 2). For instance, the model of Cai et al34 divided the population into 2 tiers, with nearly 50% of subjects in each. The model of Yeoh et al33 stratified the population into a larger middle tier of moderate risk (51.1%), a smaller average-risk tier (29.5%), and an even smaller high-risk tier (19.4%). In contrast, the population distribution was skewed in the model of Wei et al31 toward the high-risk tier (61.1%) and in the model by Driver et al28 toward the lowest-risk tier (53.1%).

Discriminatory Power Study Populations The models varied with respect to target population. Although the majority of studies used data from populations with approximately the same representation from each gender, 2 studies used data exclusively on men,28,32 and 1 study used data exclusively on women.31 Six studies were performed in populations in the United States, 1 in a population in China,34 one in a population in Japan,32 and one in populations in 11 Asian countries (Table 1).33 Four studies included individuals undergoing screening colonoscopy,26,27,33,34 and 1 study included individuals undergoing screening, diagnostic, or surveillance colonoscopy or flexible sigmoidoscopy.29

The area under the receiver operating characteristic curve or C statistic did not reach the 0.70–0.80 range that is typically considered to reflect good discriminatory power except for the risk models developed by Cai et al34 and Imperiale et al,26 which reported an area under the curve of 0.74 (95% confidence interval [CI], 0.70–0.78) and a C statistic of 0.74 (95% CI, 0.68–0.80), respectively. The risk index developed by Imperiale et al was created by using data from a predominantly white population. The study by Levitzky et al44 used an independent validation cohort to test the risk index of Imperiale et al in a population with white, black, and Hispanic patients. In that study, the C statistic values were 0.62 (95% CI,

4

Table 1. Studies of Colorectal Neoplasia Risk Prediction Models

Country

Study period

Study type

Centers

Inclusion criteria

Patient characteristics

Outcomeb

Statistical method

China

7/2006–12/2008

Prospective crosssectional

CRC, adenoma Multicenter (19) 40-year-old Mean age: 55 y; 10 mm, asymptomatic persons; female: 49.0%; villous screening colonoscopy derivation: n ¼ 5229; adenoma, or validation: high-grade n ¼ 2312 dysplasia

Logistic regression

Yeoh et al, 201133

11 Asian 7/2004–12/2004 countries

Prospective crosssectional

Logistic regression

Ma et al, 201032

Japan

1990–1993

Prospective crosssectional

Wei et al, 200931

United States

1980–2004

Prospective crosssectional

Freedman et al, 200930

United States

10/1991–9/1994, Case-control 5/1997–5/2001

Multicenter (17) Asymptomatic persons; Mean age: 51 y; screening colonoscopy female: 46%; derivation: n ¼ 860; validation: n ¼ 1892 Multicenter (11) 40-year-old persons; Mean age: 53 y; participant of Japan female: 0%; Public Health Center derivation: Study Cohort I or n ¼ 28,115; Cohort II validation: n ¼ 18,256 National 30-year-old women; Mean age: N/A; (11 states) participant of Nurses’ female: 100%; Health Study derivation: n ¼ 83,767 National 50-year-old nonAge: >50 y; (3 states) Hispanic white female: 49.0%; persons n ¼ 5097

Park et al, 200943

United States

1995–1996

Kastrinos et al, 200929

United States

Derivation: Prospective 2/2004–6/ cross2004; sectional validation: 5/2007–9/2007

Prospective cohort

National (6 states)

Single

CRC, adenoma 10 mm, villous adenoma, or high-grade dysplasia CRC

Colon cancer

Qualityc

Internal validation Good by bootstrap resampling and external validation with independent validation cohort Good Independent validation cohort

Cox regression

Independent validation cohort

Good

Poisson regression

None

Poor

Proximal colon, Logistic External validation Good distal colon, regression in companion and rectal study by Park cancers et al43 50- to 71-year-old Median age: 63 y; CRC N/A Validation of Risk N/A persons; participant of female: 61%; Index of National Institutes of n ¼ 263,402 Freedman Health-American et al30 Association of Retired Persons study 18-year-old persons; Median age: 56 y; Persons Recursive Independent Good screening, diagnostic, female: 54.8%; considered partitioning validation derivation: high risk for analysis cohort or surveillance colonoscopy or flexible n ¼ 615; hereditary CRC sigmoidoscopy validation: syndrome n ¼ 5335

Clinical Gastroenterology and Hepatology Vol.

