Our study is based on an analysis conducted for PPMCO, which is owned by Johns Hopkins HealthCare, LLC and Maryland Community Health System. This affiliation facilitated access to appropriately de-identified claims data for all 300,000-plus members served by PPMCO. In our initial analysis, we used all individual claim line items (over 10 million) from January 1, 2019, to April 23, 2020. These data provide detailed information on charges and reimbursements on behalf of all patients to all providers (facilities and professional). Since we had information in all claims, we were able to drill down precisely to identify, all eligible cases from HOPDs who could have had their procedures performed at an ASC if one offered that procedure. Conversely, we were also able to identify all procedures that could be performed by at least one ASC. By matching movable patients against eligible ASCs we were able to provide conservative estimates of potential savings.
Specifically, to develop an estimate of the potential savings involved when moving the SOS from an HOPD to an ASC, we organized the claims records into a set of episodes of care (EOCs) based on patient ID number and time of service delivery. Within each EOC, we identified the CPT (Current Procedural Terminology) code with the highest reimbursed claim entry or data element. With this in hand, we identified all ASCs that received payment for that same element. By hierarchically assigning each EOC to one CPT, we avoid duplications that could have arisen if the sorting were done at the CPT level first. This establishes a set of potentially movable cases, a set of alternate locations, and a conservative value for case-related savings in that we assumed that if an EOC were moved, only savings for that one CPT would be identified as potential savings.
Below, we specify the steps of the rules-based algorithm developed and provide an explanation of the necessity and impact of each step. We also walk through a prototypical case to highlight several nuances of our approach.
Steps of the algorithm
-
Step 1. Collect all claims data accrued for all members during the study period. Since all claims and cases are considered, there is no sampling error in the evaluation.
-
Step 2. Select all instances of outpatient service that took place at an HOPD. Each line in this data set has a Place of Service code, with location “22” [9] referring to all HOPD’s. We also omit all cases handled on inpatient status.
-
Step 3. Bundle claims to define EOCs. A 30-day window was used to accrue all reimbursement to a single EOC.
-
Step 4. Exclude all EOCs for which total reimbursement is below $500. This helps avoid cases with minimal savings – this floor was chosen by decision makers.
-
Step 5. Exclude all EOCs that are associated with an emergency room admission from 7 days before the claim to 30 days afterward [10]. Emergency room visits are unplanned, and could not have been redirected.
-
Step 6. Exclude all EOCs for patients who are 18 years of age or younger. They are ineligible under outpatient general surgery precertification initiative, 2019 [11]
-
Step 7. Exclude all EOCs that fall under the existing MNC. Medical Necessity Criteria are in place to direct the most difficult cases to the HOPD. Since no such policy was yet in place for this payer, the MNC policy from United Healthcare was used as a proxy.
-
Step 8. Identify potential ASCs based on the most expensive procedure code. We identify ASCs based on their “Place of Service Code 24” along with the corresponding Paid Amount and CPT code.
-
Step 9. Match each EOC to the candidate ASC with the largest reimbursement for the most expensive CPT code in the EOC. Matching of Outpatient EOCs with ASC records results in an expanded data set having one-to-many EOC-to-ASC combinations or “potential candidates”. Using ranking functions we select the most expensive CPT code in an EOC matched with the most expensive ASC record for a conservative savings estimate within the algorithm.
-
Step 10. Exclude all EOCs with potential savings below $100. The savings needs to exceed any projected administrative cost associated with the SOS policy.
Algorithm details
For our dataset, Step 1 identified roughly 10 million records for 300,000 members from January 1, 2019, to April 23, 2020. Step 2 selects a subset of 1.2 million records for outpatient services involving the HOPD. In Step 3, we bundled the claims by patient identifier and date to generate a set of EOCs. Step 4 removed EOCs with very small payments and resulted in a list of 64,884 EOCs. In Step 5, we sought to remove cases that could not be moved because they were not scheduled in advance. This step reduced the set to 55,744 EOCs. Step 6 excluded all cases for patients who are 18 years of age or younger and yielded 43,159 EOCs. Step 7 excluded all cases already covered by the MNC, and yielded 41,649 EOCs with 338,870 claim line items. Step 8 created a list of candidate ASC locations for each EOC, and identified 16,212 potential matches. In Step 9, we calculated potential cost savings for each of these EOCs. This is done using the minimum value of that savings to produce a conservative estimate. Step 10 pruned the set from Step 9 to focus on EOCs with potential savings above $100. This process yielded the final number of 7,679 EOCs. Thus from about 65,000 potential EOCs, roughly 12% (,7679) were identified as movable cases. These extractions are presented in Fig. 1 with each stage depicted by a circle of commensurate size.
Note that three aspects of this algorithm are in place to ensure that our estimate of potential savings is a conservative one. First, we only focus on the line in the claim with the highest payment. This ignores the technical fee (averaging $150) that HOPD’s add to each EOC. Second, we screen out instances where the potential savings is small, but still positive. Third, we ultimately consider the alternate location where the savings on the largest line of the claim is smallest.
Consider the example described in Fig. 2. The figure is split into 3 parts referring to 3 extractions from the data base. The top of the figure, consistent with Step 8, shows that this EOC was defined to include 14 line items. The aggregate reimbursement for this EOC was $6133.48. The middle part of the figure, also consistent with Step 8, shows four of these items sorted in descending order based on the paid amount, and indicates that the largest payment was the institutional claim for CPT 47,562 at $5465.86. Consequently, any ASC that had billed for this CPT code was considered an alternate SOS for this EOC. Using this rule, we found 19 potential matches. The third part of the figure, consistent with Step 9, shows four of the alternate locations displayed in descending order based on the amount paid at that ASC for the same procedure. The largest payment amount among this set was $2412.74. This difference ($5,465.86—$2,114.74 = $3,351.12) is used as our estimate of potential savings for this EOC.