Hospital Observation Status and Readmission Rates

We conducted a retrospective cohort study of hospitalizations from 50 tertiary children’s hospitals (representing 28 US states and all census divisions) participating in the Pediatric Health Information System (PHIS) (Children’s Hospital Association, Lenexa, KS) from January 1, 2018, to December 31, 2018. PHIS is an administrative and billing database containing information on all inpatient, observation, ambulatory surgery, and emergency department encounters at participating hospitals. Importantly, an observation-status encounter in PHIS indicates that the encounter was considered an outpatient encounter by the payer and payed at an observational rate but is not necessarily reflective of separate observational care. The use of a unique, consistently encrypted medical record allows patients to be tracked across encounters within the same hospital. We considered all hospitalizations as index events for a potential readmission. For each encounter in PHIS, we collected demographic characteristics; International Classification of Diseases, 10th Revision, Clinical Modification diagnosis and procedure codes; and daily detailed billing information. This study was reviewed by the Institutional Review Board at Children’s Mercy Hospital and considered to be non–human subject research.

We used the pediatric all-cause readmission (PACR) metric endorsed by the National Quality Forum at 30 days postindex encounter.¹⁶  In short, the algorithm excludes some index encounters (eg, mental health), removes readmissions considered to be likely planned, and risk adjusts hospital readmission rates by using age, sex, and the existence of some chronic conditions.

We included age at admission, sex, race, payer, and presence of a complex chronic condition.¹⁷  Length of stay was measured in hours, and the hospital-level case-mix index was assessed with the Hospitalization Resource Intensity Scores for Kids method.¹⁸ 

We compared demographic and clinical characteristics of inpatient and observation encounters using χ² or Wilcoxon rank tests, as appropriate. The PACR algorithm was run independently by using inpatient encounters only and then again by using inpatient and observation encounters. We calculated, ranked, and put into quintiles each hospital’s unadjusted and adjusted 30-day readmission rates. We determined the change in adjusted readmission rates and quintiles for each hospital when considering inpatient-only encounters and when including observation encounters.

To understand how observation status relates to readmission metric performance, we first described the percentage of observation encounters (of the total numbers of inpatient and observation encounters) for each hospital. We then examined the association between this percentage of observation encounters and the inpatient-only case-mix index, which serves as a marker of illness severity, using linear regression. Similarly, we used linear regression to determine the relationship between the percentage of observation encounters and the change in the hospital’s adjusted readmission rates with and without inclusion of observation encounters. We then ranked hospitals by adjusted 30-day all-condition readmission rate and grouped hospitals by quintiles, first including only inpatient-status hospitalizations and second including both inpatient and observation-status hospitalizations.

All statistical analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC), and P values <.05 were considered statistically significant.

There were 728 104 index hospitalizations, of which 514 226 (70.6%) hospitalizations were designated inpatient status and 213 878 (29.4%) were observation status. There were 44 433 observation and inpatient all-cause 30-day readmissions. A total of 44 of the 50 hospitals (88%) in the sample included at least 1 index hospitalization using observation status. The index hospitalization’s median duration was 25 hours (interquartile range [IQR]: 19–34) for observation status and 63 hours (IQR: 39–119) for inpatient status (Table 1). There was significant variation in the proportion of index hospitalizations designated as observation status across hospitals (Fig 1).

Compared to inpatient stays, observation-status index hospitalizations were less common for infants (18.4% vs 37.2%; P < .001) and children with complex chronic conditions (21.1% vs 40.3%; P < .001) (Table 1). In bivariate analysis, non-Hispanic white and male subjects were more likely to be observation-status index hospitalizations than inpatient-status index encounters. As noted by the increase in inpatient-only case-mix index (P < .001), the percentage of observation stays directly correlated with the severity of inpatient hospitalizations at the hospital level (Fig 2).

The 30-day all-condition readmission rate, when including all hospitalizations, was 6.1% (Table 1). The readmission rate was higher after an index hospitalization designated as inpatient status (5.5% readmitted to inpatient; 1.3% readmitted to observation), as compared to after an index hospitalization designated as observation status (1.9% readmitted to inpatient; 2.5% readmitted to observation) (P < .001). Readmissions often maintained the same inpatient or observation designation used in the index hospitalization, although this was more pronounced for index inpatient-status hospitalizations (80.9% of readmissions maintained inpatient status) compared to index observation-status hospitalizations (56.8% of readmissions maintained observation status).

Compared to hospital rankings calculated including only inpatient-status index hospitalizations, 60% (n = 30) of hospitals experienced movement of at least 1 quintile and 20% (n = 9) of hospitals moved ≥2 quintiles when both inpatient- and observation-status index hospitalizations were included (Fig 3). There was a statistically significant positive relationship between the percentage of observation-status index hospitalizations and the change in the adjusted inpatient-only readmission rate at the hospital (P < .001) (Fig 4). For example, for every 10% increase in the percentage of observation-status index hospitalizations at the hospital, the hospital’s adjusted readmission rate for inpatient-only index encounters increased by an average of 0.3% (range: 0.1–0.5).

