Observation Encounters and Length of Stay Benchmarking in Children’s Hospitals

We conducted a retrospective cohort analysis of all-condition hospitalizations for patients 0 to 18 years of age occurring between January 1, 2017, and December 31, 2017, in the Pediatric Health Information System (PHIS) database (Children’s Hospital Association, Lenexa, KS). This data set contains clinical and billing data from 52 children’s hospitals in the United States, representing 17 of the 20 major metropolitan areas across the United States and accounting for ∼20% of all US discharges for children. We included all admissions designated as IP and OBS, excluding normal newborns, patients who were transferred to or from other acute care facilities, and deaths. Hospitals not reporting LOS in hours were also excluded (n = 6). A single PHIS institution had both OBS and IP hospitalizations but chose to submit only IP data. Because the data were incomplete, this hospital was also excluded from analysis. Children's Hospital Association conducts a survey of member hospitals every 2 years to determine how they assign IP and OBS patient types in PHIS. This is based on each hospital’s use of InterQual (68.6% hospitals)¹⁸  or Milliman (31.4% hospitals)¹⁹  criteria, and as a result some hospitalizations are assigned an “outpatient” (ie, OBS) status on their UB-04 form for billing purposes.

Demographic variables included age, sex, race, payer source, rurality, and Clinical Risk Group (CRG). CRG software, developed by 3M Health Information Systems (Salt Lake City, UT),²⁶  assigns individual patients to single, mutually exclusive hierarchical condition status groups and offer a discreet categorization of chronic conditions by a hierarchy of complexity. We used CRGs to categorize patients in hospital discharge data stratifying this population as follows: (1) nonchronic condition, (2) episodic chronic conditions, (3) single lifelong chronic conditions, (4) chronic conditions in >1 body system, and (5) malignancies. Patient severity of illness was assessed as a case-mix index, calculated as the average relative weight for All Patient Refined DRGs and severity of illness levels using the Hospital Resource Intensity Scores for Kids system.²⁷ 

Risk-adjusted geometric mean length of stay (RA-LOS) for IP only and IP plus OBS was calculated for each hospital and ranked by quintile. Quintile rank comparisons were then assessed for each hospital for RA-LOS calculations with IP alone and IP plus OBS.

We summarized characteristics of the population using frequencies and percentages for categorical variables and geometric mean and SD for continuous variables. Comparisons between IP and OBS used χ² and Wilcoxon rank-sum tests as appropriate. Risk-adjustment models were built using generalized linear mixed effects models with random hospital intercepts and a log-transformed LOS in hours to account for its non-normal distribution. Models were adjusted for demographic and clinical characteristics listed above. Estimates were back-transformed to present the results on the original hours scale. Statistical analyses were performed using SAS v.9.4 (SAS Institute, Inc, Cary, NC), with a significance threshold of P < .001 because of the large sample size. This study was considered exempt from human subjects review by the Institutional Review Board of Vanderbilt University.

For calendar year 2017, 45 hospitals and 625 032 hospitalizations met the inclusion criteria (IP = 410 731 [65.7%], OBS = 214 301 [34.3%]) (Fig 1). Among PHIS hospitals, the percentage of total discharges that were OBS ranged from 0.0% to 60.3% (Fig 2). Six hospitals in the data set had 0.0% OBS, whereas in 6 hospitals, OBS accounted for >50% of all hospitalizations.

The RA-LOS (SD) was 75.2 (2.6) hours for IP vs 54.3 (2.7) hours for IP plus OBS combined (P < .001). RA-LOS (SD) for OBS alone was 25.0 (1.7) hours. There were significant differences in age, race, payer, chronic disease incidence, and case-mix index between IP and OBS patients (P < .001 for all comparisons, Table 1).

When we calculated RA-LOS for IP admissions only, there was a significant relationship between RA-LOS and the percentage of OBS such that the higher the percentage of OBS reported by hospital, the higher the RA-LOS for IP admissions (Fig 3, P = .002). For all hospitals that reported >0% OBS stays, the addition of OBS to IP LOS calculations resulted in a decrease in RA-LOS compared with IP admissions alone (Fig 4). The ratio of the calculated LOS for IP only compared with IP plus OBS averaged 0.69 for all hospitals (range: 0.51–0.93). For the 10 hospitals with the highest percentage of OBS stays, the calculated LOS including IP plus OBS averaged only 57.1% (range: 51.0%–65.7%) of the value using IP encounters alone. Overall, this difference amounted to an average calculated LOS of 24.6 hours shorter when OBS stays were included. Furthermore, hospitals reporting 0% OBS stays also saw a slight decrease in calculated RA-LOS after risk adjustment.

