Implementation of resource use measures in Medicare Advantage

Abstract

Objective

To complement the previously illustrated method to measure resource use in Medicare Advantage (MA) using Encounter data and provide technical details and SAS code to validate Encounter data and implement resource use measures in MA.

Data Sources

2015–2018 MA Encounter, Medicare Provider Analysis and Review (MedPAR), Healthcare Effectiveness Data and Information System (HEDIS), and Traditional Medicare (TM) claims data.

Study Design

Secondary data analysis.

Data Collection/Extraction Methods

We select MA contracts with high data completeness (≤10% missing hospital stays in Encounter data and ≤±10% difference in ambulatory and emergency department visits between Encounter and HEDIS data). We randomly sample TM beneficiaries with a similar geographic distribution as MA enrollees in the selected contracts. We develop standardized prices of services using TM payments, and we measure MA resource use for inpatient, outpatient, Part D, and hospice services.

Principal Findings

We report identifiers/names of contracts with high data completeness. We provide SAS code to manage Encounter data, develop standardized prices, and measure MA resource use.

Conclusions

Greater use and validation of Encounter data can help improve data quality. Our results can be used to inform studies using Encounter data to learn about MA performance.

What is known on this topic

• Medicare Advantage (MA) Encounter data are the counterpart of Traditional Medicare (TM) claims, but they have missing records and do not include payment information.

• Prior work assessed the completeness of 2015 Preliminary Encounter data, checking the data against other data sources.

• Prior work illustrated a method to measure MA resource use for inpatient and outpatient care, using pilot 2015 Encounter data from MA enrollees with cancer.

What this study adds

• We identify MA contracts with highly complete Encounter data between 2015 and 2018.

• We refine the previously illustrated method to measure MA resource use and expand the method to include Part D and hospice services.

• We provide technical details and SAS code to implement the calculation of MA resource use, including validation of Encounter data and development of standardized prices.

1. INTRODUCTION

Encounter data are an important data source to examine care provision in Medicare Advantage (MA). However, a concern exists that Encounter data have missing records. ¹ , ² Lack of payment information is another limitation of Encounter data. Recently, Jung et al. discussed how to handle these two challenges. ³ This prior work assessed the completeness of 2015 Preliminary Encounter data by checking the data against other data sources. It also illustrated a method to obtain standardized prices of inpatient and outpatient services based on Traditional Medicare (TM) payments. With those standardized prices, utilization across types and sites of care can be combined into a global measure of resource use in dollar terms. By using pilot data from cancer patients, the study showed that measuring MA resource use (price‐standardized utilization) with Encounter data is feasible.

This study complements the prior work in four ways. First, we assess the completeness of recent years of Encounter data (2015–2018). Prior work used 2015 Preliminary data, which have a shorter submission period than the currently available 2015 data. ⁴ More complete records may have been submitted during the extended 2015 submission period and in recent years. Second, we include Part D drugs and hospice services in measuring resource use. Prior work measured resource use only for inpatient and outpatient care. Third, we select the MA and TM samples to have the same distribution of US states. Our resource use measures are thus based on the geographic representation of the MA sample. Fourth, we refine the method of measuring resource use by developing standardized prices per unit and by using service units (i.e., quantities reported in a single claim) in MA data. This helps resource use measures better reflect MA practice patterns.

In addition, we provide SAS code and technical details to implement our methods, and identifiers/names of MA contracts with high data completeness. We hope that more studies, building on our approach and code, will use Encounter data to study MA performance.

2. METHODS

2.1. Assessing the completeness of Encounter data

2.1.1. Data

We use 2015–2018 Encounter data from 100% of MA enrollees. To check the completeness of data, we use three Encounter files—Inpatient, Outpatient, and Carrier—and two external data sources: Medicare Provider Analysis and Review (MedPAR) files for inpatient care, and publicly available Healthcare Effectiveness Data and Information System (HEDIS) data ⁵ for outpatient care. Medicare Beneficiary Summary Files supply demographics and the beneficiary's state of residence.

