ATS 2012, SAN FRANCISCO – In-hospital mortality for ICU patients is often used as a quality measure, but discharge practices may bias results in a way that disadvantages large academic hospitals, according to a recently conducted study.
"Hospitals differ in the number of patients they transfer to other hospitals or post-acute care facilities," said lead author Lora Reineck, MD, post-doctoral fellow at the University of Pittsburgh School of Medicine. "These differences can affect in-hospital mortality measurement if some hospitals discharge patients more frequently or earlier than others, since in these cases the mortality burden is shifted to other facilities. While it's known that discharge practices alter in-hospital ICU mortality measurement, it was previously unknown whether this effect is uniform across hospitals or whether certain types of hospitals are more affected than others."
The results will be presented at the ATS 2012 International Conference in San Francisco.
"We found that this 'discharge bias' disproportionally hurts large hospitals and academic hospitals, which frequently accept many patients in transfer from other hospitals," said Dr. Reineck. "Hospitals that care for a large number of patients insured through health maintenance organizations (HMOs) are also affected, since these organizations typical restrict transfers. Mortality measures tied to a specific time point, such as 30-day mortality, are less biased by discharge practices but are harder to calculate."
In recent years, quality measures have increasingly been publicly reported and tied to financial incentives through pay-for-performance. It is important that these quality measures accurately assess performance and are not flawed by bias. According to Dr. Reineck, "our study reveals that large academic hospitals, as well as hospitals with a high proportion of commercial HMO patients, are more negatively affected by using in-hospital ICU mortality compared to 30-day mortality than other hospitals. Accounting for this bias might prevent these hospitals from being unfairly penalized in public reporting or pay-for-performance programs."
The retrospective cohort study used data on 43,830 ICU patients admitted to 134 hospitals in Pennsylvania in 2008. Discharge bias was defined as 30-day mortality minus in-hospital mortality; greater discharge bias makes a hospital appear of relatively higher quality.
Mean risk-adjusted hospital-specific 30-day and in-hospital mortality rates were 13.1 ± 1.6%and 9.6 ± 1.3%, respectively, resulting in a mean hospital-specific discharge bias of 3.5% ± 1.3%. Discharge bias was greater in small hospitals, non-teaching hospitals, and hospitals with fewer commercial HMO patients, thereby making these facilities appear relatively better in quality compared to large teaching hospitals or those with a high proportion of commercial HMO patients. Hospital rank was greatly affected by discharge bias, with 29.1% of hospitals increasing in rank by at least one quartile and 26.9% decreasing in rank by at least one quartile. Large teaching hospitals and hospitals with the highest proportion of patients with commercial HMO insurance were more likely to decrease in rank than small, non-teaching hospitals or hospitals with a lower proportion of commercial HMO patients.
"State and national programs that use in-hospital mortality to benchmark hospitals should note how discharge bias unfairly disadvantages certain types of hospitals," concluded Dr. Reineck. "Discharge bias must be accounted for to prevent unfair performance assessments."
Future studies are planned to assess the effects that using this measure in public reporting has on outcomes of ICU patients.
"Bias In Quality Measurement Resulting From In-Hospital Mortality As An ICU Quality Measure" (Session B15, Monday, May 21, 8:45 a.m., Room 3020-3022, Moscone Center; Abstract 30009)
* Please note that numbers in this release may differ slightly from those in the abstract. Many of these investigations are ongoing; the release represents the most up-to-date data available at press time.
Bias In Quality Measurement Resulting From In-Hospital Mortality As An ICU Quality Measure
Type: Scientific Abstract
Category: 02.07 - Health Services, Policy, Financing Organization (BS)
Authors: L.A. Reineck1, F. Pike1, T. Le1, B. Cicero1, T.J. Iwashyna2, J.M. Kahn1; 1University of Pittsburgh - Pittsburgh, PA/US, 2University of Michigan - Ann Arbor, MI/US
Rationale: The National Quality Forum recently endorsed in-hospital mortality for ICU patients as a hospital quality measure. However, in-hospital mortality assessments can be affected by discharge to post-acute care facilities, such that differences in in-hospital mortality may reflect variation in discharge patterns rather than quality of care. If discharge practices vary between different types of hospitals, use of in-hospital mortality as a performance measure in quality improvement may lead to health disparities.
Methods: We performed a retrospective cohort study of ICU patients admitted in 2008 using data from the Pennsylvania Health Care Cost Containment Council (PHC4) state hospital discharge dataset. These files were linked to vital status files to obtain death information for each patient and to Medicare's Healthcare Cost Report Information System to obtain hospital characteristics. We excluded non-general acute care hospitals and hospitals with <50 ICU admissions. Standardized hospital-specific risk-adjusted in-hospital and 30-day mortality rates were calculated using hierarchical regression models controlling for severity of illness and case-mix. We defined discharge bias as 30-day mortality minus in-hospital mortality, such that greater discharge bias makes a hospital appear of relatively higher quality. We then used linear regression to compare discharge bias and change in hospital rankings due to discharge bias across hospital characteristics, including size, location, teaching status (defined using resident to bed ratio), and safety net status (defined as the proportion of Medicaid patients).
Results: The final analysis included 34,932 patients in 128 hospitals. Mean risk-adjusted hospital-specific in-hospital mortality rates and 30-day mortality rates were 10.4+1.2% and 13.2+1.6%, respectively. Mean hospital-specific discharge bias was 2.8+1.2% (i.e. the average hospital's risk-adjusted mortality improved by 2.8% due to discharge bias) and ranged from -1.4% to 6.2%. Overall, discharge bias was greater in small and non-teaching hospitals, making these hospitals appear comparatively better from a benchmarking standpoint (Table). Discharge bias led to substantial changes in hospital rank (Figure). Discharge bias led to hospitals increasing in rank by at least one quartile 27.3% of the time and decreasing in rank by at least one quartile 27.3% of the time. Large hospitals (mean change in rank: -27%, p<0.01) and teaching hospitals (mean change in rank: -16.5%, p=0.02) were more likely to decrease in rank than small and non-teaching hospitals.
Conclusion: Discharge practices bias in-hospital mortality measures in a way that disadvantages large hospitals and academic hospitals. Accounting for discharge bias will prevent these hospitals from being unfairly punished in public reporting or pay-for-performance.
Funded by: NIH