Who Has an Unsuccessful Observation Care Stay?

Background: With the recent increase use of observation care, it is important to understand the characteristics of patients that utilize this care and either have a prolonged observation care stay or require admission. Methods: We a conducted a retrospective cohort study utilizing 5% sample data from Medicare patients age ≥65 years that was nationally representative in the year 2013. We performed a generalized estimating equation (GEE) logistic regression analysis to evaluate the relationship between an unsuccessful observation stay (defined as either requiring an inpatient admission from observation or having a prolonged observation stay) compared to having successful observation care. Observation cut offs of “successful” vs. “unsuccessful” were based on the CMS 2 midnight rule. Results: Of 154,756 observation stays in 2013, 19 percent (n = 29,604) were admitted to the inpatient service and 34,275 (22.2%) had a prolonged observation stay. The two diagnoses most likely to have an unsuccessful observation stay were intestinal infections (OR 1.56, 95% CI 1.32–1.83) and pneumonia (OR 1.26, 95% CI 1.13–1.41). Conclusion: We found patients placed in observation care with intestinal infections and pneumonia to have the highest odds of either being admitted from observation or having a prolonged observation stay.


Introduction
In recent years, there has been greater use of observation services for patients by all types of providers [1][2][3] This care provides a short-term (24-72 h) treatment and assessment, is billed as an outpatient visit, and can take place in the emergency department, inpatient units, special observation units, or any other monitored settings [4] It is utilized by providers to "observe" patients in a monitored setting, usually a hospital. Patients placed in observation care are not well enough to be discharged home and not sick enough to require a prolonged admission. Due to the nature of observation care, patients placed in this care are not expected to require prolonged monitored care.
While the idea of observing a patient dates back to Hippocrates, the increased use of observation care in the US is relatively new [5]. As providers better understand the roles and uses of observation care stays, they require an improved understanding of the outcomes of patients placed in observation care. For inpatient providers and hospital administrators, patients who have unsuccessful observation stays either require an inpatient admission or to have a prolonged observation stay. It is important for both providers and administrators to understand the characteristics of these patients as unsuccessful observation stays are costly to the system, not clinically expected, and may result in unnecessary care. Currently, there are no known studies that assess the characteristics of patients who have an unsuccessful observation stay.
We evaluated 154,756 patients with Medicare Insurance age ≥65 years placed in any US hospital observation care in 2013. The objective of the study was to evaluate the characteristics of patients who utilize observation care and subsequently have an unsuccessful stay, either by being admitted to the inpatient service or by having a prolonged observation stay, defined as ≥2 midnights.

Study Design
We performed a retrospective cohort study of a 5% sample of Medicare patients that was nationally representative. All patients were placed in observation care in 2013. The IRB at the University of California, Los Angeles approved the study.

Setting and Selection of Participants
Participants in the study were age ≥65 years at the time of their first day of observation care use. If participants had multiple observation care stays, then only the first stay of the year was included in the analytic sample. Patients who had an observation stay more than 30 days or who were deceased during the observation stay were excluded.

Data Sources
Visit records used for the study analysis were obtained from the Center for Medicare and Medicaid (CMS) Outpatient File, the CMS Inpatient MEDPAR (Medicare Provider Analysis and Review) file, the Master Beneficiary File, and the Chronic Conditions file for 2013.

Measures
Patient comorbidities were derived through the CMS Chronic Conditions file which was linked to the visit records using Claim ID. The CMS Chronic Conditions file contains information regarding the sum total of chronic conditions prior to the observation stay (0-27). Medical diagnoses were obtained based on an algorithm developed by the PI of the study [6][7][8]. In brief, a cross-walk mapping process was linked to the primary ICD-9 code for each observation stay through use of the Multi-level Clinical Classification system (CCS) codes provided by the Healthcare Cost and Utilization Project (HCUP) [9]. The PI developed a total of 39 categories, which have been outlined in the Appendix A. Having used the emergency department (ED) immediately prior to the observation stay, inpatient admission, and use of a skilled nursing facility (SNF) were determined based on Revenue Center Codes as well as charges made to Medicare.
Observation care cut-offs (successful vs. unsuccessful) were based on the CMS 2 midnight rule billing criteria [10] as well as discussion with a set of hospital administrators and inpatient physicians at UCLA and other hospitals. The terms "successful" vs. "unsuccessful" were also obtained through discussion with the administrators and providers. "Unsuccessful" was defined as having at least 2 midnight observation stays or being transferred to the inpatient service. Observation stays of 0 days required at least an 8 h placement in observation care to be billed as "observation". Each day of observation care usage (i.e., 1 or 30) required the same number of midnights as days.