2.1.2. Managing Encounter data

Encounter data include duplicate records for a single encounter and some chart review records. We remove duplicate records using a key (a combination of beneficiary identifier, provider/organization identifier, service date, facility type, and service type) and retain only the latest version of the records. ⁴ Inpatient Encounter data may include multiple records for a single stay. Because discharge analysis is usually of interest, we consolidate inpatient Encounter records to the discharge level. We check discharge‐level inpatient data against MedPAR and use them to calculate resource use.

We incorporate chart review information by adding or deleting diagnosis codes in the Encounter records based on chart reviews that are “linked” to those Encounter records. We treat “unlinked” chart reviews without a related Encounter record as independent encounters. Supplement S1 describes the details of managing Encounter data. Supplement S4 includes the related SAS code (SAS Institute, Cary, NC).

2.1.3. Selecting contracts with high data completeness

Following prior work, ¹ , ³ we exclude contracts exempt from data submission and contracts with <2500 enrollees to increase the reliability of the estimates of data completeness. Among the remaining contracts, we identify those with high “data completeness,” defined as having <10% missing inpatient stays in Encounter data (denominator = total inpatient stays from Encounter and MedPAR data) and less than ±10% difference in both ambulatory visits and ED visits between Encounter and HEDIS data. Supplement S2 reports the detailed process of contract selection and identifiers/names of the contracts with high data completeness. Supplement S4 includes the associated SAS code.

Table 1 shows that 180 (48%) of 377 contracts had high data completeness in 2015, and 210 (61%) of 344 contracts did so in 2018, covering 9.2–14.1 million lives depending on the year. Sixty contracts covering 18 million enrollee‐years had high completeness in all 4 years. Among 363 contracts with at least 2 years of data, 188 contracts continuously submitted highly complete data once they achieved high completeness (Supplement S2).

TABLE 1.

Completeness of Medicare Advantage (MA) Encounter data

By year
	2015	2016	2017	2018
All contracts a
Number of contracts	377	347	335	344
Hospital stay missing rate b	9.4%	8.9%	7.7%	6.7%
Contacts with high data completeness c
Number of contracts	180	175	181	210
Hospital stay missing rate b	4.0%	4.0%	2.8%	3.2%
Number of enrollees (millions) d	9.2	10.7	11.3	14.1

By number of years with high data completeness c
	At least 1 year	At least two consecutive years	At least three consecutive years	All four years
Number of contracts	343	194	103	60
Number of enrollee‐years (millions) d	43.7	37.7	26.1	18.0

^a Cost, demonstration, program of all‐inclusive care for the elderly, employer direct contracts, and contracts with <2500 enrollees are excluded.

^b Percent of hospital stays not included in encounter inpatient data.

^c High data completeness is defined as less than 10% missing hospitalizations and less than ±10% difference from the values from the Healthcare Effectiveness Data and Information System (HEDIS) data for both ambulatory visits and emergency department visits. Only the hospitalization completeness criterion is applied to contracts that did not report HEDIS data.

^d Enrollees with continuous enrollment in MA in a given year.

The mean age of enrollees is 72 years, and 57% of the enrollees are female in both contracts with and without high completeness (Supplement Table S2C). Contracts with high data completeness have a lower representation from the West than those without data completeness. Preferred Provider Organizations tend to submit highly complete data. Health Maintenance Organizations tend to have low data completeness, possibly because their payment methods do not create incentives to record all encounters.

2.2. Developing standardized prices

We develop standardized prices based on TM payments to quantify the MA spending that would have been incurred if MA encounters were paid using TM prices.

2.2.1. Sample and data

We randomly select 50% of the enrollees in MA contracts with high data completeness to reduce the computational burden. We then randomly select a 20% sample of TM beneficiaries based on the distribution of the MA sample across states, which are the common geographic boundaries in TM payment adjustments for inpatient and professional services. We require both MA and TM beneficiaries to have continuous coverage for Part A/B/D in a given year and not to have end‐stage renal disease. MA enrollees must stay in the same contract, and TM beneficiaries must have no MA coverage in the year.

We use published TM fee schedules ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ and actual TM allowed payments (the sum of payer and patient payments) from TM claims data (MedPAR, Outpatient, Carrier, Durable Medical Equipment [DME], Part D Prescription Drug Event [PDE], and Hospice files).

