As the mandates for public reporting of healthcare-associated infections (HAIs) expand along with pay-for-performance incentives, those entrenched in the collection and analysis of this data have voiced concern about whether valid comparisons between healthcare facilities can be made. The issue: adjustment for underlying disease severity.
A recent paper published by van Mourik and colleagues1 for the Centers for Disease Control and Prevention Epicenters describes the development of a disease severity prediction model based solely on data available in the electronic health record (EHR) for risk-adjustment in a retrospective cohort of mechanically ventilated patients. The authors note that two tiers of the current NHSN ventilator-associated event definition (VAE) – ventilator-associated condition (VAC) and infection-related VAC (IVAC) – are targeted for public reporting in the future. The only case-mix adjustment currently utilized by NHSN for VAE is stratification by type of intensive care unit (ICU). The available ICU mortality risk scoring systems, such as APACHE, have not been widely adopted due to the labor-intensive manual data extraction required. As EHRs are increasingly used, there is interest in the electronically captured and stored data to determine disease severity.
The primary outcome for the model was in-hospital mortality with VACs and IVACs targeted for risk adjustment among a cohort of 20,028 patients who had recorded mechanical ventilation (MV) through an endotracheal or tracheostomy tube. Potential predictive data were extracted from the hospital’s patient data registry and included patient demographic characteristics (age, sex), type of ICU, days from hospital admission to start of MV, microbiology and laboratory data (e.g. blood culture, white blood cell count, blood gas analysis), procedure and diagnosis codes, billing information and medication use. Values for predictors were collected from the calendar day of MV onset and the day prior, if available, with the observation reflecting the worst value chosen. Comorbidity measures were assessed using a published method. The regression models were built with incremental complexity, initially including groups of predictors with the best expected feasibility and generalizability. The model with the most favorable balance of complexity to in-hospital mortality prediction was developed and stratified by ICU type. The relationship between predicted mortality risk and observed incidence of VAEs was examined to assess the use of disease severity estimates for VAE risk-adjustment.
The authors found that accurate estimation of disease severity in ventilated patients was feasible with a simple model which included demographic characteristics, type of ICU, time to intubation, blood culture sampling, 8 common laboratory tests and surgical status. It was also determined that the estimated disease severity was associated with the occurrence of VAEs.
Although more research is needed to validate and refine the proposed models, this paper illustrates how the use of readily available electronic data can provide for accurate estimates of disease severity and be used for the adjustment of VAE rates. We have advanced a step toward more valid comparison of VAE rates between hospitals and in the discovery of how EHRs can be used for risk-adjustment.
1 van Mourik M, Moons K, Murphy MV, Bonten M, Klompas M. Severity of Disease Estimation and Risk-Adjustment for Comparison of Outcomes in Mechanically Ventilated Patients Using Electronic Routine Care Data. Infect. Control Hosp. Epidemiol. 2015;36(7):807-815.