Measuring resource use in Medicare Advantage using Encounter data

Abstract

Objective

To check the completeness of Medicare Advantage (MA) Encounter data and to illustrate a process to measure resource use among MA enrollees using Encounter data.

Data sources

2015 Preliminary MA Encounter, Medicare Provider Analysis and Review (MedPAR), Healthcare Effectiveness Data and Information System (HEDIS), and 2013 Traditional Medicare (TM) claims data.

Study design

Secondary data analysis.

Data collection/extraction methods

We calculated the percentage of each contract's total hospitalizations in Encounter data after identifying total inpatient stays from Encounter and MedPAR data. We constructed each contract's ambulatory visits and emergency department (ED) visits per 1000 enrollees using Encounter data and compared those visit counts with the counts from HEDIS. We defined high data completeness as having less than 10% missing hospital stays and less than ±10% difference in ambulatory and ED visits between Encounter and HEDIS data. We used TM payments as standardized prices of services to examine resource use among MA enrollees with cancer in the contracts with high data completeness.

Principal findings

We identified 83 of 380 MA contracts with high data completeness. Total resource use per enrollee with cancer in the 83 contracts was $14,715 in 2015. Service‐specific resource use was $5342 for inpatient care, $5932 for professional services and $3441 for outpatient facility services. These represent what an MA enrollee with cancer would have cost on average if MA plans paid providers at TM payment rates, holding the observed utilization constant.

Conclusions

Checking the completeness of Encounter data is an important step to ensure the validity of research on MA resource use. Using Encounter data to measure MA resource use is feasible. It can compensate for the lack of payment information in Encounter data. It will be important to identify and refine ways to best use Encounter data to learn about care provision to MA enrollees.

What is known on this topic

• The existing literature has examined care utilization and spending on specific services in Medicare Advantage (MA).

• Evidence comparing resource use in MA and Traditional Medicare (TM) is scarce because, until recently, MA data with detailed diagnosis and utilization records were not readily available.

• The Centers for Medicare and Medicaid Services (CMS) recently released MA Encounter data as the counterpart of TM claims, but they have missing records and do not include payment information.

What this study adds

• We checked Encounter data against other data sources to identify MA contracts with high data completeness.

• We illustrated a process to measure resource use among MA enrollees with cancer using Encounter data in contracts with high data completeness.

• We used TM payments as standardized prices of services to assess what an MA enrollee with cancer would have cost on average if MA plans paid providers at TM payment rates, holding the observed utilization constant.

1. INTRODUCTION

Medicare Advantage (MA) is a private alternative to the Traditional Medicare (TM) program and is one of Medicare's significant efforts to contain costs without harming quality. MA has grown substantially in recent years, covering 43% of the Medicare population in 2021. ¹ , ² Whether MA improves efficiency or reduces costs compared with TM has long been an important inquiry.

The existing studies of health care use in MA generally examined utilization of specific services, such as recommended services to manage chronic conditions or hospital admissions. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ Prior work on MA spending was also service specific, focusing mostly on inpatient and/or postacute care settings. ⁹ , ¹⁰ A few studies used self‐reported out‐of‐pocket spending from specific samples, such as patients with diabetes. ¹¹ , ¹² To our knowledge, only one study evaluated total health care spending in MA using claims from three national MA firms. ¹³ It reported that MA had 9%–30% lower health care spending than TM in 2010.

Evidence comparing resource use in MA and TM is scarce because MA data with detailed diagnosis and utilization records were not readily available until recently. With the recent surge in MA enrollment, the Centers for Medicare and Medicaid Services (CMS) recognized the need for research on MA and began to release MA Encounter files. It released the 2015 Preliminary Encounter files in May 2018 and the 2015 Final Encounter files in July 2019. ¹⁴ The final files have no missing values for final‐action indicators, making it easier for researchers to select records for their analyses. ¹⁴ Annual data from 2015 to 2018 are now available, and later years of data (2019 and onward) will be released annually.