The addition of OBS in the calculations of RA-LOS also had a significant impact on the ranking of hospitals by LOS. For 34 of 45 (75.6%) hospitals, including IP plus OBS resulted in a change of ≥1 quintile rank compared with RA-LOS for IP alone, whereas 11 of 45 (24.4%) hospitals saw no change in their quintile rank (Table 2). Of those with a change in quintile rank, 16 hospitals increased in rank, whereas for 18, the ranking decreased.

Finally, the marked variation in percentage of OBS among study hospitals prompted us to look for clinically relevant measures to assess the similarity in patients who were most likely to meet OBS criteria. Given that a major feature of the definition of OBS is a LOS of <48 hours,^16,23  we assessed the percentage of hospitalizations with a LOS <48 hours among hospitals in the lowest and highest quintiles in the cohort (n = 9 for each). In hospitals in the lowest quintile, 46.5% of the hospitalizations had a LOS <48 hours, whereas in the highest quintile, 60.6% of hospitalizations were <48 hours. Although we observed a difference in the percentage of hospitalizations <48 hours in these quintiles, the difference in application of OBS stays was far greater (3.5% vs 51.5%, respectively). This provides strong support that clinical and demographic factors alone cannot account for the variation in OBS usage among the hospitals in the sample.

In this study, we provide new data on the impact of OBS on calculations of RA-LOS in children’s hospitals. There was marked variation in the percentage of stays designated as OBS in study hospitals, and all hospitals saw a decrease in RA-LOS when OBS was included in the calculation, with 75.6% of hospitals having a ≥1 quintile change in rank by RA-LOS. Furthermore, hospitals with a higher percentage of OBS had longer RA-LOS when only IP admissions were considered, and conversely, a lower percentage of OBS stays was associated with a shorter IP-only RA-LOS. Finally, even hospitals reporting 0% OBS stays had significant proportions of hospitalizations with LOS <48 hours, stays most likely to have a high percentage of OBS in hospitals which use that billing status designation. With these findings, we suggest that OBS stays in some hospitals are similar to short stays billed as IP in others, and including OBS in LOS calculations can have substantial impact on a hospital’s performance as gauged by the LOS measure.

Variation in the percentage of hospitalizations designated as OBS has been reported previously^22,23  although there was little reported difference in resource use between stays designated as OBS and other short-stay (1–2 day) encounters that had a billing status as IP.²³  Many of the large administrative data sets made available by the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project are derived from administrative billing data submitted by state hospital associations. Because many states do not collect data on all patients receiving OBS services and because of the inconsistent use and reporting of OBS, Healthcare Cost and Utilization Project has recommended against including OBS in its hospital data sets.²⁸  Currently, neither the Kids’ Inpatient Database nor the National Readmissions Database include encounters billed as OBS ^24,29  and therefore underrepresent the hospital-based care that is delivered to children in the United States.

The case has been made to abolish OBS as a hospital patient billing status and replace it with a low-acuity DRG equivalent.³⁰  On the basis of our findings, we recommend a national standard definition of hospital LOS that encompasses the care provided to all patients after an OBS or IP order is written from emergency department, clinic, or inpatient ward and includes the care delivered to all patients occupying inpatient hospital facilities and processes (eg, placement in inpatient ward rooms, care by inpatient nursing and medical providers). Such a standard “resource LOS” would allow hospitals a better assessment of the total cost of care they provide beyond the emergency department and admitting office and also would allow more accurate benchmarking with other hospitals.

