Development of a risk score predicting colorectal cancer among middle-aged Australian men and women
Final Report Abstract
Colorectal cancer (CRC) is the third most common malignancy in men and the second in women worldwide. Given that healthcare resources are increasingly in short supply, quantifying individual risk of CRC and identifying individuals at high risk may help targeting scarce resources efficiently to maximise health benefits. The aim of the current research project was therefore to develop a risk score to predict 5-year CRC risk using data from 197,874 individuals participating in the 45 and Up Study, a large Australian cohort study. The model’s performance and clinical usefulness was subsequently evaluated using an independent study population comprising 24,233 participants from the Melbourne Collaborative Cohort Study (MCCS). The prediction model was based on the following personal and lifestyle factors that can be easily assessed in the primary care setting: age, sex, BMI (kg/m²), prevalent diabetes (yes/no), ever having undergone CRC screening (yes/no), smoking status (never, former, current) and alcohol intake (drinks/day). Individuals who developed CRC within 5 years had higher predicted risks than persons who did not develop CRC approximately 73% of the time, which implies reasonable discriminatory accuracy. Comparison of observed risk with predicted risk indicated that the model accurately predicts absolute risk (calibration). Importantly, the model’s performance was well maintained in the external MCCS which was developed entirely separately using independent methodologies. This suggests that the model is generalizable to other populations and could potentially be implemented in clinical practice. The optimal cut-off point according to the Youden’s index to define high-risk individuals was ≥4.5 risk score points and identified 80% of individuals who developed CRC during 5-years in the validation sample (sensitivity), while 48% of individuals not developing CRC had a risk score below this threshold (specificity). Decision-curve analysis showed a net benefit of 0.0041 when using the model at the threshold of ≥4.5 (corresponding to a risk of ≥0.68%), which would finally result in 20% fewer individuals being required to undertake preventive actions and to having anxiety about their CRC risk in comparison to assume all individuals are at risk. When implemented in practice, the model may raise people’s awareness for their individual CRC risk and motivate lifestyle changes or undergo preventive strategies more regularly. The model may also be valuable in defining inclusion criteria for risk-based CRC screening programs and thus assist in focusing scarce resources in population-based screening. Before implementation, however, the effect of using the model in the envisaged field of application, including a careful evaluation of health outcomes and cost-effectiveness, needs to be quantified.
