Date and Time
August 25, 2004
1:30 PM - 3:00 PM

Availability
Open to the public
No RSVP required

Researchers investigating health care at the end of life often attempt to control for patient case mix, and one common method is to use the patient's "underlying cause of death" as reported on the death certificate. However, using this designation may not be appropriate for the objectives of many studies. Furthermore, death certificate data may not be accurate. Therefore, other methods have been used to control for case mix. Wei Yu and Samuel Richardson sought to analyze the consistency of three commonly used methods to control for case mix when measuring end-of-life healthcare utilization.

For their study, they obtained the "underlying cause of death" as reported on the death certificate for 12,808 VA patients who died in California in 2000 and 2001. Using the administrative data from VA and Medicare, they analyzed the classification consistency with three methods: (1) using the final diagnosis before death, (2) using the diagnosis responsible for the plurality of health care costs in the final year of life, and (3) using the underlying cause of death reported in the death certificate. They examined the consistency of three different levels of identification: (1) the top two causes of death, (2) the top four causes of death, and (3) the top 10 causes of death, with the CDC classification for the general population with ICD-9 codes.

The researchers found that the consistency of these methods declined with the increase in the number of case-mix groups to be classified, ranging from 65% to 70% for the top two causes of death (cancer and heart disease), from 56% to 64% for the top four cause of death, and from 47% to 58% for the top 10 causes of death. For the method that used the final diagnosis, giving higher priority to diagnoses in the common causes of death produced results that were more consistent with the other methods than treating all diagnoses equally. As a whole, these methods are more consistent in identifying cancer deaths than deaths due to other causes, such as heart disease, cerebrovascular diseases, and chronic obstructive pulmonary disease.

Based on their analyses, the researchers concluded that the three common methods used to control for case mix generate considerable discrepancies in patient classification. Therefore, they conclude, researchers should carefully consider the appropriateness of the method used for their research.

At the end of their presentation, Yu and Richardson will lead a discussion on the following question: For a study that analyzes the variations in medical treatment near the end of life, which of the three methods is the best to control for case mix?