A Scoring System to Predict Hospital Readmissions

Knowing, based on this paper, that the readmission rate to U.S. hospitals is as high as 20%, you and your colleagues decide to implement a readmissions prevention program.”  Your state-of-the-art plan includes evidence-based interventions such as frequent telephone calls, nurse home visits, telemonitoring, referral to community programs and close coordination with the outpatient physicians.

Your problem, however, is that the “reach” of your program is limited.  With only a limited budget with a limited number of nurses, you can’t afford to call, visit, telemonitor, refer and coordinate every patient discharge. 

You wish you could focus on the highest risk patients.

Jacques Donzé et al to the rescue.

As the Disease Management Care Blog understands it, this team of researchers retroactively looked at one year’s worth of medical service discharges from Boston’s Brigham & Women’s by dividing them into 3 groups: 1) no readmission within 30 days, 2) an “unavoidable” readmission within 30 days (for a new unrelated condition or a planned return to the hospital, like another round of chemotherapy for cancer) and 3) an “avoidable” readmission readmission.  The initial sorting was done using a computer algorithm followed by a chart-review that confirmed the sorting.

Then the researchers discarded the “unavoidable readmission group” and compared the “avoidables” to the “no readmission” group.  Logistic regression, based on a total of 9212 patients, was used to find the independent “signals” that were statistically and independenly associated with the avoidable readmission group: in other words, what features did they have that the no-readmission group didn’t have?

Some features had a stronger “signal” and therefore warranted a greater weight, which was reflected in a point scoring system. The authors cleverly dubbed it the HOSPITAL Score:

Low hemoglobin level at discharge (less that 12 g/dL) …1 point (H)

Discharge from an oncology service… 2 points (O)

Low sodium level at discharge (135 mEq/L)… 1 point (S)

Procedure during hospital stay (any ICD-9-CM coded procedure)… 1 point (P)

Index admission type: nonelective… 1 point (I)

No. of hospital admissions during the previous year 0… 0 points, 1-5… 2 points, >5.. 5 points (A)

Length of stay >5 days …2 points (L)

Then the point scores were arbitrarily stratified into 3 groups.  If the point score added up to 4 or less, that was a “low” risk group, while 5-6 points was intermediate risk group and 7 or more points was high risk.

If your score was low: you had a 5.2% chance of a 30 day readmission

Intermediate: 9.8% chance of a 30 day readmission

High risk: 18.3% chance of a 30 day readmission

The DMCB’s take:

1. This was a Boston academic medical system with a high readmission rate of 22%.  The results may not be applicable to settings such as this with a readmission rate of 8%.

2. There is no information on  the “planned-unavoidable” readmissions; the DMCB doesn’t know how the HOSPITAL score works on predicting readmissions for an unrelated condition.

3. The study is limited to “medical service” readmissions.  There is no information on the use of this scoring system for patients being discharged after surgery.

4. Keep in mind that Medicare’s readmissions program is based on patients with heart attack, heart failure and pneumonia.  While Medicare patients with those diagnoses were included in this study, this research didn’t focus on those particular conditions in Medicare. That means the DMCB doesn’t know if HOSPITAL will adequately predict readmissions in this key payor group.

5. It’s counter-intuitive, but some of the “signals” are associated with readmissions don’t necessarily cause them.  A casual observer might think that correction of anemia or a low blood sodium level would lead to a lower rate of readmissions. Not so. Rather, anemia of chronic disease and a low sodium level have been known for years to happen in chronically sick fragile patients.  It’s the fragility, not the labs.

6. This shows what readmission prevention programs are up against. Among the high risk patients, the algorithm only correctly spots 18%.  So, if you commit a nurse case manager to go after all patients with a score of 7 points or more, 80% are destined to not be readmitted anyway.

7. That being said, this is an evidence-based study that represents and important advance in indentifying patients at high risk for readmission, using a simple point system for information that is typically available at the time of discharge.

The DMCB likes it.