2.2.2. Standardized prices based on TM fee schedules

We use administratively determined TM fees as the “base fees” for most inpatient and outpatient services. Base fees are the relative values/weights or standard payments for a single quantity/unit of service. They differ from the actual TM payments because of adjustments for geographic areas, nonphysician providers, and other factors. They are converted to dollars when base fees represent relative weights. To account for these adjustments, we develop an adjustment factor (AF) for each service category:

$$\mathrm{A}{\mathrm{F}}_t=\frac{{\sum }_{i,j}\mathrm{TM}\text{Allowed Amoun}{\mathrm{t}}_{\mathit{ijt}}}{{\sum }_{i,j}\text{Base}\mathrm{Fe}{\mathrm{e}}_{\mathit{kt}}·\text{Units of Servic}{\mathrm{e}}_{\mathit{ijt}}},$$ (1)

where $i$ is the beneficiary and $j$ indicates a claim line for that beneficiary reimbursed at the base fee for service $k$ in the category in year $t$.

We then compute a standardized price for a unit of service $k$:

$$\text{StdPric}{\mathrm{e}}_{\mathit{kt}}=\mathrm{A}{\mathrm{F}}_t·\text{Base}\mathrm{Fe}{\mathrm{e}}_{\mathit{kt}},$$ (2)

such that the standardized prices reproduce actual total TM payments in the category.

All inpatient services and 95% of physician services comprise a single quantity per claim (e.g., one stay, one surgery, or one consultation during a visit). However, multiple quantities/units of services are delivered during a visit for 5% of physician services and most prescription drug/DME services because their fees are defined for a unit (e.g., per specimen for lab tests). Unit information is recorded in both TM claims and MA Encounter lines. Per‐unit standardized prices from Equations (1) and (2) preserve the relative weights of the services as in published TM fees and help calculate MA resource use using service units in MA data.

We apply the approach described above to the services listed below. Table 2 summarizes the base fees and standardized prices by service category. Supplement S3 describes details of developing standardized prices. Supplement S4 includes the related SAS code.

• Inpatient services: We use TM Diagnosis Related Group (DRG) weights ⁶ for acute hospital stays as the base fees for inpatient services.

• Professional services: TM pays for professional services with a fee‐for‐service system, identifying each service with a code from the Healthcare Common Procedure Coding System (HCPCS). Separate TM fee schedules exist for each service category. We use the following fee schedules as the base fees.

  • Physician services: The Resource‐Based Relative Value Schedule (RBRVS) ⁷ specific to care site (facility vs. nonfacility) and HCPCS modifier (e.g., technical vs. professional components).
  • Laboratory services: The national ceiling of the laboratory fees. ⁸
  • Part B drugs: The Average Sales Price. ⁹
  • Durable medical equipment, prosthetics/orthotics, and supplies (DMEPOS): The national limits of DME fees depending on HCPCS modifier. ¹⁰
  • Anesthesia services: The base anesthesia units ¹¹ and time units with adjustments by HCPCS modifier. ¹²

• Hospice services: Medicare fees for hospice services are determined by a combination of revenue and HCPCS codes. We use published per‐diem rates ¹³ and facility RBRVS for HCPCS‐based professional services (details are described in Supplement S3).

• Part D drugs: Fees for Part D drugs are not administratively determined. To be consistent with our method for other services, we calculate the fee per dosage unit for each drug from allowed payments in the TM sample. We identify drugs by National Drug Code (NDC).

TABLE 2.