Cai et al, 201234

Validation

Ma and Ladabaum

Study, yeara

-,

No. -

1982–2004

Prospective cohort

National

Lin et al, 200627

United States

10/2001–10/2004

Retrospective Single crosssectional

Imperiale et al, 200326

United States

Derivation: Retrospective Multicenter (8) 9/1995– cross12/1998; sectional validation: 1/1999–6/2001

Levitzky et al, 201144

United States

1/2000–12/2005

Retrospective Single crosssectional

40-year-old persons; no Age range: 40–84 y; female: 0%; history of n ¼ 21,851 cardiovascular disease, cancer, or other serious illness; participant of Physician’s Health Study 50-year-old persons; Mean age: 60 y; screening colonoscopy female: 50.8%; derivation: n ¼ 1512; validation: n ¼ 1493 50-year-old Mean age: 57 y; asymptomatic persons; female: 43%; screening colonoscopy derivation: 1994; validation: 1031 50-year-old Mean age: N/A; asymptomatic persons; female: 53.2%; screening colonoscopy derivation: n ¼ 3499

CRC

Logistic regression

Internal validation by bootstrap resampling

Fair

CRC, adenoma 10 mm, 25% villous, or high-grade dysplasia

Based on Internal validation Risk Index cohort of Imperiale et al26

Fair

CRC, adenoma 10 mm, villous adenoma, or high-grade dysplasia CRC, adenoma 10 mm, villous adenoma, or high-grade dysplasia

Logistic regression

Internal validation Good cohort and external validation by Levitzky et al44

N/A

Validation of Risk Index of Imperiale et al26

2014

United States

-

Driver et al, 200728

N/A

N/A, not applicable. a Studies are listed in reverse chronological order, except for 2 studies (Park et al and Levitzky et al) that tested an existing risk prediction model (these are listed below the initial publication of the risk prediction model). b Most studies included CRC as an outcome; one study (Wei et al) included colon cancer only. c Quality rating: Good ¼ externally validated with an independent validation cohort; Fair ¼ internally validated with bootstrapping, cross-validation, or random split-sample methods; Poor ¼ no internal or external validation.

Colorectal Neoplasia Risk Prediction Models 5

6

Table 2. Risk Stratification and Statistical Characteristics of Colorectal Neoplasia Risk Prediction Models

Study, year Cai et al, 201234

Yeoh et al, 201133

Risk factors included

Risk factors evaluated but not included

Age, gender, smoking, Hypertension, diabetes mellitus, green triglycerides, coronary vegetables, pickled artery disease, calcium, food, fried food, white vitamin D, nonsteroidal meat anti-inflammatory drugs, fresh fruits, milk, egg, red meat, tea, coffee, alcohol Gender, age, FDR with Alcohol, diabetes mellitus CRC, smoking

Ma et al, 201032

Age, body mass index, smoking, alcohol

Physical activity, dietary factors, diabetes mellitus, family history

Wei et al, 200931

Smoking, body mass index, physical activity, red or processed meat, folate

Family history, aspirin, postmenopausal hormone use, height

Kastrinos et al, 200929

Subjects (%)

AN/CRC prevalence

0–3 >3

49.2 50.9

AN: 2.6% AN: 10.0%

Average Moderate

0–1 2–3

29.5 51.1

AN: 1.3% AN: 3.2%

High

4–7

19.4

AN: 5.2%

Risk tiers not 1–3 defined, 4–7 arbitrary tiers 8–10 created Low — Moderate —

— — — 16.0 22.9

High

—

61.1

Risk tiers not defined

—

—

Risk tiers not defined

—

—

CRC/10 y: 0.2%–0.9% CRC/10 y: 1.3%–3.3% CRC/10 y: 4.6%–7.4%

RR/OR (95% CI)a — —

— RR: 2.6 (1.1–6.0) RR: 4.3 (1.8–10.3) — — —

Discriminatory powerb AUC: 0.74 (0.70–0.78)

Other features Sensitivity, 80.3%; specificity, 51.2%; NNS, 10

C statistic: 0.64 (0.60–0.68)

C statistic: 0.69 (0.67–0.71)

— AUC: 0.61 (0.59–0.63) RR: 1.44 (1.05–1.96) N/Ac RR: 3.84 (1.61–9.16) From Freedman: CRC/10 y — From Park: AUC: (men), 0.16–9.95; CRC/ 0.61 (0.60–0.62) 10 y (women), 0.09– men; 0.61 8.16; CRC/20 y (men), (0.59–0.62) 0.53–19.4; CRC/20 y women (women), 0.31–17.49

N/Ac N/Ac

77% of high-risk patients detected with 3 questions

—

—

-,

No.

