Risk prediction models for colorectal neoplasia or cancer in symptomatic individuals: a systematic review
The problem
Colorectal cancer (CRC) is the second leading cause of cancer-related death in Europe. Detecting the disease at an early stage improves outcomes and whilst screening can make important contributions to achieving this, the majority of cases are still diagnosed symptomatically. Longer time from symptom onset to diagnosis is associated with increased mortality. Individual symptoms, such as rectal bleeding and change in bowel habit, are associated with CRC but are also common in populations without cancer. Risk prediction models which combine multiple risk factors and symptoms of CRC could improve the consistency and quality of clinical decision-making. We aimed to systematically identify and compare the performance of models that predict the risk of primary CRC for symptomatic individuals.
The approach
We searched Medline and EMBASE to identify primary research studies reporting or validating CRC risk models. To be included, models needed to assess a combination of risk factors that included CRC symptoms, and be applicable to the general population. Screening of studies for inclusion and data extraction were completed independently by at least two researchers.
Findings
A total of 12,808 papers were identified from the literature search and 2 through citation searching. Of these, 26 papers describing 36 risk models were included. The majority of models predicted the risk of a single outcome: 25 models predicted the risk of CRC; 1 CRC or pre-malignant adenoma; 1 ‘significant colorectal (CR) neoplasm’; 3 ‘advanced CR neoplasm’; and 3 adenoma. We also included 4 papers validating published models and 1 paper describing the usability and acceptability of a model amongst GPs.Of the 36 models, 8 were developed using primary care data sets, 27 in referred patient groups and 1 in a screened population. Where reported, the discrimination of models was good (AUROC range: 0.76-0.92) and similar for those developed using primary care data (0.79-0.92) and referred patient groups (0.76-0.88). All of the best performing models included anaemia and therefore required a blood test.
Consequences
:A large number of models predicting the risk of colorectal neoplasia or cancer in symptomatic individuals have been developed. Despite the majority of models being derived in referred patient groups, the highest discriminatory statistics were reported using primary care derived models. Further work is needed to evaluate the best performing models in the same data set, perhaps alongside the proposed NICE guidelines for suspected cancer. Additional research is also required to assess the comparative usability and acceptability of these risk prediction models, and to identify optimal thresholds to minimise both the duration of undiagnosed disease and the potential harms of unnecessary investigation and treatment.
Credits
- Thomas Williams, The Primary Care Unit, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK
- Jon Emery, General Practice and Primary Care Academic Centre, University of Melbourne, Melbourne, Australia
- Aung Win
- Simon Griffin, General Practice and Primary Care Academic Centre, University of Melbourne, Melbourne, Australia
- Fiona Walter, General Practice and Primary Care Academic Centre, University of Melbourne, Melbourne, Australia
- Juliet Usher-Smith, General Practice and Primary Care Academic Centre, University of Melbourne, Melbourne, Australia