Hospital Readmission Risk Prediction Models Work Poorly

Most models currently used to predict hospital readmission risk perform poorly, and better approaches are needed as policy makers increasingly use hospital readmission rates to calculate and publicize quality of care comparison information.

Twenty-six different methods to calculate readmission risk were reviewed, yet only one model specifically addressed preventable readmissions, according to the study by Dr. Devan Kansagara of the Portland (Ore.) Veterans Affairs Medical Center and colleagues published in the Oct. 16 issue of JAMA. Meanwhile, most models performed poorly when used to predict readmission rates, regardless of whether they were developed to compare hospitals or to improve quality of care, the researchers said.

Still, several of these models are being used in clinical settings, in research projects, and for policy making. "Readmission risk prediction remains a poorly understood and complex endeavor," Dr. Kansagara said. "Although in certain settings such models may prove useful, better approaches are needed to assess hospital performance in discharging patients, as well as to identify patients at greater risk of avoidable readmission."

The analysis broke the models down into three groups: models relying on retrospective administrative data, models using real-time administrative data, and models incorporating primary data collection (JAMA 2001;306:1688-98).

In the group relying on retrospective administrative data, most of the 14 models studied included variables for medical comorbidity and use of prior medical services, while a few models also considered mental health, functional status, and other social variables, all of which may be important when determining readmission risk. All performed poorly except in specific subsets of patients; for example, one model was able to predict some readmissions in asthma patients, the researchers noted.

Three models attempted to identify high-risk patients during their initial hospitalizations in an effort to target them for interventions that might be able to prevent readmissions. Two of these – including a model implemented in one urban U.S. hospital to predict readmissions for heart failure – worked modestly well, especially in certain populations, but none had excellent predictive ability overall, the study said.

Finally, in the group of nine models that incorporated primary data collection, hospitals used questionnaires and other data in an effort to predict potential readmission risks early in patients’ initial hospital stays. Although several of these models had some predictive value, according to the study, none are in wide use, and a couple were developed more than 20 years ago.

Most of the models examined in the study included data on medical comorbidity, but few considered variables associated with illness severity, overall health and function, and socioeconomic factors that can affect a patient’s health, according to the study.

Public reporting of readmission rates, coupled with financial penalties for hospitals with high 30-day readmission rates, both are spurring organizations to implement quality improvement programs, Dr. Kansagara said. However, since the current available models for predicting readmission risk don’t work well, it may not be fair to use them to compare hospitals and possibly penalize them.

"Use of readmission rates as a quality metric assumes that readmissions are related to poor quality care and are potentially preventable," Dr. Kansagara said. "However, the preventability of readmissions remains unclear and understudied. We found only one validated prediction model that explicitly examined potentially preventable readmissions as an outcome, and it found that only about one-quarter of readmissions were clearly preventable."

The researchers noted that hospital and health system–level factors likely contribute to readmission risk, but are not included in any current models used to calculate readmission risk. For example, coordination with the patient’s primary care physician, plus the timing and frequency of postdischarge follow-up visits, can help determine if a patient will be readmitted or not, they said.

No conflicts of interest were reported. The study was supported by funding from the Department of Veterans Affairs and the National Institutes of Health.

Most models currently used to predict hospital readmission risk perform poorly, and better approaches are needed as policy makers increasingly use hospital readmission rates to calculate and publicize quality of care comparison information.

Twenty-six different methods to calculate readmission risk were reviewed, yet only one model specifically addressed preventable readmissions, according to the study by Dr. Devan Kansagara of the Portland (Ore.) Veterans Affairs Medical Center and colleagues published in the Oct. 16 issue of JAMA. Meanwhile, most models performed poorly when used to predict readmission rates, regardless of whether they were developed to compare hospitals or to improve quality of care, the researchers said.

Still, several of these models are being used in clinical settings, in research projects, and for policy making. "Readmission risk prediction remains a poorly understood and complex endeavor," Dr. Kansagara said. "Although in certain settings such models may prove useful, better approaches are needed to assess hospital performance in discharging patients, as well as to identify patients at greater risk of avoidable readmission."

The analysis broke the models down into three groups: models relying on retrospective administrative data, models using real-time administrative data, and models incorporating primary data collection (JAMA 2001;306:1688-98).

In the group relying on retrospective administrative data, most of the 14 models studied included variables for medical comorbidity and use of prior medical services, while a few models also considered mental health, functional status, and other social variables, all of which may be important when determining readmission risk. All performed poorly except in specific subsets of patients; for example, one model was able to predict some readmissions in asthma patients, the researchers noted.