Vitamins, alcohol, red Sigmoidoscopy results, meat, fruit colonoscopy results, history of polyps, relative with CRC, aspirin/nonsteroidal anti-inflammatory drug use, smoking, vegetables, body mass index, leisure time activity (men only), leisure exercise time (women only), estrogen status (women only) FDR with CRC or Lynch None syndrome related cancer, CRC or polyps diagnosed before age 50, more than 3 relatives with CRC

Low High

Score

Clinical Gastroenterology and Hepatology Vol.

Freedman et al30 and Park et al, 200943

Tier

Ma and Ladabaum

Risk stratification

-

Age, gender, distal findings As above on sigmoidoscopy Levitzky et al, 201144; validation of Imperiale et al26

Imperiale et al, 200326

AN, advanced neoplasia; AUC, area under the receiver operating characteristic curve; FDR, first-degree relative; N/A, not applicable; NNS, number needed to screen; OR, odds ratio; RR, relative risk; SDR, second-degree relative. a RR and OR refer to the higher-level tiers vs the comparator, lowest-level tier. b AUC or C statistic reported is for the validation cohort if available. c Prevalence was not reported. Outcome was cumulative CRC risk up to age 70 years.

—

—

Proximal AN: 0.4% Proximal AN: 1.9% Proximal AN: 3.8% AN (highest vs lowest risk): white, 3.7% vs 1.0%; black, 4.2% vs 1.0%; Hispanic, 3.7% vs 0.6% 46.8 40.3 12.8 0–1 2–3 4–7 Low Intermediate High As above

AN: 3.0% AN: 5.7% 40.2 59.8 0–3 4–6 Low High

Gender, age, FDR with None adenoma or CRC, SDR with CRC Age, gender, distal findings None on sigmoidoscopy Lin et al, 200627

CRC/20 y: 5.6% 9.9 7–10

CRC/20 y: 1.0% CRC/20 y: 3.1% 53.1 36.9 0–3 4–6

Driver et al, 200728 Age, smoking, body mass Vegetables, vitamins, Lowest index, alcohol use cereal, physical activity, Intermediate diabetes mellitus Highest

C statistic: white, NNS: white, 27; 0.62 (0.54–0.70); black, 30; black, 0.63 Hispanic, 34 (0.54–0.73); Hispanic, 0.68 (0.53–0.82)

Colorectal Neoplasia Risk Prediction Models

C statistic: 0.74 (0.68–0.80)

2014

— AUC: 0.695 OR: 3.07 (2.46–3.83) OR: 5.75 (4.44–7.44) — —

-

7

0.54–0.70) for whites, 0.63 (95% CI, 0.54–0.73) for blacks, and 0.68 (95% CI, 0.53–0.82) for Hispanics.

Discussion This systematic review summarizes the available colorectal neoplasia risk prediction models and highlights their heterogeneity. Many models still require external validation. The majority of models were developed by using data from primarily white populations, and validation in more diverse populations would be necessary to demonstrate broader generalizability. Statistically validated models must also prove clinically valuable.45 The discriminatory power of most currently available models is relatively weak, which raises questions about their potential clinical impact. More univariate associations with colorectal neoplasia have been reported than the number of risk factors included in each of the current risk models.19–25 For some models, certain risk factors were not considered a priori; for others, certain potential predictors were not ultimately retained.28,30–34 No clear patterns emerged regarding risk factors that were evaluated but ultimately excluded. For instance, Ma et al32 and Wei et al31 evaluated but did not retain family history in the final models. In the models that included family history, the definition of this complex variable was not uniform (Table 2).27,29–33 Gender is one of the most studied colorectal neoplasia risk factors. Some have argued that women should begin CRC screening at a later age than men on the basis of studies demonstrating gender as an independent predictor of advanced neoplasia prevalence.46–48 The 4 indices that included gender as a risk factor did not weigh gender more heavily than other risk factors. For example, the index of Imperiale et al26 assigned 1 point for male gender out of a total of 7 possible points. Cai et al34 noted that when not stratified by gender, their risk index retained the same discriminatory power. In contrast with lung cancer, for which a single dominant risk factor (smoking) has been identified,49 there is no single dominant predictor of CRC risk. In the absence of a single dominant preventive measure such as smoking avoidance, screening is likely to remain the most powerful public health measure to decrease CRC incidence and mortality. Although higher intensity screening is recommended for persons with a family history of CRC, the majority of the population is currently considered to be at average risk and is eligible for the same screening protocols.14 A risk prediction model could allow for risk stratification within this average-risk group; such risk stratification might improve the already notable mortality benefits of CRC screening by increasing screening intensity in those at higher risk (thereby increasing CRC prevention and early detection) and decreasing screening in those at lower risk (thereby decreasing the burdens and risks of complications

