The simplest approach is to group diagnoses into a manageable number of morbidity categories which can then be included in regression models of patient costs as a set of dummy variables indicating the presence or absence of specific diagnoses. This assumes that the effect of diagnoses is additive. But the cost of one patient with both diabetes and depression may be greater than the cost of two patients, one with diabetes and the other with depression,because it may be more difficult to control blood sugar levels for a depressed patient. Conversely, there may be cost savings with some multimorbid patients. For example, heart disease and diabetes are conditions where monitoring of cholesterol may be required but the associated costs need only be incurred once in a given period for a patient with both conditions. Allowing for the possible non-additive effects of multi morbidity is potentially important since the proportion of the population who are multi morbid is non-trivial(20% to 61% in our data set depending on the multi morbidity measure used) and has been growing over time (Hippisley-Cox and Pringle, 2007).