Seminar

alternative econometric approaches to dealing with health care costs and expenditures.

The one day masterclass aimed to provide health economists with the latest economic and statistical modelling tools for examining health care costs and expenditures. Professor Manning explained that the traditional ordinary least squares (OLS) approach does not work well when modelling patient-level health care costs or expenditures. Cost data tends to be highly skewed, with some patients having exceptionally high rates of utilisation; whilst other individuals incur zero costs. This means OLS estimates can mis-predict average costs for a group under investigation , are inefficient (have large variance) and will provide poor forecasts if the results are generalised to other groups.

In data sets for which observations of zero costs are not a problem, one potential solution is to perform OLS on costs transformed by a suitable function, for example the square root. However, re-transforming costs back to their original scale may itself cause problems. An alternative approach is to use one of a class of generalised linear models (GLM). Professor Manning referred to some of his own applied work with a variety of different models to argue that there is no single preferred solution but that some will perform better than others depending on the characteristics of the particular data set being used.

The GLM estimator can also be applied to data sets which have a substantial number of zeros, as the second stage of a two-part modelling strategy. In the first part, the analyst addresses the distinction between utilisers and non-utilisers of health care and, in the second, estimates a separate model for cases with positive costs. A final application of GLM methods is in the modelling of count data (e.g. number of physician visits, number of admissions).

The audience, drawn from academic, pharmaceutical and international policy making organisations, were extremely positive about the masterclass, with over 95% of delegates saying they would attend another.