8

Ma and Ladabaum

associated with screening). Better use of screening resources might decrease costs. Despite the known mortality benefit of CRC screening, approximately one-third of average-risk adults in the United States are not up-to-date with CRC screening.15 The approaches for improving screening uptake and adherence differ in settings that rely on opportunistic screening vs a population-based screening program. Strategies designed to increase screening adherence have achieved varying levels of success.50 Physician recommendation remains one of the strongest predictors of CRC screening adherence.16,51–53 Some studies suggest that knowledge of CRC screening guidelines is suboptimal for both family physicians and, to a lesser extent, gastroenterologists.54 It is conceivable that a reliable risk score could facilitate recommendations for screening. Studies suggest that high-risk individuals are more likely to be up-to-date with CRC screening and adhere to physician recommendations.16,55–57 Thus, identifying a high-risk group could optimize both outcomes and allocation of screening resources. However, in 1 previous study, patients falling in an intermediate-risk group had the lowest rates of CRC screening.16 The impact of any risk stratification strategy will be influenced by how it affects screening uptake by persons in the various risk tiers. Risk stratification might improve the costeffectiveness of CRC screening. Multiple economic analyses support the cost-effectiveness of average-risk CRC screening.7–11,58 One model suggested that screening might be more cost-effective in black persons starting at younger ages than in other groups starting at older ages.59 Although current guidelines do not address quantitative levels of risk, the American College of Gastroenterology recommends screening in blacks beginning at the age of 45 rather than 50 years.60 One modeling study predicted only modest improvements in effectiveness and cost when individualizing CRC screening by sex and race,61 but greater risk discrimination might have substantial clinical and economic impacts. The costs of chemotherapy have been rising,62–64 which has improved the costeffectiveness of prevention.65,66 The impact of risk stratification on the use of limited resources, such as colonoscopy,67,68 is likely to depend on the relative sizes of the subpopulations identified as low risk vs high risk. Several barriers exist to the incorporation of a colorectal neoplasia risk prediction model into practice.69 Most models have not been validated in diverse populations, and most have relatively weak discriminatory power. Risk prediction models for breast cancer,70 prostate cancer,71 and lung cancer72 have faced similar problems. The risk models included in this study used clinical risk factors, but it is conceivable that risk prediction might improve with the addition of genetic and other biomarkers of risk. Biomarkers that meet the requirements for clinical use remain to be discovered or validated. Because of the technical advances and decreasing costs of genomic and other laboratory

Clinical Gastroenterology and Hepatology Vol.

-,

No.