Encounter data are conceptually equivalent to claims, containing records of interactions between patients and providers, such as diagnoses and services delivered to patients. MA Encounter data thus offer a remarkable opportunity to obtain insights on care provision in MA. However, the data impose two challenges for analyzing MA performance. First, the quality of the initial 2015 data is uncertain. ¹⁵ MedPAC (2019) documented that the 2015 Encounter data are incomplete compared with other data sources. ¹⁶ For example, 18% of MA hospitalizations were missing in the 2015 Encounter data. The completeness of Encounter data improved between 2015 and 2017, but the 2017 data were still incomplete. ¹⁷ Second, the Encounter data do not include payment information, which is proprietary. This imposes a challenge for examining spending in MA. Service utilization can be studied with Encounter data but only at the specific service level (e.g., hospitalization, receipt of a test).

We discuss how to handle these two challenges. First, we check the completeness of the Encounter data against other data sources and select contracts with high data completeness. This restricts us to data from a sub‐group of MA contracts. However, analysis of Encounter data from those contracts provides valuable information on patient care in MA. It also provides useful information on how to improve the quality of Encounter data for future research.

To compensate for the lack of payment information in the Encounter data, we apply standardized prices to services delivered in MA. This approach can help infer potential financial implications of MA at fixed service prices that do not vary by plan. We refer to this price‐standardized utilization as resource use throughout the article. Resource use does not reflect the competitive prices that MA plans might have negotiated with providers; however, it offers two advantages over examining service utilization. First, we can combine utilization of different types of services and construct a global measure of resource use in dollars. Second, resource use can capture the intensity of services via the standardized price of each service (e.g., intensive care versus management visits). This is not feasible with a utilization measure, such as the number of outpatient visits.

We use TM payments as the standardized prices for two reasons. First, TM payments are administratively determined and do not vary by providers' negotiating power or specialties for the same service. Second, resource use based on TM payments represents what MA would have cost if it had the same payments as TM, holding constant the observed utilization by MA enrollees. To illustrate the process, we used pilot 2015 Encounter data from a sample of cancer patients to construct inpatient and outpatient resource use in MA. This exercise assesses the feasibility of using Encounter data to measure total resource use in MA. We report only descriptive data on resource use in MA enrollees with cancer. Future studies using our methods can compare resource use in MA and TM, expanding their study samples to broader Medicare populations.

2. METHODS

2.1. Assessing the completeness of MA Encounter data

2.1.1. Data

We used three sources of 2015 preliminary Encounter data: the 100% Inpatient file for hospital stays; the 100% Outpatient file for outpatient facility use; and the 20% Carrier file for professional services. CMS provides only a 20% random sample of the Carrier file to researchers due to the large size of that data set. We used the 2015 Medicare Provider Analysis and Review (MedPAR) file and the publicly available Healthcare Effectiveness Data and Information System (HEDIS) data ¹⁸ as external data sources to check the completeness of Encounter data. The Medicare Beneficiary Summary File supplied information on MA enrollees' age, sex, and race. Patient race in Medicare data is identified from the information used by the Social Security Administration and the origin of last names for Asian or Hispanic beneficiaries. ¹⁹ , ²⁰

2.1.2. Managing Encounter data

MA Encounter data are similar to claims, but two issues need to be managed before using the data: (1) MA Encounter data may have multiple records for the same encounter and (2) MA Encounter data include both encounters and chart review records, defined below.

To address the first issue, we identified duplicative records following the user guide from the CMS Chronic Condition Warehouse (CCW). ²¹ We identified all the records from a unique encounter by creating a “key” from a combination of beneficiary identifier, provider identifier, service date, facility type, and facility‐specific service type. Among multiple records with the same key, we selected the record indicated as the final action, or the record with the latest update for cases with no final action. We retained this record and discarded the duplicates.

MA plans submit chart reviews to add or delete diagnoses. Some chart review records are “linked” to the encounter records to provide complementary diagnosis information. Other “unlinked” chart reviews standalone without a related encounter record. We added or deleted diagnoses codes in the encounter records using the information in “linked” chart reviews. This gave us relatively accurate diagnoses, which we used to construct condition indicators for MA enrollees. Depending on the Encounter file, between 1% and 8% of encounter records had a related chart review (Table S1). We treated “unlinked” chart reviews as independent encounters to partially address the incompleteness of Encounter data. Chart review records have the same data fields as encounter records. The unlinked chart reviews accounted for 1%–5% of the total records (Table S1). MedPAC (2019) showed that inclusion of unlinked chart reviews slightly increased the matching rate between inpatient Encounter records and another data source. ¹⁶ We performed the analysis without chart review information to check the sensitivity of the findings (Table 5).

