Use of Statistical Control Charts to Assess Outcomes of Medical Care: Pneumonia in Medicare Patients

Abstract

Detection of nonrandom variation in outcomes with statistical control charts is at the heart of quality improvement techniques. The authors examined the charts' ability to detect variations in outcome of pneumonia. They surveyed Medicare claims data for DRG 89, pneumonia with complications or co-morbidities, from November 1988 through October 1991 at 20 Illinois hospitals with the most Medicare discharges for DRG 89. Control charts were constructed on five outcomes--mean length of stay, range of length of stay, mortality, readmissions, and complications. Standard techniques from industrial statistics were used to construct the historical means and control limits derived from 2 years of data, to plot the monthly samples from the 3rd year of data and to score the control charts for nonrandom variation at less than 1% probability. The observed number of control charts with nonrandom variation was 33 of 100; the expected number was 9.18 (p < 0.0001). Nineteen hospitals had 1 to 3 control charts with nonrandom variation on the five outcomes, whereas only one hospital had none. The number of control charts with nonrandom variation per hospital did not correlate with hospital size, occupancy, teaching status, location, or payer-mix. Statistical control charts provide simple tools for identification of nonrandom variation in outcomes. To the extent that these variations can be related to quality issues, the charts will be useful for quality management.