-

methods, inclusion of such predictors might be economically and practically feasible in the future.73–75 However, to date, the addition of genetic risk factors to risk prediction models for other diseases has yielded only modest gains in discriminatory power.76–81 Ultimately, sufficient discriminatory power and generalizability, which are demonstrated by rigorous external validation, will be necessary for a model to be integrated successfully into clinical practice.69 A practical challenge for implementation of a risk score is the collection of the necessary inputs. Application of a risk score would be aided by routine inclusion of the necessary predictors in an electronic medical record. Our review has limitations. We could not formally assess publication bias. Non-English language studies were excluded, which could lead to language bias. However, no non-English language studies met inclusion criteria. The methodological heterogeneity of the identified studies did not allow for quantitative analysis. We were unable to evaluate in detail 2 risk scores that have been described in abstract form only recently.82,83 One tool derived in a population undergoing screening colonoscopy in the United States considered age, sex, waist circumference, cigarette smoking, and family history of CRC and was used to construct 4 tiers of risk; in the validation set, the prevalence of advanced neoplasia was 1.65% in the very low–risk tier, 3.3% in the low-risk tier, 10.9% in the intermediate-risk tier, and 22% in the highrisk tier.82 A second tool derived in a population undergoing screening colonoscopy in Poland considered age, sex, body mass index, cigarette smoking, and family history of CRC and yielded a risk score of advanced neoplasia ranging from 0 to 8, which corresponded to risk of advanced neoplasia ranging from 2.5% to 19.4%.83 In this systematic review, we identified 9 colorectal neoplasia risk prediction models. The models demonstrate substantial heterogeneity with regard to study population, risk factors included, and outcomes predicted. Several models include age, gender, smoking, some measure of obesity, and/or family history of CRC among the risk factors. Three models have not been externally validated, and none have been demonstrated to be generalizable across diverse populations. Most models have relatively weak discriminatory power. The potential clinical and economic benefits of successfully integrating into clinical practice a risk prediction model with acceptable discriminatory power could be substantial. It remains to be defined what role the currently available and emerging models can have in practice.

Supplementary Material Note: To access the supplementary material accompanying this article, visit the online version of Clinical Gastroenterology and Hepatology at www.cghjournal.org, and at http://dx.doi.org/10.1016/j.cgh.2014.01.042.

-

2014

References 1. Siegel R, Ward E, Brawley O, et al. Cancer statistics, 2011: the impact of eliminating socioeconomic and racial disparities on premature cancer deaths. CA Cancer J Clin 2011;61:212–236. 2. Winawer SJ, Zauber AG, Ho MN, et al. Prevention of colorectal cancer by colonoscopic polypectomy: the National Polyp Study Workgroup. N Engl J Med 1993;329:1977–1981. 3. Schoen RE, Pinsky PF, Weissfeld JL, et al. Colorectal-cancer incidence and mortality with screening flexible sigmoidoscopy. N Engl J Med 2012;366:2345–2357. 4. Atkin WS, Edwards R, Kralj-Hans I, et al. Once-only flexible sigmoidoscopy screening in prevention of colorectal cancer: a multicentre randomised controlled trial. Lancet 2010;375: 1624–1633. 5. Hewitson P, Glasziou P, Watson E, et al. Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (hemoccult): an update. Am J Gastroenterol 2008; 103:1541–1549. 6. Segnan N, Armaroli P, Bonelli L, et al. Once-only sigmoidoscopy in colorectal cancer screening: follow-up findings of the Italian Randomized Controlled Trial—SCORE. J Natl Cancer Inst 2011; 103:1310–1322. 7. Frazier AL, Colditz GA, Fuchs CS, et al. Cost-effectiveness of screening for colorectal cancer in the general population. JAMA 2000;284:1954–1961. 8. Sonnenberg A, Delco F, Inadomi JM. Cost-effectiveness of colonoscopy in screening for colorectal cancer. Ann Intern Med 2000;133:573–584. 9. Lieberman DA. Cost-effectiveness model for colon cancer screening. Gastroenterology 1995;109:1781–1790. 10. Ladabaum U, Song K. Projected national impact of colorectal cancer screening on clinical and economic outcomes and health services demand. Gastroenterology 2005;129:1151–1162. 11. Sharaf RN, Ladabaum U. Comparative effectiveness and cost-effectiveness of screening colonoscopy vs sigmoidoscopy and alternative strategies. Am J Gastroenterol 2013;108: 120–132. 12. U.S. Preventive Services Task Force. Screening for colorectal cancer: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med 2008;149:627–637. 13. Qaseem A, Denberg TD, Hopkins RH Jr, et al. Screening for colorectal cancer: a guidance statement from the American College of Physicians. Ann Intern Med 2012;156:378–386. 14. Levin B, Lieberman DA, 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;134:1570–1595. 15. Joseph DA, King JB, Miller JW, et al. Prevalence of colorectal cancer screening among adults: Behavioral Risk Factor Surveillance System, United States, 2010. MMWR Morb Mortal Wkly Rep 2012;61(Suppl):51–56. 16. Felsen CB, Piasecki A, Ferrante JM, et al. Colorectal cancer screening among primary care patients: does risk affect screening behavior? J Community Health 2011;36:605–611. 17. Beydoun HA, Beydoun MA. Predictors of colorectal cancer screening behaviors among average-risk older adults in the United States. Cancer Causes Control 2008;19:339–359. 18. Straus WL, Mansley EC, Gold KF, et al. Colorectal cancer screening attitudes and practices in the general population: a