TABLE 5.

Sensitivity checks of resource use among Medicare Advantage (MA) enrollees with cancer

	Mean (SD) a
	Excluding contracts with missing HEDIS data b	Not using chart review information c
	N = 37,422 (57 contracts)	N = 34,529 (68 contracts)
Total	$14,682 (22,346)	$14,468 (21,871)
Inpatient services	$5261 (14,206)	$5238 (13,316)
Professional services	$5977 (8406)	$5860 (8357)
Physician services	$4476 (1925)	$4396 (5024)
Facility based	$1558 (2832)	$1581 (2940)
Nonfacility based	$2918 (3627)	$2815 (3741)
Lab services	$215 (401)	$229 (404)
Anesthesia	$120 (228)	$116 (213)
Ambulance	$157 (608)	$153 (606)
Part B drugs	$963 (5564)	$921 (5486)
Durable medical equipment	$45 (1046)	$45 (663)
No fee schedule	$0.60 (57)	$0.54 (57)
Services at outpatient facility	$3443 (9094)	$3371 (9345)
Ambulatory surgical centers	$111 (488)	$119 (532)
Nonambulatory surgical centers	$3332 (9084)	$3252 (9332)

Note: Calculated based on 2015 Preliminary Encounter data from a random 20% sample of MA enrollees with cancer in the contracts with ≥2500 enrollees.

^a Among the contracts with high data completeness, which was defined as having 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 was applied to contracts that did not report HEDIS data. Eighty three contracts were selected based on these criteria.

^b Among the 83 contracts with high data completeness, 26 contracts met the hospitalization criterion but had missing HEDIS data were excluded.

^c When chart review records were not used, 68 contracts met the high data completeness criteria.

2.1.3. Selecting MA contracts with high data completeness

A contract refers to an organization entering into a contract with CMS to offer MA plans. It can offer multiple plans, which are specific insurance policies or benefit packages. We excluded contracts that are not required to submit Encounter data (e.g., cost, demonstration, and PACE [Program of All‐inclusive Care for the Elderly] contracts). Following MedPAC (2019), we excluded contracts with <2500 enrollees to increase the reliability of the estimates of data completeness. Table S2 compares the number of contracts selected with and without this enrollment requirement.

We assessed the completeness of Encounter data at the MA contract level using two external data sources. First, we checked records from the 100% Encounter Inpatient file against MedPAR, which includes MA enrollees' hospital stays from hospitals covering 92% of all Medicare hospital stays. ⁷ We defined each contract's total hospitalization counts as the sum of hospitalizations identified only in the Encounter Inpatient file, those only in MedPAR, and those in both files (hospitalizations matched by beneficiary identifier and dates). We calculated the percentage of the contract's total hospitalizations included in the Encounter file to examine the completeness of Encounter inpatient data.

Second, we checked the Encounter data for outpatient care against outpatient service utilization measures from HEDIS. ¹⁸ Using the 20% Outpatient and Carrier Encounter files, we constructed two contract‐level utilization measures following the HEDIS specifications ²² : ambulatory visits and emergency department (ED) visits per 1000 enrollees. This check assesses the internal consistency of data because HEDIS data are also reported by MA contracts.

We selected contracts with high “data completeness,” defined as having less than 10% missing hospitalizations and less than ±10% difference from the HEDIS values for both ambulatory visits and ED visits. We applied only the hospitalization completeness criterion to contracts without HEDIS data because the HEDIS exemption is for reasons not related to data quality, such as administrative reasons, small sample sizes, and new contracts. If we excluded contracts meeting the hospitalization criterion but missing HEDIS, we might have left out contracts with good data quality. We performed sensitivity analyses excluding the contracts without HEDIS data (Table 5 and S2).