In our study, we provide new information about the impact of excluding observation-status stays from traditional pediatric readmission rate calculations. Including observation stays in readmission rate calculations significantly changed hospital readmission ranking, with 60% of the hospitals moving at least 1 quintile. The impact of observation status on these metrics and the composition of patients included in hospital data are profound, although, currently, these encounters are excluded from most readmission rate analyses. Policy makers and payers setting readmission rate penalties that are based on national data are encouraged to include observation-status stays to ensure more standardized, equitable, and meaningful hospital comparisons.

Our study agrees with previous literature suggesting that observation status is applied variably across children’s hospitals.¹²  This variation is likely driven by regional payer contracts, local practice patterns, and state laws. Because of the differential application of observation across hospitals, it is difficult to accurately define the designation in a clinically meaningful way. The PHIS hospitals represented in the current study are tertiary children’s medical centers treating a wide but similar range of pediatric health conditions. It is likely that hospitals billing a lower percentage of observation cases may have more short-stay, lower-acuity encounters designated as inpatient status, whereas similar cases might be designated as observation status in greater numbers at other institutions. This variability could significantly impact comparisons of hospitals on a wide array of metrics beyond readmission rates. However, we are not aware of other large, multihospital analyses published to date in which authors assess the impact of observation status on other metrics, making this a fruitful topic for future study.

This issue is particularly pressing when considering large administrative data sets, such as the Nationwide Readmission Database, which does not include observation stays.¹⁹  Variable use of observation status results in arbitrary inclusion or exclusion of hospital encounters from these databases based on an administrative label, which is applied to encounters on the basis of inconsistent interpretation of third-party criteria (Milliman and InterQual). Including all hospital encounters (inpatient and observation) would allow for consistent comparison across hospitals for benchmarking and population analyses.

Our findings are also potentially problematic for the financial health of children’s hospitals because inpatient-to-inpatient readmission rates may be used by payer and state Medicaid pay-for-performance programs and/or penalties. Hospitals with high observation use effectively exclude lower-acuity patients from their overall inpatient population. Siphoning off lower-acuity patients into observation status explains our finding that hospitals with high observation use have higher inpatient case-mix index and higher inpatient-to-inpatient readmission rates. Conversely, hospitals with low observation use may have comparatively lower inpatient-to-inpatient readmission rates because lower-acuity patients are being included in the inpatient population and subsequent calculations instead of being labeled as observation.

Readmission metrics as indicators of quality may possess some value; however, the differential use of observation status across hospitals alters inpatient populations, making comparisons challenging and casting doubt on the validity of readmission rate comparisons as the basis for financial penalties. Most readmission metrics use some form of risk adjustment to help account for these differences; however, claims-based risk-adjustment models cannot adjust for the exclusion of observation encounters. In addition, we found that the risk adjustment for a particular hospital is sensitive to whether observation-status patients from other hospitals are included in the model. This may lead to a change in the hospital’s relative performance as a result of whether, or how much, observation status is used by their peers. Given the high financial stakes for many children’s hospitals, more universal standardization of readmission metrics will provide hospitals with opportunities for targeted improvements and more accurate comparisons with peer institutions.

Targeted interventions to lower readmission rates may not only improve hospital finances but also reduce the harm that readmissions have on vulnerable patient populations. Although all-cause readmissions are relatively rare after an index hospitalization (4%–6%),^20,21  the impact on individual patients and their families can be profound. The emotional, social, and financial consequences of an unexpected readmission have been shown to be severe and potentially long lasting.^22–24  In recent years, there has been national focus on identifying and reducing unnecessary readmissions in an effort to reduce the harm of these encounters.^25–27  Including observation-status patients in readmission metrics will optimize our understanding of hospital readmissions and promote health equity by avoiding exclusion of certain patient populations on the basis of billing designation alone.

Our study has several limitations. First, the PHIS database includes only tertiary children’s hospitals. Although our results are not generalizable to all hospitals that care for children, they do allow for a large-scale relative comparison of the impact of admission status designation. Second, although our results indicate that observation-status stays affect readmission metrics, there may be variables not captured in this study that may also impact these metrics. We accounted for these effects using the PACR algorithm, which excludes planned returns and adjusts for patient-level factors, including age, sex, and the existence of chronic conditions. Third, inpatient versus observation designations are generally finalized at discharge; however, after discharge, payers may dispute the designation assigned, and, as a result, inpatient stays may be downgraded to observation status. PHIS captures the hospitalization designation assigned at discharge and does not account for changes that occur after payer use review. Finally, we did not evaluate other clinical quality indicators (ie, central line–associated bloodstream infections) to determine the effects of administrative designations on the rates of these adverse events. Further work is needed to determine if these metrics are also impacted by billing status.

Including observation-status encounters in readmission metrics changes readmission rates and subsequent hospital rankings. There is great variability in the use of observation status among children’s hospitals, and observation encounters make up a substantial proportion of patient encounters at children’s hospitals contributing to the PHIS. Their exclusion alters the composition of the population of hospitalized patients under evaluation and makes national studies and comparison of hospitals difficult. We recommend including observation stays when calculating readmission metrics to ensure valid hospital-to-hospital comparisons.

IQR

interquartile range

PACR

pediatric all-cause readmission

PHIS

Pediatric Health Information System