Colorectal Neoplasia Risk Prediction Models

9

risk-adjusted survey. J Public Health Manag Pract 2005; 11:244–251. 19. Sturmer T, Glynn RJ, Lee IM, et al. Lifetime cigarette smoking and colorectal cancer incidence in the Physicians’ Health Study I. J Natl Cancer Inst 2000;92:1178–1181. 20. Chao A, Thun MJ, Jacobs EJ, et al. Cigarette smoking and colorectal cancer mortality in the cancer prevention study II. J Natl Cancer Inst 2000;92:1888–1896. 21. Miller PE, Lesko SM, Muscat JE, et al. Dietary patterns and colorectal adenoma and cancer risk: a review of the epidemiological evidence. Nutr Cancer 2010;62:413–424. 22. Frezza EE, Wachtel MS, Chiriva-Internati M. Influence of obesity on the risk of developing colon cancer. Gut 2006;55:285–291. 23. Sung JJ, Lau JY, Goh KL, et al. Increasing incidence of colorectal cancer in Asia: implications for screening. Lancet Oncol 2005;6:871–876. 24. Giovannucci E. Modifiable risk factors for colon cancer. Gastroenterol Clin North Am 2002;31:925–943. 25. Tomeo CA, Colditz GA, Willett WC, et al. Harvard report on cancer prevention, volume 3: prevention of colon cancer in the United States. Cancer Causes Control 1999;10:167–180. 26. Imperiale TF, Wagner DR, Lin CY, et al. Using risk for advanced proximal colonic neoplasia to tailor endoscopic screening for colorectal cancer. Ann Intern Med 2003;139:959–965. 27. Lin OS, Kozarek RA, Schembre DB, et al. Risk stratification for colon neoplasia: screening strategies using colonoscopy and computerized tomographic colonography. Gastroenterology 2006;131:1011–1019. 28. Driver JA, Gaziano JM, Gelber RP, et al. Development of a risk score for colorectal cancer in men. Am J Med 2007;120: 257–263. 29. Kastrinos F, Allen JI, Stockwell DH, et al. Development and validation of a colon cancer risk assessment tool for patients undergoing colonoscopy. Am J Gastroenterol 2009;104: 1508–1518. 30. Freedman AN, Slattery ML, Ballard-Barbash R, et al. Colorectal cancer risk prediction tool for white men and women without known susceptibility. J Clin Oncol 2009;27:686–693. 31. Wei EK, Colditz GA, Giovannucci EL, et al. Cumulative risk of colon cancer up to age 70 years by risk factor status using data from the Nurses’ Health Study. Am J Epidemiol 2009;170: 863–872. 32. Ma 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;34:534–541. 33. Yeoh KG, Ho KY, Chiu HM, et al. The Asia-Pacific Colorectal Screening score: a validated tool that stratifies risk for colorectal advanced neoplasia in asymptomatic Asian subjects. Gut 2011; 60:1236–1241. 34. Cai QC, Yu ED, 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;175:584–593. 35. Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Open Med 2009;3:e123–e130. 36. Bouwmeester W, Zuithoff NP, Mallett S, et al. Reporting and methods in clinical prediction research: a systematic review. PLoS Med 2012;9:1–12. 37. Morgan JW, Cho MM, Guenzi CD, et al. Predictors of delayedstage colorectal cancer: are we neglecting critical demographic information? Ann Epidemiol 2011;21:914–921.

10

Ma and Ladabaum

Clinical Gastroenterology and Hepatology Vol.

-,

No.

-

38. Win AK, Macinnis RJ, Hopper JL, et al. Risk prediction models for colorectal cancer: a review. Cancer Epidemiol Biomarkers Prev 2012;21:398–410.

57. Sung JJ, Choi SY, Chan FK, et al. Obstacles to colorectal cancer screening in Chinese: a study based on the health belief model. Am J Gastroenterol 2008;103:974–981.

39. Chung SJ, Kim YS, Yang SY, et al. Five-year risk for advanced colorectal neoplasia after initial colonoscopy according to the baseline risk stratification: a prospective study in 2452 asymptomatic Koreans. Gut 2011;60:1537–1543.

