A. Gbegnon1, J. G. Armstrong1, J. Monestina1, J. W. Cromwell1 1University Of Iowa,General Surgery,Iowa City, IA, USA
Introduction: Hospital readmissions are costly and rates of these are increasingly being used as measures of quality. Several predictive models have been developed to aid in the identification of patients at high risk of readmission so that valuable readmission-prevention resources may be appropriately assigned. The LACE index (LI) has become the most widely used of these tools because of its ease-of-implementation using electronic health record data, even being embedded into some EHR systems. A LI of 10 or higher is frequently used to identify patients at high risk of readmission. The LI was developed primarily on non-surgical patients and has not been validated in the surgical populations to which it is now being applied. Poor discrimination of readmission risk in this population would likely result in under-resourcing of this group of patients. Our goal is to evaluate the performance of the LI on encounters of general surgery patients in our hospital.
Methods: We performed a retrospective analysis of patients who underwent a general surgery operation between January 2011 and March 2014, and whose readmission data was submitted to the National Surgical Quality Improvement Program (NSQIP). The primary outcome measure was unplanned, related readmissions within 30 days of operation. Exclusion criteria included patients who did not have a LI, who died within 30 days of their operation, and patients who had not been discharged within 30 days of their operation. To examine LI discrimination we generated a receiver operative characteristic (ROC) curve, and calculated the area under the curve (AUC). The LI was calculated by the method of van Walraven et al. from discrete elements within the EHR.
Results: There were 219 patient encounters that met inclusion criteria. The overall readmission rate in the study population was 12.8%. The readmission rate for encounters with a LI=<9 was 13.6%, while the readmission rate for LI >=10 was 9.5%. The positive predictive value using this threshold was 0.14. The AUC (c-statistic) for the LI was 0.51, indicative of poor discrimination.
Conclusion: This study is the first to attempt to validate the LI for identifying patients at high risk of readmission in a general surgery population. The LI exhibited poor positive predictive value and discrimination approaching that of random guesses in this population. With the LI being widely applied to hospital populations for the purpose of identifying patients in need of readmission prevention resources, general surgery patients may be under-resourced where this index is being used. Surgeons and hospitals should be aware of the limitations of the LI and seek other strategies for identifying surgical patients at high risk of readmission.
