Table 2

Characteristics of a population risk algorithm

To be useful, a risk algorithm should:
 Be feasible and practicalThe risk algorithm can be easily applied to population data. Variables contained within the risk algorithm are available within routinely collected population health data (eg, population health surveys)
 Be applicable for important target populationsThe risk algorithm can be calculated for important target populations such as by sex and for different age groups
 Offer an equity lensThe risk algorithm can be used to assess or compare risk in different equity settings (eg, can produce accurate risk assessment across socioeconomic position)
To be valid and robustly support its intended uses, a risk algorithm should:
 Be well calibratedThe predicted risk estimates closely approximate observed or actual risk. Calibration should be assessed for intended target populations
 Be easily recalibratedThe algorithm can be easily calibrated or recalibrated for different target populations
 Be discriminatingThe algorithm correctly identifies which individuals or populations are at high and low risk
 Have considered important risk exposures and predictorsDuring algorithm development, risk exposures are assessed for their predictive ability if they are prevalent in the general population and/or have a strong causal association with the outcome
 Have assessed causal relationships for risk exposuresDuring algorithm development, developers compare predictive risk of individual exposures to causal studies