Few dispute the need to adjust for severity of illness to produce meaningful comparisons of resource use, clinical outcomes, or medical effectiveness across patients and providers. Despite this consensus, two major questions persist: what precisely is """"""""severity,"""""""" and how should it be measured? In the last decade, over a dozen severity of illness measures have been developed or have matured. Many health services researchers have adopted one or more of these existing systems for their investigations which require severity adjustment. This is reasonable given the conceptual complexities and potential expense of de novo development of a severity measure. However, each existing system has unique conceptual underpinnings and methods for rating patients. This conceptual concern as well as preliminary research findings suggest that research results may be highly dependent upon the method chosen to adjust for severity. Little information is available to help researchers make informed choices about which severity measure is appropriate to their needs. The overall purpose of this project is to produce a """"""""road map"""""""" to assist researchers in the choice and use of severity of illness measures appropriate to their study goals. Nine existing """"""""severity' measures will be examined: the Acuity Index Method, the revised Acute Physiology and Chronic Health Evaluation, the Body Systems Count, the Charlson comorbidity index, the Computerized Severity Index, Disease Staging Q-Stage and Q-Scale, MedisGroups, Patient Management Categories, and a Yale University refinement of the diagnosis related groups. The analyses will use four large databases containing both discharge abstract data as well as the information required by at least one of the systems dependent on medical record review. Our analyses will focus on the following: (1) the conceptual foundation of the severity measures; (2) the way the severity algorithms use clinical and other information to rate patients; (3) the abilities of these systems to adjust for risk of three major outcomes (resource use [charges or adjusted charges and length of stay], in-hospital and post-discharge mortality, and in-hospital complications) at the patient- and hospital-levels; and (4) various statistical considerations in using severity information for judgments about patient outcomes (e.g., reliability of summary statistics of model performance, such as the R-squared; impact of outlier definitions; delineating the types of cases in which errors in predictions occur). The final report of this two and one-half year project will detail important considerations for choosing among severity of illness measurement systems for different research objectives, particularly patient outcome studies.