Base fees and standardized prices for each category of services

Service category	Base fees (Traditional Medicare [TM] payments)	Standardized prices	Resource use per claim line
Inpatient services	TM Diagnosis Related Group (DRG) weights for acute hospital stays	Base fee times an adjustment factor a	Standardized price of the DRG in MA data
Professional services
Physician services	TM RBRVS fee schedules specific to site of care (facility vs. nonfacility) and HCPCS modifier	Base fee times an adjustment factor	Standardized price times average TM units b per HCPCS (by care site and modifier)
Laboratory services	TM Clinical Lab Payment: National ceiling	Base fee times an adjustment factor	Standardized price times service units in MA data
Part B drugs	TM Average Sale Price (ASP) for each HCPCS	Base fee times an adjustment factor	Standardized price times service units in MA data
DMEPOS	TM DMEPOS fees: National ceiling depending on HCPCS/HCPCS modifier	Base fee times an adjustment factor	Standardized price times service units in MA data
Anesthesia	TM anesthesia base units specific to HCPCS and HCPCS modifier, plus time units (in 15‐min increments) c	Base fee times an adjustment factor	Standardized price times base and time units in MA data
Ambulance	Average TM payments for each HCPCS: Obtained from TM claims	Average TM payments (per‐line standardized prices)	Average TM payments
Services at outpatient facility	Average TM payments specific to facility type (ambulatory surgical center [ASC] vs. non‐ASC) for each HCPCS: Obtained from TM claims	Average TM payments (per‐line standardized prices)	Average TM payments
Part D drugs	Dosage unit fee for each National Drug Code (NDC); calculated from Part D Prescription Drug Event (PDE) data for TM beneficiaries	Calculated fee per drug dosage unit	Standardized price times drug quantities in Part D data from MA enrollees
Hospice services	Per‐diem rates d ; and TM RBRVS weight for HCPCS d	Base fee times an adjustment factor	Standardized price times days in Hospice clams from MA enrollees
Services with no TM base fees	Average TM payments for each HCPCS: Obtained from TM claims	Average TM payments (per‐line standardized prices)	Average TM payments

Abbreviations: DMEPOS, durable medical equipment, prosthetics, orthotics and supplies; HCPCS, Healthcare Common Procedural Coding System; RBRVS, resource‐based relative value system.

^a The adjustment factor is the ratio of the sum of actual payments for TM claims to the sum of base fees x units per claim for TM claims in each category.

^b About 95% of professional physician services comprise a single unit per claim. Thus, the use of average TM units instead of MA units is limited to a very small fraction of professional service claims.

^c Supplement S3 describes details of base fees and the calculation of standardized prices for anesthesia services.

^d Separate per‐diems for the first 60 days and days 61 and later for revenue code 651 beginning in 2016; per‐diem rates are adjusted for 15‐min time intervals for revenue code 652. HCPCS‐based professional services are reported in revenue code 657.

2.2.3. Standardized prices based on actual average TM payments

We use actual average TM payments as the standardized prices when: (1) published TM fees are not available; (2) TM fee schemes are too complex to apply to claims; or (3) the payment and billing systems are not aligned.

We develop the standardized price for a unit of service $k$ and claim line $c$ as:

$$\text{StdPric}{\mathrm{e}}_{\mathit{kt}}=\frac{{\sum }_{i,j}\text{Allowed Amoun}{\mathrm{t}}_{\mathit{ijt}}}{{\sum }_{i,j}\text{Units}{\text{of Service}}_{\mathit{ijt}}},$$ (3)

$$\text{StdLinePric}{\mathrm{e}}_{\mathit{kct}}=\text{StdPric}{\mathrm{e}}_{\mathit{kt}}·\text{Average}\mathrm{TM}\text{Units}\mathrm{per}\mathrm{Lin}{\mathrm{e}}_{\mathit{kct}}.$$ (4)

Subscripts $i,j,\text{and}t$ are the same as in Equations (1) and (2). To apply this approach, we require the services to have at least three TM claim lines to increase the stability of the standardized prices. Only 0.3% of total TM allowed amounts are attributed to services with fewer than three TM claims.

We use the standardized prices from Equations (3) and (4) for the following services:

• Professional services without published fees: Some HCPCS do not have TM fee schedules.

• Ambulance services: TM pays for ambulance services based on mileage calculations by location/transportation routes that are too complex to apply to claims.

• Part D drugs lacking consistency in units: A few NDCs have dosage forms whose units are unclear (e.g., “crystals”) or inconsistently recorded.

• Hospice services without published rates: Some revenue codes or HCPCS do not have published TM rates.

• Outpatient facility services: TM pays for these services with the Prospective Ambulatory Payment System, while facilities submit bills using HCPCS. ¹⁴ We calculate the actual average TM payment for each HCPCS, separately for Ambulatory Surgical Centers and other outpatient facilities.

2.3. Calculating MA resource use

We calculate MA resource use by assigning the standardized prices to the services delivered in MA. Supplement S3 describes details of this process, and Supplement S4 includes the related SAS code.