We explored variation in the completeness of Encounter Inpatient data across plans within contracts, checking the data against MedPAR records (Supplement S3). We used data only from MA contracts offering multiple large plans to obtain stable estimates of hospitalization counts.

2.1.4. Checking enrollee characteristics

We compared enrollee demographics and condition prevalence between contracts with and without high data completeness using standardized differences. ²³ We constructed condition indicators using Encounter Inpatient/Outpatient/Carrier files from the 20% random sample, following the standard algorithm that CMS uses to construct the indicators for TM beneficiaries. ²⁴

2.2. Measuring resource use in MA

2.2.1. Data and sample

We used Encounter data from the contracts with high data completeness to measure MA resource use. We used three sources of TM claims data to calculate the average TM payments per service for certain categories of services (discussed in the next section): MedPAR, Outpatient, and Carrier files. TM data available to us were limited to records from a random sample of TM beneficiaries with cancer in 2013. We thus selected MA enrollees with cancer to calculate resource use. This sample selection does not affect our illustration of how to use the Encounter data to measure MA resource use. We required MA and TM beneficiaries to meet the following criteria: (1) having both Part A and Part B coverage for the full year, (2) not having end‐stage renal disease, and (3) staying in the same contract for MA enrollees.

2.2.2. Standardized prices for inpatient care

We used TM Diagnosis Related Group (DRG) payments for the standard prices of acute hospital stays. We obtained the standardized prices in two steps. First, we identified 2015 TM “base payments” for acute inpatient services. Base payments are the standard DRG payments before any adjustments. Each year, Medicare sets base rates for hospital discharges to cover operating and capital costs (e.g., the operating rate is $5961 and the capital rate is $466 in 2021). ²⁵ The base payment for each DRG is determined as the base amount times the “weight” of the DRG. For example, the weight for radiotherapy is 2.4875, so the base payment for radiotherapy in 2021 is $15,978.16. This base payment is different from the actual payment for the DRG due to payment adjustments for wage differences across geographic areas, payment reductions for cases with a short length of stay and transfers to postacute care, and additional payments to hospitals offering graduate medical education. ²⁵

Medicare pays inpatient psychiatric facilities on a per diem basis and long‐term hospitals at prospective payment rates per discharge. However, we could not use facility‐type specific payments because the Encounter data do not distinguish inpatient facility types. Thus, we estimated resource use for inpatient psychiatric stays as if the stays had occurred in acute care hospitals.

In the second step, we addressed the gap between TM base payments (i.e., base rate times DRG weight) and actual payments. We multiplied the TM base payments by an “adjustment factor” calculated as the ratio of the sum of actual payments for all TM inpatient claims to the sum of base payments for all TM inpatient claims regardless of area. This adjustment factor is thus affected by the geographic representation of TM beneficiaries because Medicare's payment adjustments vary by area.

2.2.3. Standardized prices for professional services

TM pays for professional services with a fee‐for‐service (FFS) system, identifying each service with a code from the Healthcare Common Procedure Coding System (HCPCS). Professional services refer to services delivered by physicians, other practitioners, and suppliers (e.g., clinical laboratories, ambulances, Part B‐covered drugs, and durable medical equipment [DME]). Separate fee schedules exist for these categories. We used those fees as the base payments for certain service categories. Table 1 reports the base payments and standardized prices for each category of services.

TABLE 1.

Standardized prices for each category of services

Service category	Base payments (Traditional Medicare [TM] payments)	Standardized prices
Inpatient services	TM Diagnosis Related Group (DRG) base payments for acute hospital stays 26	Base payment times an adjustment factor a
Professional services
Physician services b	TM RBRVS fee schedules specific to site of care (facility vs. nonfacility) 27	Base payment times an adjustment factor
Laboratory services b	TM Clinical Lab Payment: National ceiling 28	Base payment times an adjustment factor
Anesthesia c	Average TM payments specific to provider/service type (under supervision vs. performing alone) for each service (HCPCS): Obtained from TM claims	Average TM payments
Ambulance	Average TM payments for each HCPCS: Obtained from TM claims	Average TM payments
Part B drugs	Average TM payments for each HCPCS: Obtained from TM claims	Average TM payments
DME	TM DME payment: National ceiling 29	Base payment times an adjustment factor d
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

Abbreviations: DME, durable medical equipment; HCPCS, healthcare common procedural coding system; RBRVS, resource‐based relative value system.

