How Can Surgeons Predict Risk of Readmission?
A new study finds that 3 industry-leading indices for predicting mortality and healthcare utilization fail as models for predicting which patients may need to be readmitted after hip and knee replacement surgery.
About 1 million Americans each year undergo total knee or total hip replacements, but complications bring as many as 1 in 12 back to the hospital and result in higher use of post-acute services within 90 days.
To compel hospitals to do better, the Centers for Medicare and Medicaid Services (CMS) launched the Comprehensive Care for Joint Replacement (CJR) program in April 2016, which penalizes hospitals for readmission of joint replacement patients within 90 days.
A new study, however, finds that CMS and healthcare providers lack the predictive models needed to assess which patients are at risk for readmission.
Some hospital systems fear being penalized inadvertently because CJR’s current payment model does not include a risk adjustment method to account for patients’ medical complexity or their functional status, said study lead author Amit Kumar, PT, MPH, PhD, a postdoctoral research associate at the Brown University School of Public Health in Providence, Rhode Island.
In the new study, Dr. Kumar and his co-authors tested the 3 best candidates – including an index developed by CMS – but found that none were useful in predicting readmissions among patients who underwent joint replacement for osteoarthritis.
Therefore, a model, or index, is needed that will accurately predict the risk of readmission to improve patient care and help CMS judge hospitals on the quality of their care rather than on how inherently risky their patients are, Dr. Kumar said.
“In the absence of that risk adjustment, when sick patients have worse outcomes, hospitals will be penalized,” said Dr. Kumar, whose paper appears in Arthritis Care & Research.
“If we could find an index that was working for this population, we could recommend that – but unfortunately none of them are working very well.”
Existing Models Are Not Predictive
Dr. Kumar and his former colleagues at the University of Texas Medical Branch tested the applicability of the 3 industry-leading indices for predicting mortality and healthcare utilization:
- Charlson Comorbidity Index
- Elixhauser Comorbidity Index
- Hierarchical Condition Category from CMS
They analyzed Medicare data on every beneficiary who survived for 90 days after a total knee or total hip replacement for osteoarthritis between January 2009 and September 2011. In all, the study covered a total of 605,417 patients. Data showed that 46.3% of patients were discharged home, 40.9% went to skilled nursing facilities, and 12.7% stayed in inpatient rehabilitation.
The analysis sought to determine whether any of the 3 indices would make a meaningful difference in predicting where patients would be discharged and whether they would return to the hospital within 30, 60, or 90 days.
They did not, the study authors found. In fact, none of the indices they evaluated significantly improved on a “base model” of merely accounting for a mix of demographic and medical factors.
To rate the base model and the 3 indices, the study authors relied on the calculation of the C-Statistic, which essentially measures the probability that an index would identify as high risk a person who actually turned out to be high risk.
By convention, a C-statistic has to be higher than 0.7 to be considered clinically relevant. The base model scored in the 0.63 to 0.65 range, and the indices only nudged those numbers up in the hundredths place, never rising above the 0.7 threshold.
What’s Missing in the Current Models?
Dr. Kumar said the models, which he acknowledged weren’t created for this exact purpose, likely break down in the case of joint replacement because they don’t account for patients’ functional status or other relevant health conditions.
Functional status includes measures of:
- Postoperative pain
- Ability to move the affected joint
- Ability to perform activities of daily living
Medicare does not require hospitals to report functional status, but in a study earlier this year, Dr. Kumar was able to obtain inpatient rehabilitation data for patients who had strokes or hip fractures, as well as data for some joint replacement patients. He and his co-authors found that adding functional status data into a predictive risk model yielded a substantial improvement.
“The reason we do joint replacements is to reduce pain and improve functional status, but this information is missing from our risk indices,” Dr. Kumar said.
In the current study, the researchers were able to assess other relevant health conditions. The following health conditions were most frequently associated with hospital readmission:
- Pulmonary disease
- Heart disease
In addition, prior research suggests obesity is likely an important determinant, although that was not tracked in the study.
In the near term, Dr. Kumar said, CMS should begin tracking functional status of patients who undergo joint replacements. Ultimately, he said, that data should be evaluated in a new index that will help hospitals assess which patients are at greatest risk to struggle and will help CMS assess which hospitals are taking on riskier patients.