The American Journal of Managed Care

Learning About 30-Day Readmissions From Patients With Repeated Hospitalizations | Page 2

Published Online: June 27, 2014
Jeanne T. Black, PhD, MBA
Patients with Medicaid coverage alone made up 27.6% of the very frequent readmission group, more than double the proportion among the frequent readmission subgroup (13.4%), while only 10.0% of the comparison patients with 1 or 2 hospital stays had Medicaid coverage. The proportion of dual-eligible patients also was significantly higher in the frequent readmissions subgroups. The geographic distribution of patients with frequent readmissions was generally similar to that of the comparison patients. There was a trend toward a larger proportion of those with 3 to 5 hospitalizations living in the area immediately surrounding the hospital, while a smaller percentage of those in the very frequent readmission subgroup lived in this core service area. A significantly higher proportion of patients in the very frequent readmission subgroup did not live within the hospital’s primary service area, but further away within the county.

Just 2.4% of the total cohort died during the index stay. The highest proportion of in-hospital deaths was found in the frequent readmissions subgroup (7.7%), significantly higher than the 6.9% of patients with 6 or more stays who died. Within the comparison group, 3.4% of patients with a single hospital stay died, while 6.1% of those with 1 rehospitalization died during that second stay. Only 11.7% of the index stays in the comparison subgroup had principal diagnoses associated with any of the 10 chronic conditions. A larger percentage of patients in the 2 frequent readmission subgroups had an index hospital stay for all but 1 of these conditions; the exception was coronary artery disease (CAD). For example, 93.6% of patients with an index stay for CAD were in the subgroup with 1 or 2 inpatient stays in 6 months, whereas 76.3% of patients with a chronic renal failure index stay and 60.3% of patients with a sickle cell anemia index were in this subgroup. Nevertheless, only 26.9% of the patients with very frequent readmissions were categorized as having any chronic condition as the principal diagnosis for their index stay. Using the principal diagnosis of all of the hospitalizations resulted in an even smaller proportion associated with chronic conditions, although this approach showed that sickle cell anemia and HIV/AIDS were among the most common principal diagnoses for patients with very frequent readmissions (5.0% and 4.0%, respectively), in addition to malignancies with poor prognosis (4.0%) and heart failure (3.9%). For patients in the frequent readmissions subgroup (3 to 5 stays in 6 months), the most prevalent principal diagnoses were heart failure (6.7%), malignancies with poor prognosis (5.2%), and chronic pulmonary disease (2.3%). Finally, categorizing patients according to the principal diagnosis of any stay (which allowed patients to be counted more than once) resulted in 55.2% of patients with very frequent readmissions and 42.5% of those with frequent readmissions being associated with a chronic condition, versus only 13.7% of comparison group patients.

Although heart failure was the most prevalent chronic condition in the cohort, the 726 patients whose index stay had a principal diagnosis of heart failure comprised only 3.8% of all cohort patients. As shown in Table 3, 19.0% of these patients had 3 or more hospital stays in 6 months. However, an additional 216 patients (1.1% of the cohort) had an index stay for some other diagnosis and subsequently had a heart failure stay. A much larger proportion of these patients (57.9%) had 3 or more hospitalizations. For both groups of patients with at least 1 hospitalization for heart failure, the table shows that as the number of total hospitalizations increased, the proportion attributable primarily to heart failure decreased.

Examination of secondary diagnoses also revealed significant differences among the subgroups. The overall prevalence of mental health diagnoses coded from the medical record was 21.0%, but 43.5% of patients with very frequent readmissions had such diagnoses, compared with 34.7% of those with 3 to 5 hospital stays and 19.3% of those with 1 or 2 stays. The corresponding prevalence of substance abuse diagnoses was 27.6%, 9.9%, and 6.3%. Patients with very frequent readmissions also had a significantly higher prevalence of documented morbid obesity (9.0%) compared with the frequent readmission subgroup (6.0%) and the comparison subgroup (3.3%). A different pattern was observed in the prevalence of documented tobacco-use disorder, which was 6.0% in the comparison subgroup, 4.4% in the subgroup with 3 to 5 hospital stays, and 2.8% in the subgroup with very frequent readmissions. This finding was associated with the fact that a higher proportion of patients with an index stay for CAD had a documented diagnosis of tobacco-use disorder than did patients with other index diagnoses, and nearly all of the CAD patients were in the comparison subgroup. The proportion of patients documented as lacking housing using a diagnosis code was too small to report.