^a An adjustment factor is calculated as the ratio of the sum of actual payments for TM claims to the sum of base payments for TM claims in each category.

^b The average TM payment was used as the standardized price for a small fraction of services where TM fees are defined for a specific unit (Supplement S1).

^c Supplement S2 describes TM payments specific to provider/service type for anesthesia services.

^d The adjustment factor for DME is the ratio of total actual payment to the sum of the base payment times units per claim for all TM DME claims.

Physician services and lab tests

We obtained the standardized prices for physician services and lab tests in two steps. First, we identified TM base payments for each category of service from the TM fee schedules. We used the resource‐based relative value scale (RBRVS) for physician services. Fee schedules for lab tests vary across states, so we used the national ceiling values as the base payments for lab tests.

RBRVS reflects clinician work (effort, time, skill, and stress), practice expenses, and liability insurance. RBRVS does not vary by provider type, but it differs by facility versus nonfacility site of care (e.g., hospital outpatient departments vs. physician offices). Using the base payment specific to site of care allows the estimates of resource use to capture patterns of care sites in MA. We assigned appropriate values to each claim according to the place of service. We then obtained base payments for physician services as the RBRVS times the standard dollar amounts set by Medicare each year ($36.09 in 2020).

Second, we multiplied the TM base payments for each category of service by an “adjustment factor” calculated as the ratio of the sum of actual TM payments to the sum of base payments for TM claims in each category of services.

We applied the two‐step approach above to most physician/lab services that comprise a single quantity during a visit (e.g., one surgery or one disease management consultation). However, about 4% of the TM records for physician/lab services had multiple units or quantities per claim. This was because fees for certain services are defined for a specific unit (e.g., per muscle for motion analysis procedures or per specimen for pathology services). We used average TM payments per HCPCS as the standardized prices for these multiunit services. To increase the stability in the standardized prices, we used this method only for HCPCS with at least three TM claims in applying the average TM payment. We used the TM base payments for HCPCS with fewer than three TM claims. We performed sensitivity analysis without applying the three‐claim criterion. Supplement S1 describes identification of multiunit services.

Anesthesia, ambulances, and Part B‐covered drugs

TM fee schedules for anesthesia, ambulances, and Part B drugs are based on specific units (15 min of anesthesia time, mileage, 1 mg of an active drug ingredient). We thus used the average TM payment for each HCPCS in these service categories as the standardized prices because of lack of mileage units in MA Encounter data and potential inconsistency in drug units (e.g., an injectable with 10 mg of the active ingredient may be recorded as 1 or 10 units).

For anesthesia services, we used the average TM payment for each HCPCS that is specific to provider/service type (e.g., nurses under physician direction vs. physicians) based on HCPCS modifiers. Supplement S2 describes differences in TM payments for anesthesia services by provider/service type. Use of specific provider/service type payments helps the estimates of resource use reflect provider types used in MA.

We obtained the standardized prices for HCPCS with at least three TM claims for anesthesia, ambulance, and Part B drugs. We excluded HCPCS with fewer than three TM claims because using specific unit‐based TM fees was inappropriate for these services.

Durable medical equipment

TM fee schedules for DME vary by state. We used the national ceiling as the base payment to apply the same price to a service regardless of area. About 19% of TM DME claims recorded multiple units of DME per claim. We thus calculated an adjustment factor as the ratio of the sum of the actual TM payments to the sum of the base payment times units per claim for all TM DME claims. We obtained standardized prices for DME services by multiplying the base payments by the adjustment factor.

Services with no fee schedule

We used the average TM payment for each HCPCPS for services where no TM fees were identified.

2.2.4. Standardized prices for outpatient facility services

We used average TM payments for services delivered in outpatient facilities. Medicare uses a prospective payment system for these services based on Ambulatory Payment Classification codes, but outpatient facilities use HCPCS codes for billing. Claims data include HCPCS codes and the Medicare payments applied to those HCPCS codes.³⁰ We thus used average HCPCS‐specific TM payments as the standardized prices for outpatient facility services with at least three TM claims. We obtained these payments separately for hospital outpatient departments and ambulatory surgical centers because Medicare payment rates differ between the two facility types. This helps our estimates of resource use reflect types of outpatient facilities used in MA.