2.3.1. Inpatient services

We identify inpatient services by DRG in the discharge‐level inpatient data (see Supplement S1). MA Encounter data record DRGs for about 70% of inpatient stays. DRGs for 30% of stays may be missing because some MA contracts do not use a DRG system to pay hospitals. For these stays, we use the “derived DRG” included in Encounter data. Derived DRGs are constructed by the Chronic Condition Warehouse (CCW) based on diagnoses. About 96%–97% of the stays recorded in both Encounter and MedPAR data have the same DRG in the two files. About 4% of inpatient stays have missing DRGs because those stays are identified from unlinked chart reviews. Following the CCW guide, ⁴ we do not use derived DRGs for chart reviews.

The Encounter data capture over 95% of hospital stays by MA enrollees. We add hospital stays identified only from MedPAR data to the stays in Encounter data to measure inpatient resource use. This helps address the missing stays in Encounter data (2.3%–4.0% in the selected contracts; Table 1).

2.3.2. Professional, Part D, and hospice services

We identify professional services by HCPCS/HCPCS modifier, Part D drugs by NDC, and hospice services by revenue/HCPCS codes. We multiply the standardized prices of these services by the service quantities/units in a claim line. We first assess the reliability of units in MA data by (1) comparing the average units per HCPCS between TM claims and MA encounters, and (2) checking the reasonableness of service units for a specific HCPCS. MA units are reliable for all services except 5% of RBRVS‐based services. We thus use the average TM units per HCPCS for RBRVS‐based services and services with no published TM fees. MA resource use measures for these services are based on the distribution of TM service units. However, this impact is limited only to 5% of professional services.

2.3.3. Outpatient facility services and services without published TM fees

We identify these services by HCPCS and assign per‐claim line standardized prices to MA records.

3. DISCUSSION

About 48%–61% of MA contracts had highly complete Encounter data between 2015 and 2018. The completion rate in 2015 is higher than what we reported using 2015 Preliminary data. ³ Contracts with high data completeness covered more than 9 million lives each year. This is encouraging because researchers can do various analyses using data only from those contracts. However, with half of all contracts still having incomplete data, data validation will remain important before using the currently available Encounter data. The validation process is necessary yet nontrivial. We provide the SAS code to implement the process and identifiers/names of the selected contracts, which can facilitate contract selection in future work.

This paper refines the previously illustrated method of measuring resource use, including constructing per‐unit standardized prices and using MA units to measure MA resource use. These improvements help resource use measures better reflect practice patterns in MA. This study also extends resource use measures to Part D and hospice services. Finally, it samples TM and MA enrollees with similar geographic distributions. We use state‐based sampling to incorporate several types of services from nationwide data. However, one can use other geographic sampling units depending on their study aim (e.g., using hospital wage areas to study resource use for inpatient care).

The standardized prices based on TM payments do not vary across MA plans. This lets us quantify what MA would have spent if it had paid providers at TM rates, holding observed utilization constant. This task requires a substantial amount of work to navigate TM payment details, TM claims, and MA data. We provide technical details of our methods and SAS code to implement these methods. While not perfect, our methods and code could be useful instruments to measure MA resource use.

The study has five limitations. First, we check the completeness of inpatient Encounter data against MedPAR files, which include over 90% of MA hospital stays but not all stays. ¹⁵ Second, providers may up‐code some claims, but this behavior represents resources allocated to providers. Third, we do not consider providers' responses to price changes if MA paid providers at TM rates. Fourth, we do not analyze postacute care because Encounter data for those services are highly incomplete. Finally, we do not consider facility‐type specific payments for inpatient care because facility‐type information is missing in Encounter data.

As MA enrollment grows, Encounter data will become increasingly important for examining care provision in MA. Greater exposure of the research community to the data can lead to better data quality. Future studies could build on the methods and code we present and improve them to best use Encounter data.

Supporting information

Data S1. Supporting information.

Data S2. SAS code.

Jung J, Carlin C, Feldman R, Tran L. Implementation of resource use measures in Medicare Advantage. Health Serv Res. 2022;57(4):957‐962. doi: 10.1111/1475-6773.13970