Data Analysis
Patient characteristics (demographic and clinical) as well as the diagnoses were summarized for the two clinical outcomes following an observation stay (successful observation care stay and unsuccessful observation care stay). In addition, both descriptive statistics and frequency distributions for continuous and categorical variables were generated.
Candidate factors included demographic characteristics, patient comorbidities proxied by the number of CMS chronic conditions, and observation care diagnoses. Clinical Outcomes were modeled using a Generalized Estimating Equation (GEE) logistic regression [11]. The model included all candidate factors as fixed effects and provider-level random effects that accounted for multiple observations within providers.
The model evaluated the factors associated with unsuccessful observation care (inpatient admission from observation care or observation care 2-30 days) vs. successful observation care (0 or 1 days/midnights of observation care which equates to a maximum of 47 h and 59 min). Adjusted odds ratios (AOR) and 95% confidence interval estimates were generated from this analysis. The reference groups for all analyses were the following: Age 65-69, female gender, weekday initial observation placement, observation placement from a non-ED, never used a SNF, no chronic conditions, and observation care diagnosis of "Urinary Tract Infection". In addition, the study group conducted additional sensitivity analyses regarding patients who attended the ED and weekend vs. weekday visits. Table 1 describes the characteristics of the sample. There were close to twice the number of female patients as compared to male (96,742 vs. 57,994). Fifteen percent of the cohort were placed in observation on the weekend and over half of the number of patients placed in observation came from the emergency department. Of all patients placed in observation care, a total of 64,215 (41.5%) came from the ED on a weekday and 21,500 (13.9%) came from the ED on a weekend day. Table 2 describes the diagnoses of the patients with observation stays and their outcomes. Of all diagnoses, diseases of the musculoskeletal system resulted in the highest number of patients placed in observation care (N = 17,401). This diagnosis also had the highest percent of placement (76.3%) in successful observation care. The diagnosis with the greatest number of admissions from observation was pneumonia (1077/1857, 58%). The diagnosis with the greatest percent of prolonged observation was abdominal pain (36.1%).

Discussion
In recent years, there has been a greater use of observation care [1,2,12,13] This type of "temporary" care allows providers to place patients in a monitored setting, usually a hospital, where they can be watched for 0-48 h while being considered an outpatient encounter [5] For providers, administrators, and health policy experts, it is important to understand the type of patients that have an unsuccessful observation stay, defined as either having a prolonged observation stay or getting admitted from observation care, as having an unsuccessful observation care stay is not only unexpected to the health care system but it may result in greater cost and unnecessary care for the system. We found that patients with intestinal infections and pneumonia have the highest likelihood of having an unsuccessful observation care stay. In addition, we also found that patients coming from the ED, seen on a weekend as compared to weekday, and having been placed in a skilled nursing facility to have a higher rate of an unsuccessful observation stay.
The diagnosis with the highest odds of having an unsuccessful observation care stay was an intestinal infection, ranging from a rare diagnosis such as Cholera or Shigella to an ill-defined diagnosis. An intestinal infection is commonly a condition that is transitory in nature and while physically uncomfortable, less likely to require aggressive treatment. The findings of this study suggest that if a patient requires placement in the hospital, there may be additional factors not identifiable in administrative data that could lead to prolonged care such as dehydration and/or requirement of an extended course of treatment.
Pneumonia had the second highest odds of an unsuccessful observation care stay. Over 50% of pneumonias are classified as community acquired pneumonia [14]. While the epidemiology and bacteriology of all the types of pneumonia are different, on initial presentation a provider is unable to distinguish between the different kinds of pneumonia until further testing is done [15]. As pneumonia is an infection that can have an unpredictable course, it is understandable that patients with pneumonia had a high rate of an unsuccessful observation care stay. It is also possible that patients with pneumonia were misdiagnosed.
We found that originating from the emergency department had a high odds of an unsuccessful observation care stay. Patients placed in observation care can range between having come from an acute encounter or a scheduled procedure and providers in the ED often lack historical information on patients [5]. The unpredictability of the type of patents presenting to the ED as well as the lack of history may lead to ED providers not understanding the complexity of care patients may need. It is important for health care administrators to be aware of these finding so that if patients do originate from the ED, they receive a more defined method of management.
Patients placed in observation care on a weekend had a higher likelihood of an unsuccessful observation care stay. This could be a result of multiple factors. Care delivered to patients on weekends does not often include the complete staff and services needed. In addition, patients may have prolonged seeing a provider until the weekend and the condition could have worsened. Although the study controlled for number of comorbidities and conditions, it was unable to account for severity of illness.
Patients in a skilled nursing facility (SNF) usually have a greater number of medical problems and require more ancillary care [16] As these patients are more "complex" it would be expected that they would have a greater likelihood of having a prolonged observation stay or requiring admission following their observation care. In the same light, it would lead to an excess in resource utilization if all patients from a SNF were admitted. Providers seeing these patients should continue to evaluate and develop a disposition plan based on need but should keep in mind that these patients have a higher likelihood of not being successful in their observation care stay.

Limitations
The study has some limitations. First, the analysis is based on data derived from claim ID, billing data, and ICD-9 codes, which are limited in that they are retrospective and can reflect incomplete coding. Second, a majority of patients who use Medicare insurance do not visit Federal hospitals, so these findings are not generalizable to Federal facilities [3]. Third, the analysis did not include information from prior year observation stays as that would require use of data from a prior year that the team did not have. Also, the files lack clinical variables such as vital signs and physical exam. The files also lack information regarding hospital characteristics such as teaching vs. non, rural vs. non, average income of hospitals, etc. Finally, the data is several years old as a result of the time it took to acquire (2 years), link and clean the files (2 years). Despite these limitations, this study provides important information regarding older Medicare beneficiaries that experience observation stay.

Conclusions
With the rise of observation care utilization, we assessed the factors associated with having an unsuccessful observation care stay. Patients with either an intestinal infection or pneumonia had the highest odds of an unsuccessful observation care stay. In addition, patients coming from the emergency department, placed in observation care on a weekend, or requiring a skilled nursing facility had the highest likelihood of lack of observation success. This study provides relevant and essential information for both providers and hospital administrators.

Conflicts of Interest:
The authors declare no conflict of interest. Figure A1.

Name of Chronic Condition
Variable Name in the Dataset Table A2. Cont.