There are several potential explanations for the variability in the percentage of OBS in children’s hospitals. Whereas Medicare guidelines for determination of OBS apply across all 50 states, state Medicaid agencies and private insurance companies may vary in their definitions of OBS and its application as a status for hospital patients. For example, California’s Medicaid program (Medi-Cal) does not recognize OBS as a billing status,³¹  whereas Medicaid programs in a number of other states do.³²  Because a high percentage of patients in children’s hospitals are covered by Medicaid programs nationwide, different policies among state agencies may contribute to the variation observed. Also, it has been postulated that in response to the Medicare Hospital Readmissions Reduction Program created by the Affordable Care Act, some hospitals may place more patients in OBS to avoid readmission penalties, which are often calculated using IP-only data. Theoretically, the same could apply for hospitals treating patients covered by Medicaid or commercial insurance. However, at least for Medicare patients, this possibility has not been supported by available evidence.³³  Because hospitals receive higher payments for IP status,²³  there may be incentive to reduce the percentage of OBS patients to obtain greater hospital reimbursement. Furthermore, Medicare recipients may have higher out-of-pocket expenses if placed in OBS as opposed to IP,³⁰  and at least in some settings, OBS patients covered by Medicaid and commercial insurance may also have more out-of-pocket expenses for hospital services. Thus, some hospitals may respond to competitive forces by billing fewer encounters as OBS. Finally, some hospitals may bill OBS and IP days independently within the same hospitalization, and only the final billing status at discharge would appear in our data.

The hospitals represented in the current study are tertiary children’s medical centers treating a wide but similar range of pediatric health conditions. Coupled with previous use data, our finding that hospitals billing a lower percentage of OBS cases had significantly shorter IP-only RA-LOS suggests that these hospitals may have more short-stay encounters that are designated IP in their data, whereas similar cases might be designated as OBS in greater numbers at other institutions. In this instance, if only encounters billed as IP were counted, hospitals with a lower percentage of OBS stays would have an artificially lower RA-LOS because similar short-stay cases would be designated as OBS and excluded at other hospitals. If, in the future, value-based care and pay-for-performance payment structures take LOS into account as a measure of resource use and efficiency of care, hospitals with fewer or no designated OBS stays could be advantaged by the inclusion of more IP status stays of short LOS.

From the consumer’s perspective, if one were comparing hospitals by reported LOS, one might tend to choose a hospital with a lower average LOS, hoping that this would result in less time away from home and work and less income loss for the family. Therefore, a hospital designating few or no short-stay admissions as OBS would mislead the consumer when compared with a hospital with a higher percentage of OBS cases if LOS were measured for IP only. Another potential impact of the OBS versus IP status designation is that families may be responsible for a greater share of costs associated with an OBS stay than one deemed IP status. Litigation has challenged Medicare policy on OBS billing,³⁴  but the financial burden of OBS in the pediatric population has not been studied. In contrast, these hospitals may be disadvantaged if measures of severity of illness are used (eg, a lower case-mix index) because of the inclusion of greater numbers of short-stay encounters with presumably lower acuity. Thus, it will be important to have an accurate measurement of the LOS and case-mix index of patients who actually occupy hospital beds, regardless of their designated billing status, to truly reflect the care provided and afford the most accurate comparisons among hospitals.

This study should be considered in light of several limitations. Data used in this study are from tertiary children’s hospitals and thus may not reflect the impact of OBS stays in general hospitals which treat children. Also, as with most studies in which researchers rely on administrative billing data, our data were subject to errors in accuracy and capture of all relevant codes as well as OBS versus IP designations. The standardized submission of IP versus OBS charges on the CMS billing form (UB-04, National Uniform Billing Committee)³⁵  and data quality assurances by Children's Hospital Association and PHIS member hospitals lessen the contribution of this factor to our results. Some hospitals may bill OBS and IP days for a given hospitalization independently and only the final billing status at discharge would appear in our data. Finally, some hospitals may have chosen not to submit OBS data despite billing some encounters as OBS at their institution. The uniform processes and oversight of submission of data by member hospitals to PHIS make this less likely.

Children’s hospitals exhibit significant variation in the assignment of OBS to admitted patients and inclusion of OBS significantly impacts RA-LOS calculations. The variability in the use of the OBS designation and the overlap between resource use measures between OBS and other short-stay hospitalizations support the recommendation that OBS stays be included in calculations of LOS for purposes of benchmarking and comparative analysis.

CRG

Clinical Risk Group

DRG

diagnosis related group

IP

inpatient status

LOS

length of stay

OBS

observation status

PHIS

Pediatric Health Information System

RA-LOS

risk-adjusted geometric mean length of stay