reports

Ambulatory Care Groups: an evaluation for military health care use

Fort Sam Houston, TX, USA: United States Army Center for Healthcare Education and Studies, publication #HR 94-004.

United States Army Center for Healthcare Education and Studies, Fort Sam Houston, TX, USA

The Ambulatory Care Evaluation Study (ACES) team, part of the U.S. Army Health Care Studies and Clinical Investigation Activity, evaluated the Ambulatory Care Groups (ACGs) System developed at Johns Hopkins University for possible military use. The ACGs are unique in that they were developed to categorize patients or populations rather than visits or services. The ACGs categorize diagnoses according to their likelihood or persistence. They are conceptually simple and require only patient age, gender and ICD-9-CM diagnoses over the period of time under study. The ACGs are based on 34 broad clusters of ICD-9-CM diagnoses called ambulatory diagnostic groups (ADGs). The ACGs were developed from enrolled population data during a one-year time period. The ACG developers made three specific claims about the ACG grouper performance: First, 30-50% of the variance in number of visits and ambulatory charges can be explained by the ACG designation. Second, the ACG grouper is twice as powerful in predicting resource use in the same year of morbidity than in the subsequent year of utilization (the ACGs explained over 20% of the variance in visits and ambulatory charges in the subsequent year). Third, the number of unique ADGs explain up to 50% of the variance in utilization for the year in which they were assigned and up to 42% of the variance in charges for the year of assignment. The ACG developers do not recommend the grouper as an appropriate tool for making individual decisions about the financial impact of a single patient’s health status. The ACG grouper system is recommended for application to research as well as payment and management of Medical Service.
The ACES team’s evaluation of the ACGs addressed the following four issues: (a) clinical evaluation, (b) user friendliness (programming and administrative issues), (c) statistical analyses of the grouper results, and (d) military applicability.
The team found that the ACG groups are conceptually sound. However, there were some problems with the grouper algorithm in the pilot version evaluated by the study group. The pilot version of the ACGs used over 5,000 common ICD-9-CM diagnoses in the grouping algorithm. The ACG grouper algorithm should be modified to assign the majority of ICD-9-CM diagnoses to ambulatory diagnostic groups (ADGs). The limited list of ICD codes resulted in an underestimation of morbidity levels. There are also some inconsistencies in the assignment of diagnoses to major, versus minor, ADG categories. For example, the assessment of psychiatric diagnoses to ADGs 23 and 24 should be revised because minor psychiatric conditions are assigned to the ADG for Psychosocial: major. The titles of ACGs arising from Major Ambulatory Categories (MAC) 10, 17, 21, and 23 should be revised, as the use of the term psychosocial to describe psychophysiologic conditions is misleading.
The ACG grouper is available in a personal computer or mainframe version. Both versions were used on a test file and provided the same results. The ACES team used a 12-month sample of data containing 774,750 patient records representing 260,515 unique patients (called the Year Sample) to evaluate the ACG grouper. The mainframe version was used for the evaluation of the Year Sample because it rapidly processed the large sample. The input data had to be sorted by patient ID and the output was an ACG code per unique patient. The grouper appeared to appropriately group all patient records presented to it. Analyses of variance (ANOVA) procedures were performed to assess the relationship between the ACG category and a variety of dependent variables which included using ACES cost formulas and logarithmic cost, total number of ambulatory visits, and total number of diagnoses made in the year. These results are in Table A Costs (including Logarithmic Cost).

These r-squares approximate the r-squares reported by the ACG developers. The ACG grouper explains approximately 50% of the variables in number of visits, from 35% to 44% of the variance in costs and 43% to 47% of the variance using logarithmically transformed costs in the year studied and 71% in the variance in the number of diagnoses (roughly equivalent to the level of morbidity).