2.2.5. Calculating resource use

We identified services delivered to MA enrollees by DRG or HCPCS codes in the Encounter data. We assigned the standardized prices described above to Encounter records and calculated the average resource use for each category of services per MA enrollee with cancer. We incorporated per‐claim service units/quantities in MA records in calculating average resource use for DME. We measured resource use separately for each service category, as well as total resource use.

3. RESULTS

3.1. Contracts with high data completeness

Among 540 MA contracts that were required to submit Encounter data in 2015, 380 had ≥2500 enrollees. We identified 83 of these 380 contracts with high data completeness as defined above. On average, 5.7% of hospital stays were missing in the 83 contracts. We found similar missing rates in hospital stays across plans within the same contract (Supplement S3). Table S2 reports the number of contracts and missing rates by selection criterion.

Table 2 compares the distributions of patient demographics in the contracts with and without high data completeness. The mean age was 72, and 56.3% of the enrollees were female in both groups. The 83 selected contracts had a lower share of black enrollees (9.6% vs. 11.9%) and a higher share of Hispanic enrollees (4.2% vs. 3.7%) compared with contracts with low data completeness.

TABLE 2.

Characteristics of Medicare Advantage (MA) enrollees by data completeness

Characteristic	Enrollees from 83 contracts with high data completeness	Enrollees from 297 contracts without high data completeness	Standardized difference
Age (mean)	72.51	72.95	−0.04
Female (%)	56.30%	56.39%	0.00
Black (%)	9.63%	11.68%	−0.07
White (%)	80.14%	79.35%	0.02
Hispanic (%)	4.15%	3.20%	0.05
Region
Northeast (%)	32.42%	15.69%	0.40
West (%)	22.05%	25.78%	−0.09
Midwest (%)	12.24%	20.15%	−0.22
South (%)	25.02%	36.04%	−0.24
US territories (%)	8.27%	2.33%	0.27
Plan type
HMO (%)	71.19%	64.23%	0.15
PPO (%)	26.49%	25.27%	0.03
Regional PPO (%)	1.92%	8.60%	−0.30
MSA (%)	0.39%	0.00%	0.09
PFFS (%)	0.00%	1.91%	−0.20

Note: The data are based on a random 20% sample from contracts with ≥2500 enrollees. Cost, demonstration, and Program of All‐inclusive Care for the Elderly contracts were excluded. High data completeness, which was defined as having ≤10% missing hospitalizations and within ±10% difference from the values from the Healthcare Effectiveness Data and Information System (HEDIS) data for ambulatory visits and emergency department visits. Only the hospitalization completeness criterion was applied to contracts that lacked HEDIS data.

Abbreviations: HMO, Health Maintenance Organization; MSA, medical saving account; PFFS, private fee‐for‐service; PPO, Preferred Provider Organization.

Table 3 shows the prevalence of chronic conditions. Encounter data with fewer missing records can better identify patients with specific conditions. The table indicates slightly higher prevalence rates of almost all conditions in the 83 selected contracts, confirming that these contracts had more complete records than the other contracts.

TABLE 3.

Prevalence of chronic conditions in contracts with and without high data completeness

Condition a	Enrollees from 83 contracts with high data completeness	Enrollees from 297 contracts without high data completeness	Standardized difference
Alzheimer's disease and related dementia	7.4%	7.0%	0.02
Acute myocardiac infarction	0.8%	0.7%	0.00
Anemia	21.8%	19.3%	0.06
Breast cancer	2.8%	2.9%	0.00
Colorectal cancer	1.2%	1.2%	0.01
Depression	19.5%	18.3%	0.03
Diabetes	29.3%	28.2%	0.03
Heart failure	11.5%	11.2%	0.01
Hyperlipidemia	52.5%	50.5%	0.04
Hypertension	62.3%	60.0%	0.05
Ischemic heart disease	23.2%	22.3%	0.02
Lung cancer	0.9%	0.9%	0.00
Osteoporosis	10.0%	9.5%	0.02
Prostate cancer	3.1%	3.1%	0.00

Note: The data are based on a random 20% sample from contracts with ≥2500 enrollees. Enrollees from cost, demonstration, Program of All‐inclusive Care for the Elderly, and employer direct contracts were excluded. High data completeness, which was defined as having 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 was applied to contracts that did not report HEDIS data.

