People with high health spending tend to have multiple conditions, making identifying which disease is responsible for how much spending tricky. Aside from an academic interest in attribution of spending to disease, such knowledge can guide care management efforts for complex patients and even help with setting national health priorities. A new study at the National Bureau of Economic Research evaluates several methods. (NBER Paper) One long-used method is simply to assign each claim to a primary disease. That relies on the accuracy of diagnosis coding by physicians. Other methods attempt to use regression analyses to attribute all spending in a year, for example, to the set of diseases or conditions the patient had. These methods often have a large residual set of spending which is hard to attribute. These researchers develop what they call a propensity score model that partly compares patients with a disease to those without it and also attempts to take into account the number of comorbidities.
They test the various approaches using a Medicare data set, focusing on 78 higher-spending conditions. The claims method tends to assign more spending to an acute condition or disease exacerbation, while the regression or propensity score methods tend to assign more spending to comorbid, usually chronic, conditions. The authors’ propensity score model seems to do a better job of matching spending and not leaving large amounts of unattributed spending. But the issue remains of how the presence of one condition may affect spending in others, for example, mental health issues may cause more spending on various medical diseases, or treating hypertension may lower spending on heart attacks. No method seems to handle this issue well. One way to try to attribute spending for comorbid patients is to first identify spending on patients with only the one disease and use that to help allocate spending when patients have multiple diseases. Unfortunately few institutionalized patients have just one condition. The authors use DRG weights to help create average spending amounts per condition.
Using this modified claim attribution method they find that diseases of the circulatory system were the most expensive, costing about $148 billion in 2009 or 23% of all spending. Nervous system disorders were next at $64 billion and then musculoskeletal at $62.2 billion. The top five disease groups were 59% of all spending. The regression model methods are problematic because spending is heavily skewed, only a few people account for a high percent of all expenses. The authors include a variety of demographic and other individual characteristics to try to adjust the models. This regression model actually has circulatory diseases second, and with a lower percent of all spending. The largest category is a generalized other conditions and the third largest is a residual of unexplained spending. Not very satisfactory. Under the authors model of propensity scoring, circulatory disease were the top category, at about 23% of spending, as in the claim attribution approach. Other conditions came in second, followed by respiratory diseases. One conclusion is that none of the methods is really great at attributing spending by disease.