58. Khandker RK, Dulski JD, Kilpatrick JB, et al. A decision model and cost-effectiveness analysis of colorectal cancer screening and surveillance guidelines for average-risk adults. Int J Technol Assess Health Care 2000;16:799–810.

40. Church JM. A scoring system for the strength of a family history of colorectal cancer. Dis Colon Rectum 2005;48:889–896.

59. Theuer CP, Wagner JL, Taylor TH, et al. Racial and ethnic colorectal cancer patterns affect the cost-effectiveness of colorectal cancer screening in the United States. Gastroenterology 2001;120:848–856. 60. Agrawal S, Bhupinderjit A, Bhutani MS, et al. Colorectal cancer in African Americans. Am J Gastroenterol 2005;100:514–523.

41. Betes M, Munoz-Navas MA, Duque JM, et al. Use of colonoscopy as a primary screening test for colorectal cancer in average risk people. Am J Gastroenterol 2003;98:2648–2654. 42. Colditz GA, Atwood KA, 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;11:477–488.

61. Lansdorp-Vogelaar I, van Ballegooijen M, Zauber AG, et al. Individualizing colonoscopy screening by sex and race. Gastrointest Endosc 2009;70:96–108, 108 e1–e24.

43. Park Y, Freedman AN, Gail MH, et al. Validation of a colorectal cancer risk prediction model among white patients age 50 years and older. J Clin Oncol 2009;27:694–698.

62. Brown ML, Riley GF, Potosky AL, et al. Obtaining long-term disease specific costs of care: application to Medicare enrollees diagnosed with colorectal cancer. Med Care 1999;37:1249–1259.

44. Levitzky BE, Brown CC, Heeren TC, et al. Performance of a risk index for advanced proximal colorectal neoplasia among a racially/ethnically diverse patient population (risk index for advanced proximal neoplasia). Am J Gastroenterol 2011;106: 1099–1106.

63. Yabroff KR, Lamont EB, Mariotto A, et al. Cost of care for elderly cancer patients in the United States. J Natl Cancer Inst 2008; 100:630–641.

45. Altman DG, Royston P. What do we mean by validating a prognostic model? Stat Med 2000;19:453–473. 46. Ferlitsch M, Reinhart K, Pramhas S, et al. Sex-specific prevalence of adenomas, advanced adenomas, and colorectal cancer in individuals undergoing screening colonoscopy. JAMA 2011; 306:1352–1358. 47. Regula J, Rupinski M, Kraszewska E, et al. Colonoscopy in colorectal-cancer screening for detection of advanced neoplasia. N Engl J Med 2006;355:1863–1872. 48. Nguyen SP, Bent S, Chen YH, et al. Gender as a risk factor for advanced neoplasia and colorectal cancer: a systematic review and meta-analysis. Clin Gastroenterol Hepatol 2009;7:676–681. 49. Peto R, Darby S, Deo H, et al. Smoking, smoking cessation, and lung cancer in the UK since 1950: combination of national statistics with two case-control studies. BMJ 2000;321:323–329. 50. Sabatino SA, Lawrence B, Elder R, et al. Effectiveness of interventions to increase screening for breast, cervical, and colorectal cancers: nine updated systematic reviews for the guide to community preventive services. Am J Prev Med 2012;43:97–118. 51. Wee CC, McCarthy EP, Phillips RS. Factors associated with colon cancer screening: the role of patient factors and physician counseling. Prev Med 2005;41:23–29. 52. Klabunde CN, Schenck AP, Davis WW. Barriers to colorectal cancer screening among Medicare consumers. Am J Prev Med 2006;30:313–319. 53. Dan YY, Chuah BY, Koh DC, et al. Screening based on risk for colorectal cancer is the most cost-effective approach. Clin Gastroenterol Hepatol 2012;10:266–271. 54. Schroy PC 3rd, Barrison AF, Ling BS, et al. Family history and colorectal cancer screening: a survey of physician knowledge and practice patterns. Am J Gastroenterol 2002;97:1031–1036. 55. Blalock SJ, DeVellis BM, Afifi RA, et al. Risk perceptions and participation in colorectal cancer screening. Health Psychol 1990;9:792–806. 56. Neale AV, Demers RY, Herman S. Compliance with colorectal cancer screening in a high-risk occupational group. J Occup Med 1989;31:1007–1012.