^a Conditions were constructed based on the standard algorithm that the Centers for Medicare and Medicare Services use to create condition indicators for Traditional Medicare beneficiaries. ²⁴

3.2. Resource use

Table 4 shows descriptive statistics on resource use from MA enrollees with cancer in the 83 contracts. Total resource use for inpatient and outpatient care was $14,715 per MA enrollee with cancer in 2015. These descriptive data represent what an MA enrollee with cancer would have cost on average if MA plans paid providers at the TM payment rates, holding observed utilization constant.

TABLE 4.

Resource use and service utilization among Medicare Advantage (MA) enrollees with cancer

	Mean (SD), N = 45,466
	Resource use/enrollee ($)	Utilization/enrollee (visits)
Total	$14,715 (22,313)
Inpatient services	$5342 (14,063)	0.27 (0.44)
Professional services	$5932 (8413)
Physician services	$4434 (4924)	23.76 (18.88)
Facility‐based	$1584 (2878)	7.03 (11.41)
Nonfacility based	$2850 (3597)	17.78 (14.21)
Lab services	$211 (395)	4.81 (5.87)
Anesthesia	$118 (223)	0.52 (0.94)
Ambulance	$160 (612)	0.36 (1.50)
Part B drugs	$964 (5611)	1.80 (3.81)
DME	$45 (935)	0.23 (1.09)
No fee schedule	$0.58 (56)	0.04 (1.40)
Services at outpatient facility	$3441 (9203)	5.91 (7.86)
Ambulatory surgical centers	$115 (519)	0.16 (0.50)
Nonambulatory surgical centers	$3326 (9190)	5.77 (7.89)

Note: Calculated based on 2015 Preliminary Encounter data from a random 20% sample of MA enrollees with cancer in the 83 contracts with high data completeness and ≥2500 enrollees. High data completeness, which was defined as having 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 was applied to contracts that did not report HEDIS data.

Abbreviations: DME, durable medical equipment; SD, standard deviation.

Inpatient resource use was $5342 per MA enrollee with cancer, accounting for 36.2% of total resource use. Resource use for professional services was $5932 per MA enrollee with cancer. The largest share of professional services resource use was allocated to physician services ($4434 per MA enrollee with cancer) and the second largest share to Part B drugs ($964 per enrollee with cancer) probably due to chemotherapy for cancer patients. Other professional services accounted for only a small portion of resource use. Resource use for outpatient facility services was $3441 per MA enrollee with cancer.

We found little difference in resource use estimates when we included HCPCS codes with fewer than three claims—probably because such codes made negligible contributions to total spending.

Table 4 also reports MA utilization data, which can be obtained only at the service category level. About 27% of MA enrollees with cancer were hospitalized in 2015. On average, an MA enrollee with cancer had 23.7 physician visits, 4.8 visits for lab tests, and 1.8 visits for Part B drugs. These utilization data do not reflect the intensity of each visit. The resource use measures indicate that visits for Part B drug services were expensive, while visits for lab tests were the least resource intensive.

Estimates from the sensitivity checks were lower than the primary analysis (Table 5). Estimated total resource use was $14,682 when the contracts with missing HEDIS data were excluded and $14,468 when chart review records were not used.

4. DISCUSSION

The initial MA Encounter data present two challenges for measuring resource use. First, the quality of the data is uncertain. Second, the Encounter data do not include payment information, which is proprietary.

