Practical Analytic Approaches to Healthcare Challenges
Evidence-based measures (EBMs) and many other health insurance data analytics require counts of inpatient admission events and associated costs. There are multiple ways to identify and count inpatient admission events, each of which may produce a different result. As per the California Department of Health Services, the methodologies used to identify and count individual inpatient admission events vary significantly across different organizations, due to unique data structures and availability.[i] An inpatient event can be identified by various data points like an inpatient Evaluation and Management (E&M) Current Procedure Terminology (CPT) code; an inpatient Uniform Billing (UB) revenue code; or an inpatient bill type. However, when it comes to utilization or counting of inpatient events, the facility claims that contain inpatient Uniform/Universal Billing form (UUB) revenue codes or inpatient bill types are the most credible, and therefore should be the ones used to identify inpatient admissions.
It is important to highlight that an inpatient admission event typically produces more than one claim line, and all of these individual claim lines must be grouped together to constitute the complete inpatient admission event. Various methods are available to do this, with specific advantages and limitations in each method. We have reviewed several of these methods with their potential advantages and shortcomings and have explored the impact of applying these different methods in a representative claims data set. Inpatient claims can come from different types of facilities, including acute hospitals, long-term acute care (LTAC) centers, acute rehabilitation centers, and skilled nursing facilities (SNFs), where there is evidence that the insured stayed overnight.[ii]
Methods
Method 1: All claim lines for a member with same admission date and discharge date constitute a single inpatient admission event when any one of the claim lines have a UB revenue code for inpatient services.
Method 1 accurately identifies an inpatient facility claim line but it fails to account for overlapping dates on claim lines. A single inpatient admission event can have individual claim lines with different admission and discharge dates. This is referred to as “overlapping dates.” Method 1 counts each claim line with a different admission or discharge date as a unique inpatient admission event and calculates the length of stay (LOS) separately for each.
Method 2: Method 1’s logic is used as the first step to identify admission events, followed by a second step grouping all claim lines with overlapping admission and discharge dates into one inpatient admission event.
Method 2 assumes that claim lines with overlapping admission and discharge dates are part of the same continuous stay and counts them as one inpatient admission event.
Method 3: All claim lines for a member with the same admission date and discharge date constitute a single inpatient admission event when any one of the claim lines has either a UB revenue code for inpatient services, or a bill type for inpatient services.
Method 3 casts a wide net to identify inpatient admission events, as it employs more claim types and service codes than Methods 1 or 2. However, similar to Method 1, claim lines with overlapping admissions or discharge dates are not grouped into a single inpatient admissions event. Expanding the data used to identify inpatient admission events increases the probability of having claim lines with overlapping admission or discharge dates.
Method 4: A step to combine claim lines with overlapping admission or discharge dates is added to Method 3.
Method 4 assumes that claim lines with overlapping admission and discharge dates are part of the same continuous stay and counts them as one inpatient admission event.
Method 5: All claim lines for a member with same the claim/encounter ID are grouped into one inpatient admission event when any of the claim lines have a UB revenue code for inpatient services or a bill type for inpatient services.
Method 5 assumes that the claim/encounter identification (ID) remains the same throughout an inpatient admission event, and combines all claim lines with the same claim/encounter ID. This method counts all claim lines with the same claim/encounter ID, even when there is no overlap between admission and discharge dates.
Variation in Results
We applied these different methods for counting inpatient stays on a test database composed primarily of commercial plan members with over 2.6 million lives and 18,551,986 member months for the year 2012 to compare inpatient utilization results for a Healthcare Effectiveness Data and Information Set (HEDIS) EBM: Inpatient Utilization General Hospital/Acute Care (IPU). Results of this comparison are provided in the table in Figure 1.
Figure 1: Comparison of Results Using Different Inpatient Stay Methodologies
Inpatient admission event counts varied up to 25% across the different methodologies. These results clearly confirm that counts of inpatient admission events can vary significantly depending on the methodology used. Average LOS is also influenced considerably by the method used to identify inpatient admission events.
The outcomes of this analysis emphasize the importance of using a comprehensive methodology when defining inpatient admissions events. It also highlights the need to know the underlying methodology when comparing results for EBMs, for other inpatient admissions event-related measures between organizations, or for different time periods in order to understand if variations in results may be due to the methodology used to define inpatient admissions event rather than operational or quality differences.
[i] California Department of Healthcare Services. Methodology for Identifying Inpatient and Emergency Room Encounters, Appendix A. Retrieved September 7, 2016, from https://www.dhcs.ca.gov/provgovpart/Documents/Appendix%20A.%20Methods.pdf
[ii] Health Care Cost Institute (September 2012). Health Care Cost and Utilization Report: 2011: Analytic Methodology. Retrieved September 7, 2016, from https://www.healthcostinstitute.org/files/HCCI_HCCUR2011_Methodology.pdf