Three models attempted to identify high-risk patients during their initial hospitalizations in an effort to target them for interventions that might be able to prevent readmissions. Two of these – including a model implemented in one urban U.S. hospital to predict readmissions for heart failure – worked modestly well, especially in certain populations, but none had excellent predictive ability overall, the study said.

Finally, in the group of nine models that incorporated primary data collection, hospitals used questionnaires and other data in an effort to predict potential readmission risks early in patients’ initial hospital stays. Although several of these models had some predictive value, according to the study, none are in wide use, and a couple were developed more than 20 years ago.

Most of the models examined in the study included data on medical comorbidity, but few considered variables associated with illness severity, overall health and function, and socioeconomic factors that can affect a patient’s health, according to the study.

Public reporting of readmission rates, coupled with financial penalties for hospitals with high 30-day readmission rates, both are spurring organizations to implement quality improvement programs, Dr. Kansagara said. However, since the current available models for predicting readmission risk don’t work well, it may not be fair to use them to compare hospitals and possibly penalize them.

"Use of readmission rates as a quality metric assumes that readmissions are related to poor quality care and are potentially preventable," Dr. Kansagara said. "However, the preventability of readmissions remains unclear and understudied. We found only one validated prediction model that explicitly examined potentially preventable readmissions as an outcome, and it found that only about one-quarter of readmissions were clearly preventable."

The researchers noted that hospital and health system–level factors likely contribute to readmission risk, but are not included in any current models used to calculate readmission risk. For example, coordination with the patient’s primary care physician, plus the timing and frequency of postdischarge follow-up visits, can help determine if a patient will be readmitted or not, they said.

No conflicts of interest were reported. The study was supported by funding from the Department of Veterans Affairs and the National Institutes of Health.

Most models currently used to predict hospital readmission risk perform poorly, and better approaches are needed as policy makers increasingly use hospital readmission rates to calculate and publicize quality of care comparison information.

Twenty-six different methods to calculate readmission risk were reviewed, yet only one model specifically addressed preventable readmissions, according to the study by Dr. Devan Kansagara of the Portland (Ore.) Veterans Affairs Medical Center and colleagues published in the Oct. 16 issue of JAMA. Meanwhile, most models performed poorly when used to predict readmission rates, regardless of whether they were developed to compare hospitals or to improve quality of care, the researchers said.

Still, several of these models are being used in clinical settings, in research projects, and for policy making. "Readmission risk prediction remains a poorly understood and complex endeavor," Dr. Kansagara said. "Although in certain settings such models may prove useful, better approaches are needed to assess hospital performance in discharging patients, as well as to identify patients at greater risk of avoidable readmission."

The analysis broke the models down into three groups: models relying on retrospective administrative data, models using real-time administrative data, and models incorporating primary data collection (JAMA 2001;306:1688-98).

In the group relying on retrospective administrative data, most of the 14 models studied included variables for medical comorbidity and use of prior medical services, while a few models also considered mental health, functional status, and other social variables, all of which may be important when determining readmission risk. All performed poorly except in specific subsets of patients; for example, one model was able to predict some readmissions in asthma patients, the researchers noted.

Three models attempted to identify high-risk patients during their initial hospitalizations in an effort to target them for interventions that might be able to prevent readmissions. Two of these – including a model implemented in one urban U.S. hospital to predict readmissions for heart failure – worked modestly well, especially in certain populations, but none had excellent predictive ability overall, the study said.

Finally, in the group of nine models that incorporated primary data collection, hospitals used questionnaires and other data in an effort to predict potential readmission risks early in patients’ initial hospital stays. Although several of these models had some predictive value, according to the study, none are in wide use, and a couple were developed more than 20 years ago.

Most of the models examined in the study included data on medical comorbidity, but few considered variables associated with illness severity, overall health and function, and socioeconomic factors that can affect a patient’s health, according to the study.

Public reporting of readmission rates, coupled with financial penalties for hospitals with high 30-day readmission rates, both are spurring organizations to implement quality improvement programs, Dr. Kansagara said. However, since the current available models for predicting readmission risk don’t work well, it may not be fair to use them to compare hospitals and possibly penalize them.

"Use of readmission rates as a quality metric assumes that readmissions are related to poor quality care and are potentially preventable," Dr. Kansagara said. "However, the preventability of readmissions remains unclear and understudied. We found only one validated prediction model that explicitly examined potentially preventable readmissions as an outcome, and it found that only about one-quarter of readmissions were clearly preventable."

The researchers noted that hospital and health system–level factors likely contribute to readmission risk, but are not included in any current models used to calculate readmission risk. For example, coordination with the patient’s primary care physician, plus the timing and frequency of postdischarge follow-up visits, can help determine if a patient will be readmitted or not, they said.

No conflicts of interest were reported. The study was supported by funding from the Department of Veterans Affairs and the National Institutes of Health.