What is the most complete source of data on the use and costs of health care for Americans?
Using multimode methods to gather data for the MEPS-HC
Policy changes in the U.S. health care delivery and financing system affecting how Americans use and pay for health care services and prescription drugs require continuous large-scale data updates. These updates not only project how much the government needs to spend on federal health care programs—current and future—as well as on drug benefits, but also alert policymakers about pressure points in the private sector. For example, these might include increases in the percentage of Americans without health insurance and their impact on the use of emergency room services. All these critical data need to be timely and ongoing as the nation’s health care system is constantly shifting.
Gathering this data efficiently and accurately through the Household Component (HC) of the Medical Expenditure Panel Survey (MEPS) requires a highly skilled research firm with extensive large-scale survey experience, an extensive staff of specialists, including data science experts and field data collectors, state-of-the-art technology, and multimode approaches, to efficiently tackle data collection and processing to create public-use files.
Westat conducts MEPS-HC to support the mission of the Agency for Healthcare Research and Quality (AHRQ) to produce evidence to make health care safer, higher quality, equitable, more accessible, and affordable.
Each year, Westat fields a new panel of approximately 10,000 households that responded to the National Health Interview Survey in the prior year. These households are invited both to participate in a series of 5 interviews over a 2.5-year period and to allow MEPS to obtain data from their health care providers’ records. This design results in a series of overlapping panels so that each year we can combine data from 2 panels to create annual estimates related to the U.S. civilian population. To manage this vast amount of data, Westat has produced 4,000 class variables to impute health care expenditures and reduce bias in MEPS expenditure estimates.
4,000 class variables Westat produces 4,000 class variables to impute health care expenditures and reduce bias in MEPS expenditure estimates.
Our interviewers typically conduct the main interview in person, supported by computer-assisted personal interviewing (CAPI), computer-assisted telephone interviews (CATI), and computer-assisted video interviews (CAVI). In addition, we use web and paper self-administered questionnaires (SAQs) to gather additional information about individual adults in each household.
When the COVID-19 pandemic interrupted in-person interviews in 2020 and into 2021, we responded with creativity and flexibility, seeking other modes to resume data collection without jeopardizing the estimates moving forward. We shifted immediately to telephone interviewing and built out an existing respondent website to provide households with materials and “show cards” so they could see response categories for specific items. For a new 2021 SAQ on social determinants of health, we created both web and paper versions of the questionnaire and used email and SMS to invite respondents to complete the web version. In 2022 we introduced CAVI for the main interviews, enabling our interviewers to build rapport with respondents through eye contact and a dedicated window to display show cards, eliminating the burden on the respondent to visit a website to see them.
MEPS-HC data collectors are equipped with advanced technology to support Westat’s data collection process. Our advanced field operating system supports case assignments, local and nationwide travel, and all time and expense reporting. These systems operate on laptops and mobile devices, and we use GIS data to assist data collectors in navigating to, and making contact with, respondents. The systems also collect an extensive set of operational data that support rapid feedback to field staff, reporting to managers and clients, quality control, and survey methods research.
MEPS leverages data science methods among them natural language processing (NLP), machine learning (ML), and deep learning techniques to improve survey processing and imputation of missing data. For example, our interviewers can make open-ended comments to clarify respondent answers. About 20,000+ comments are entered by interviewers each year into the CAPI system. We have trained a classification model to automatically label the comments into 10 predefined classes. This increases processing efficiency while maintaining exacting standards for data quality. In addition, throughout the data collection process, our large staff of specialists look for anomalies in the data to assure its quality. As part of our quality control, our data science unit, using artificial intelligence (AI), scans transcripts of thousands of computer-assisted recorded interviews (CARI) to determine if there are any falsification of responses that would jeopardize the data’s quality. Increasing the efficiency of data management through innovative data science techniques is critical since MEPS requires a rapid turnaround to field the cases for the next survey cycle.
We are also working with AHRQ to develop web SAQs for MEPS respondents with specific health concerns, such as diabetes, cancer, or heart disease. We are implementing continuous improvements using multimode and advanced technology in order to reduce under-reporting and perceived respondent burden. In this way, we can gather health event data more continuously, streamline the in-person interviews, improve data quality, and reduce perceived respondent burden.
Since 1996, we have conducted nearly 1 million in-person MEPS-HC survey interviews, providing the government with annual national estimates concerning the use and costs of health care services by Americans, how they pay for them, their health insurance coverage, health conditions, and other items related to health care usage.
Each year, we produce 3 major data files and a series of topic-specific files, combining the data reported by households with data from their health care providers, and imputing for missing data. These data continuously serve to inform the government on how best to improve the quality of federal health care programs and expand access to care.
The reason the MEPS-HC surveys have been so successful in efficiently supplying the government important data is because we are dedicated to being responsive, nimble, and creative, and we have retained year after year an enormous staff of specialists and interviewers.
Rick Dulaney, Vice President, Large Surveys Practice
Focus AreasHealth Services Research and Health Policy
CapabilitiesAdvanced Technologies Data Collection Data Collection Modes Data Collector Recruitment, Hiring, and Training Data Science Field Management Survey Design
TopicsComplex Surveys COVID-19 Data Science Multimode Data Collection
Issue BriefYoung Children’s Consumption of 100% Fruit Juice by Racial-Ethnic Characteristics of Their Mothers
Congress authorized the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) as a pilot program in 1972 and made WIC permanent in 1974.…
PerspectiveCDISC Conformance and Compliance: So Many Resources, So Little Time!
Implementing the Clinical Data Interchange Standards Consortium (CDISC) standards required for data management, programming, documentation, and eCTD data submission can be challenging for many reasons…
Expert InterviewHow Will the Marine Corps Integrate Men and Women at Recruit Training?
Marine Corps recruit training is every bit as intense as it looks. Each morning starts hours before sunrise with a chorus of screaming drill instructors…