Value-based reimbursement, which judges a hospital’s outcomes (and, subsequently, its payments) by comparing them with other hospitals’ outcomes, has boosted interest in risk adjustment in an attempt to level a playing field that is, in reality, quite uneven.
At the heart of this issue is one indisputable fact: patient populations differ.
“[U]sing mortality rates to compare hospitals or health care systems is invalid if the analysis fails to take into account the differences in a population’s overall risk, making it appear that hospitals caring for sicker populations provide worse care,” write physician researchers in a recent issue of the journal Quality Management in Health Care. “To use mortality as an outcome measure, it is essential to rigorously risk-adjust for the illness of the patient population.”
In 2008, the journal Medical Care published a powerful and externally validated predictive model by Escobar et al designed to risk-adjust for 30-day mortality among inpatients, explain Douglas Tremblay, MD, Julia H. Arnsten, MD, MPH, and William N. Southern, MD, MS, in the Quality Management in Health Care article.
Unfortunately, they add, the model is complex, requiring elements that may be difficult to pull from electronic health records (EHR). As a result, the Escobar model is not widely used.
The authors, physicians from Mount Sinai Medical Center and Montefiore Medical Center in New York, explain their efforts to extend Escobar’s work by developing and validating a simpler predictive model. Their model provides an easier way hospitals can gauge 30-day mortality for inpatients, they report. Better yet, the three-element model offers predictive power similar to the more complex risk-adjustment models.
The model factors in three elements to predict 30-day mortality for inpatients: comorbidity (measured using the Charlson Comorbidity Index score), acuity of illness (measured using the Laboratory-based Acute Physiology Score, or LAPS), and age.
The authors developed —and then validated — the model using data from adult admissions between July 1, 2002, and April 30, 2008, at Montefiore Medical Center, an urban academic medical center with a diverse patient population in the Bronx, New York.
“Overall the model performed well predicting 30-day mortality and was well calibrated across risk deciles,” they report. “The model offers a compelling balance of simplicity of use and predictive power and is therefore a useful risk-adjustment tool.”
The model is also appropriate for a varied patient population.
“It is worth noting that our model was derived and validated in a racially diverse inpatient population with high rates of poverty, poor support systems, and high burdens of chronic comorbid disease,” they write. “It is encouraging that this model performed so well as a predictor of mortality in such a diverse population, suggesting that it will also perform well when tested in other patient populations.”
Adjusting for risk is crucial, they write, in pay-for-performance programs to protect a hospital’s resources as well as its patient population.
“Risk adjustment is particularly important in areas with many disadvantaged patients because providers who care for disadvantaged patients tend to perform poorly on the type of quality measures used in pay-for-performance programs,” the authors explain.
“Failure to appropriately adjust for the increased risk associated with the care of disadvantaged patients may worsen health care disparities by steering resources away from hospitals that care for disadvantaged patients.”
Their risk-adjustment model, they add, is helpful for research efforts and quality improvement initiatives, too. In addition, the necessary data elements are already available in most institutions’ electronic health records and, at the most, would require some rather straightforward programming from the IT department.
“It is relatively easy to calculate the elements of the model using clinical/administrative data,” they observe.
How could this model benefit your hospital? Leave us a comment below!