Table 2.

Systematic review studies summarising risk prediction models for risk-stratified screening

Systematic review Author	Focus of review/Review question	Search date	Search sources	Inclusion/exclusion criteria	Number of included studies	Key findings
Ma and Ladabaum (2014)	To review existing risk prediction models for colorectal neoplasia	January 1990 -March 2013	MEDLINE, Scopus, and Cochrane Library	Case control, cohort and cross-sectional studies that developed or tested risk prediction models for colorectal neoplasia for average risk populations were included. Abstracts only and non-English language articles were excluded	9 CRC risk prediction models	6 models were from the US, 1 from China, 1 from Japan and 1 from 11 Asian countries. The main risk factors included age, gender, smoking, a measure of obesity, and/or family history of CRC. 6 of the models were considered good (externally validated), 2 were fair (internally validated) and 1 was poor (unvalidated). Most of the risk prediction models have weak discriminatory power with only two (Cai et al. and Imperiale et al.) reaching the 0.70 C statistic. The majority of the models were developed among primarily White populations thus validation is required among more diverse populations to determine generalisability
Peng et al. (2018)	An overview on the development and validation of risk scores and their composition and discriminatory power for identifying people at high or low risk of AN	Until March 2018	PubMed, Embase, Web of Science	Included studies met ALL of the following criteria: 1) original research in peer reviewed journal, 2) using data from cohort, cross-sectional or RCTs to develop or validate a risk score. 3) considered at least age and sex and other risk factors, laboratory tests, genetic scores or their combination. 4) only included asymptomatic, average risk patients who underwent screening colonoscopy and 5 reported presence of AN as an outcome	22 studies evaluating 17 different risk scores	Risk scores included a median number of 5 risk factors. The most commonly considered and included factors were age, sex, FH in first-degree relatives (FDR), body mass index (BMI) and smoking; other frequently considered factors were alcohol, diabetes, NSAIDs, aspirin, physical activity, red meat and vegetable consumption, CVD and hypertension Only 7 scoring systems showed at least modest discriminatory power (AUC ≥ 0.70) in internal or external validation and meta-analysis of AUCs in 1 risk score indicated that the overall performance was relatively good
Peng et al. (2019)	Head to head validation and comparison of scores identified in Peng 2018 review against 2 large scale screening cohorts (KolosSal and BliTz)	As above	As above	As above	17 risk scores were compared: 14 from Peng 2018 and 3 additional models	Risk models used were: 6 tools from the United States, 3 tools from Korea, 2 tools from Hong Kong, 1 each from Germany, Spain, Poland, China, and Japan, and a cluster of 11 Asian cities Advanced neoplasms were detected in 1,917 (11.8%) KolosSal and 848 (11.4%) BliTz AUCs of all risk scores ranged from 0.57 to 0.65 in both studies, indicating variable, but overall modest performance in predicting presence of at least 1 advanced neoplasm
Raut et al. (2019)	To systematically review and summarise studies addessing the association of whole-blood DNA methylation markers and risk of developing CRC and its precursors	Until November 2018	PubMed and Web of Science	Not reported	19 studies reporting 102 methylation markers	5 studies in China, 3 in the US, 3 in Italy, 2 in the UK, and 1 each from Canada, Germany, Finland, Sweden, France and Lithuania. None of the risk predictions were validated in independent cohorts. AUCs were only reported for 2 studies (Heiss et al. 2017 and Nugsen et al. 2015) only two genes from the Heiss et al. 2017 study reached good discriminatory power (≥ 0.70): KIAA1549L promoters cg04036920 (0.70, p < 0.05) and cg14472551 (0.72, p < 0.05)
Stegeman (2013)	They examined to what extent the validity and performance of these cancer risk models have been evaluated	Until August 2010	Medline and Embase	Inclusion criteria were that published papers (any study design) examined multivariate risk models for breast, cervical or colon cancer (only colon analysed here). Models containing laboratory measurements were excluded	2 CRC risk prediction models	Only 2 CRC risk prediction models were identified: Freedman et al. 2009 (externally validated by Park et al. 2009) and Driver et al. 2007, both of which were based in the US. Neither of the models reached good discriminatory power. Freedman et al.'s model has C statistics of 0.610 (men) and 0.605 (women) for the model which including gastro history, medication use (aspirin/nsaid), lifestyle factors, hormone status (women only) and BMI. Driver et al.'s CRC model AUC was 0.695 for the model consisting of age, smoking, BMI and alcohol use
Usher Smith et al. (2016)	To conduct a comprehensive analysis of risk prediction tools for risk of primary colorectal cancer in asymptomatic individuals within the general population	January 2000—March 2014	Medline, EMBASE, and the Cochrane Library	Inclusion criteria: (i) primary studies published in a peer-reviewed journal; (ii) studies which identify risk factors for developing colon, rectal or colorectal cancer, or advanced colorectal neoplasia at the level of the individual; (iii) provide a measure of relative or absolute risk using a combination of two or more risk factors that allows identification of people at higher risk of colon and/or rectal cancer; and (iv) are applicable to the general population. Exclusion: Studies including only highly selected groups, or those with a previous history of colon and/or rectal cancer and conference proceedings were excluded	40 papers describing 52 risk models for inclusion in the analysis and six external validation studies	Multiple risk models exist for predicting the risk of developing colorectal cancer, colon cancer, rectal cancer, or advanced colorectal neoplasia in asymptomatic populations, and that they have the potential to identify individuals at high risk of disease. The discrimination of the models, as measured by AUROC, compare favourably with risk models used for other cancers, including breast cancer and melanoma, and several include only variables recorded in routine medical records and so could be implemented into clinical practice without the need for further data collection. Further research should focus on the feasibility and impact of incorporating such models into stratified screening programmes