Patients in the frequent and very frequent readmissions subgroups also had a higher proportion of 30-day readmissions following their index stay. Table 4 shows the rate of readmission within 30 days after the index stay for all patients in the cohort who were discharged alive from their index stay, as well as for patients whose index admission diagnoses were heart failure, chronic pulmonary disease, or not attributable to any chronic condition. Although patients with only 1 or 2 hospital stays did have a higher 30-day readmission rate when their index stay was for heart failure, the 30-day readmission rate among patients with frequent readmissions was similar for patients with heart failure and for those whose index stay was not primarily associated with any chronic condition.


This analysis affirmed the common finding that a small proportion of patients accounts for a disproportionate share of resource use: in our analysis, 10% of a cohort of patients initially hospitalized with a medical diagnosis incurred 72% of the 30-day readmissions that occurred within 180 days of the initial discharge. Of the 3 conditions currently targeted by the HRRP, heart failure has received the most attention. Given the intense national focus on reducing 30-day readmissions among patients with heart failure, one might expect them to comprise a large share of such rehospitalizations. This is not the case. Jencks’ analysis of Medicare FFS claims in 2003-2004 showed that patients with an index DRG of heart failure accounted for 7.6% of 30-day readmissions.2 Our cohort included all adult patients, not just those with Medicare coverage, but the relative magnitude of the population with heart failure was consistent with Jencks’ findings. Heart failure was the most prevalent chronic condition in our cohort, but the more notable finding was that 87.3% of index hospital stays could not be categorized as any chronic condition based on the patient’s principal diagnosis. Although it had seemed reasonable to assume that most nonsurgical patients with frequent hospitalizations were suffering from a chronic condition, the principal diagnosis code for the inpatient stay did not perform well in identifying these conditions. Even when the principal diagnosis for every hospitalization was included, about half the patients with 3 or more hospitalizations remained uncategorized. Is it plausible that the hospitalizations of patients with multiple inpatient stays in 6 months were not associated with any major chronic condition? Probably not. More likely, this finding illustrates that the principal diagnosis obtained from administrative data reflects the acute manifestations and complications—or the side effects of treatment—that led to the hospitalization, rather than the underlying disease itself. This, in turn, results from a coding system based on individual body systems that is applied for reimbursement purposes.

This analysis has several implications for efforts to reduce readmissions. Overall, it suggests that intervention strategies should take into account patients’ readmission histories. Attempts to target patients by condition using inpatient MS-DRGs or ICD-9 codes are unlikely to be successful. Numerous efforts have been made to develop an algorithm to predict which patients are at highest risk of a single 30-day readmission. Most of the resulting models have had rather poor discriminative ability and have not been able to generate predictions in real time.15 This may be due in part to the heterogeneity of patients categorized on the basis of a single index discharge, as illustrated by this analysis. The Hospital Compare rate for 30-day readmissions following a heart failure discharge at the medical center combines patients with few readmissions, the majority of which are for heart failure, and patients with both multiple conditions and multiple hospitalizations. This analysis does suggest a simple way to identify which patients are at greatest risk of multiple 30-day readmissions: those who have already had 2 or more hospitalizations in the previous 6 months. The real challenge is not in predicting which patients are at highest risk, but in identifying which interventions are likely to be most effective for specific patients.

PDF is available on the last page.

Issue: June 2014
More on AJMC.COM