1. average; nearly all members are likely to be high to extreme-high risk. Finally, Group Z is the extreme low risks; ranking within Group Z is not easy and the range of predicted costs how sensitive are the models to the non-cost covariates. Table 4 summarizes our results. We observe that while nearly all models give similar values for the mean expected costs (and predictive ratios close to 1), the range of predicted costs varies substantially. For example, on the high-risk diabetes group, the predictive ratios of Lasso and GBM are almost the same (8524/10150 and 8448/10150 respectively);Group D is very high-risk
2. Algorithmic prediction of health-care costs;D Bertsimas;Operations Research,2008
3. National health expenditure projections, 2012-22: Slow growth until coverage expands and economy improves;G A Cuckler;Health Affairs,2013