64. Schrag D. The price tag on progress: chemotherapy for colorectal cancer. N Engl J Med 2004;351:317–319. 65. Lansdorp-Vogelaar I, van Ballegooijen M, Zauber AG, et al. Effect of rising chemotherapy costs on the cost savings of colorectal cancer screening. J Natl Cancer Inst 2009;101:1412–1422. 66. Parekh M, Fendrick AM, Ladabaum U. As tests evolve and costs of cancer care rise: reappraising stool-based screening for colorectal neoplasia. Aliment Pharmacol Ther 2008;27:697–712. 67. Vijan S, Inadomi J, Hayward RA, et al. Projections of demand and capacity for colonoscopy related to increasing rates of colorectal cancer screening in the United States. Aliment Pharmacol Ther 2004;20:507–515. 68. Seeff LC, Manninen DL, Dong FB, et al. Is there endoscopic capacity to provide colorectal cancer screening to the unscreened population in the United States? Gastroenterology 2004;127:1661–1669. 69. Toll DB, Janssen KJ, Vergouwe Y, et al. Validation, updating and impact of clinical prediction rules: a review. J Clin Epidemiol 2008;61:1085–1094. 70. Anothaisintawee T, Teerawattananon Y, Wiratkapun C, et al. Risk prediction models of breast cancer: a systematic review of model performances. Breast Cancer Res Treat 2012;133:1–10. 71. Shariat SF, Karakiewicz PI, Roehrborn CG, et al. An updated catalog of prostate cancer predictive tools. Cancer 2008;113: 3075–3099. 72. Cassidy A, Duffy SW, Myles JP, et al. Lung cancer risk prediction: a tool for early detection. Int J Cancer 2007;120:1–6. 73. Shendure J, Ji H. Next-generation DNA sequencing. Nat Biotechnol 2008;26:1135–1145. 74. Shendure J, Lieberman Aiden E. The expanding scope of DNA sequencing. Nat Biotechnol 2012;30:1084–1094. 75. Wetterstrand KA. DNA sequencing costs. Available at: http:// www.genome.gov/sequencingcosts/2013. Accessed February 21, 2013. 76. Wacholder S, Hartge P, Prentice R, et al. Performance of common genetic variants in breast-cancer risk models. N Engl J Med 2010;362:986–993. 77. Mealiffe ME, Stokowski RP, Rhees BK, et al. Assessment of clinical validity of a breast cancer risk model combining genetic

-

2014 and clinical information. J Natl Cancer Inst 2010;102: 1618–1627.

78. Cornelis MC, Qi L, Zhang C, et al. Joint effects of common genetic variants on the risk for type 2 diabetes in U.S. men and women of European ancestry. Ann Intern Med 2009;150: 541–550.

Colorectal Neoplasia Risk Prediction Models

11

82. Imperiale TF, Stump TE, Tabbey R, et al. A risk index to stratify the risk of advanced colorectal neoplasia in asymptomatic adults. Gastroenterology 2013;144:S138. 83. Kaminski MF, Kraszewska E, Rupinski M, et al. A risk score to facilitate informed decision-making about colonoscopy screening. Gastroenterology 2013;144:S–603.

79. Talmud PJ, Hingorani AD, Cooper JA, et al. Utility of genetic and non-genetic risk factors in prediction of type 2 diabetes: Whitehall II prospective cohort study. BMJ 2010;340:b4838. 80. Paynter NP, Chasman DI, Pare G, et al. Association between a literature-based genetic risk score and cardiovascular events in women. JAMA 2010;303:631–637. 81. Thomas G, Jacobs KB, Yeager M, et al. Multiple loci identified in a genome-wide association study of prostate cancer. Nat Genet 2008;40:310–315.

Reprint requests Address requests for reprints to: Uri Ladabaum, MD, MS, Division of Gastroenterology and Hepatology, Stanford University School of Medicine, 300 Pasteur Drive, Alway Bldg, Room M211, Stanford, California 94305. e-mail: uri. [email protected]; fax: (650) 723-5488. Conflicts of interest The authors disclose no conflicts.

11.e1

Ma and Ladabaum

Clinical Gastroenterology and Hepatology Vol.

-,

No.

-

Supplementary Figure 1. Colorectal Neoplasia Risk Prediction Model Data Extraction Table. AN, advanced neoplasia; AUC, area under the receiver operating characteristic curve; RR, relative risk.