We showed how to address these concerns. To mitigate the concern about data quality, we checked the Encounter data against other data sources and identified MA contracts with high data completeness. This task was neither trivial nor perfect, and it restricted us to using data from a sub‐group of MA contracts. However, the task is critical for researchers who want to use the currently available data to study MA resource use. MedPAC (2020) reported that the quality of Encounter data improved between 2015 and 2017, but the 2017 data were still incomplete. ¹⁷ This suggests that checking Encounter data for completeness will remain important, although complete data from more contracts may be available for years after 2015. The process we introduced can help ensure the validity of studies of MA resource use that rely on the initial Encounter data.

MedPAC (2020) expects that data quality will improve as Medicare places more weight on Encounter data in determining risk‐adjusted payments. ¹⁷ MA contractors will also become more familiar with data submission requirements ²¹ and will likely report more complete records over time. In the meantime, studies using initial Encounter data could refine and advance methods for validating and managing Encounter data, providing constructive feedback on the data. Greater exposure of the research community to the Encounter data can steer MA plans to improve data quality. All these steps can help establish data sets that will allow researchers to evaluate MA performance.

To compensate for the lack of payment information in Encounter data, we introduced a method to measure MA resource use, using TM payments as the standardized prices. Although the complexities of TM reimbursement methods make this task nontrivial, it allows us to estimate what MA would have cost if it paid providers at TM rates holding observed utilization constant.

Using TM payments as the standardized prices was straightforward for inpatient care and most physician/lab services—the two main categories of services that accounted for about 77% of resource use among MA cancer patients. However, it was not feasible for services where TM payments are based on units whose information is inconsistent or unavailable in MA data. The method was also challenging when payment and billing systems are not aligned, as in outpatient facility services (Ambulatory Payment Classification system versus HCPCS codes). We thus used the average TM payments for these services as the standardized prices. This means that our estimates of MA resource use for those services are influenced by the distribution of service units or practice patterns in TM. Future studies identifying ways to mitigate those impacts would help obtain more precise estimates of MA resource use.

Using the standardized prices is a viable option to measure MA resource use with the information available in Encounter data. It is also a useful approach to combine different types of services into a measure of total resource use and to capture the intensity of services via their standardized prices. While not perfect, our exercises showed that measuring MA resource use using Encounter data is feasible. Future studies could build on and refine the method we introduced to perform rigorous analyses of MA performance.

Our analysis has several limitations. First, we assessed the completeness of Encounter Inpatient data against MedPAR, which may not include all MA inpatient stays. Our estimates of inpatient data completeness may thus be a lower bound.

Second, we could not assess data completeness at the plan level because HEDIS data are available only at the contract level. However, data from large MA contracts indicated similar hospitalization missing rates across plans within contracts. We could not assess contract data completeness at the county level because county‐level enrollment was usually too small to obtain stable estimates of data completeness.

Third, we did not consider payment adjustments by area, so the geographic representation of TM beneficiaries is reflected in our resource use estimates.

Fourth, Encounter Inpatient data lack indicators distinguishing inpatient facility types (e.g., acute care hospitals or psychiatric hospitals). We thus applied TM DRG payments for acute inpatient care to all inpatient records.

Fifth, we did not look at resource use for postacute care because MA Encounter data from skilled nursing facilities and home health agencies had only 46%–49% matching rates with other assessment files. ¹⁶ Also, we did not examine resource use for Part D drugs and hospice services because Encounter data do not include Part D or hospice claims. However, it will be straightforward to apply our method to those types of services.

Sixth, we used data only from MA cancer patients because we had TM data only from beneficiaries with cancer. Thus, we may not have captured patterns of resource use in other patient groups. However, this restriction does not affect our illustration of the methodology for measuring MA resource use.

Finally, our estimates of inpatient resource use in MA based on DRG codes may partially capture some upcoding by providers.

In conclusion, CMS released MA Encounter data as part of its efforts to ensure appropriate and efficient care provision to MA enrollees. Encounter data are the first national counterpart of TM claims, thus, they offer an unprecedented opportunity to evaluate MA performance. It will be important to identify ways to best use Encounter data to learn about care provision to MA enrollees.

Supporting information

Data S1. Supporting information.

Jung J, Carlin C, Feldman R. Measuring resource use in Medicare Advantage using Encounter data. Health Serv Res. 2022;57(1):172‐181. doi: 10.1111/1475-6773.13879