Changes in Health Services Use Among Commercially Insured US Populations During the COVID-19 Pandemic

Key Points Question How did health services use among commercially insured populations change during the initial phase of the coronavirus disease 2019 (COVID-19) pandemic in the United States? Findings This cross-sectional study with a claims-based analysis of 6.8 million commercially insured individuals found that during the initial phase of the COVID-19 epidemic in March and April of 2020, patients significantly reduced use of preventive and elective care and increased use of telemedicine but not enough to offset reductions in in-person care. Racial/ethnic and income disparities were seen in changes in use of in-person care and telemedicine. Meaning In this study, the initial 2 months of the COVID-19 pandemic were associated with large reductions in use of health services; future policy initiatives should ensure that these reductions do not adversely affect patient health.


eAppendix 1. Description of Data Sample
This study used medical and pharmacy claims data from employers who purchased access to the Castlight Health platform, which provides price transparency, wellness and other health benefits tools. This analysis did not include the digital tools provided by Castlight Health, but instead used the claims data that participating employers provide to Castlight as a way to implement the digital tools. For each of approximately 200 self-insured employers that provide access to this tool, the claims data covers all in-network procedures that are reimbursed through insurance. The claims data includes reimbursement amounts, procedure codes, and patient diagnoses. The data also includes demographic (e.g., geographic location, age, gender) and employer information (e.g., industry). We did not have access to individual-level data, but instead data aggregated to the year-month-state-gender-age group category level.
One potential concern is that the population included in this sample may not be representative of the broader U.S. population. To assess differences between our study population and the broader commercially insured population, we used data from the American Community Survey (ACS). 2018 is the most recent year available in the ACS, and so we limited our comparison to 2018. We limited the ACS sample to individuals who receive insurance through an employer or union and are under the age of 65. We applied the nationally representative population weights in the ACS data. As shown in Table A1, the Castlight population is similar in gender, age, and geographic distribution to the ACS population.

Description of Regression Model
To examine the association between health care utilization and the first month of the COVID-19's declaration of a national emergency in the U.S., we estimate a regression model that quantifies the change in health care utilization in March 2020, relative to previous periods. We defined healthcare utilization by measuring the number of persons per 10,000 persons who received each of the 145 services grouped by the IBM Watson Health procedure categories. We defined utilization rates ( ) for procedure received by patients (age and gender-level) who live in state during time period (year and month) . Ages are categorized as 0-2, 3-18, 19-26, 27-45, and 46-64. For colonoscopies, we restricted the denominator population to persons ages 46 to 64; for mammograms, women ages 46 to 64; and for infant vaccines, children ages 0 to 2.
With these utilization measures we estimated a regression model of the form = + ℎ2020 + 1 + 2 + 3 + 4 + 5 ℎ + In this model, the ℎ2020 term is an indicator for the March 2020 time period. We included fixed effects controls for the age categories, gender, state, year, and month. We estimated this model separately for the 10 procedures of interest. For the model where we pooled across procedures, we included a fixed effect for procedure ( 6 ). We estimated this model using linear regressions with Stata version 16.
For the models that assess differences by the five-digit zip code level income and race, we first linked the utilization/ measures to zip code. We then used data from the 2018 American Community Survey (ACS) on zip code level household income and race. For income, we defined zip code mean income relative the federal poverty line (FPL) for a family of four ($26,200). We categorized mean zip code income as below 200% of FPL ($52,400), between 200% and 400% ($104,800) and above 400% of FPL. For race, we categorized zip codes as those with 80% or more non-white residents, 79% to 21% non-white residents, and 80% or more white residents (and 20% or fewer non-white residents).
To measure the differences in changes in health care utilization after the national emergency declaration based on zip code income, we estimated a regression model of the form = + 1 ℎ2020 + 2 1 + 3 2 + 4 ℎ2020 × 1 + 5 ℎ2020 × 2 + 1 + 2 + 3 + 4 + 5 ℎ + In this model, 1 represents zip codes with mean household income below 200% of FPL and 2 represents zip codes with household income between 200% and 400% of FPL. The omitted category is zip codes with mean household income 400% or more of FPL. The regression model includes indicator controls for the income level, plus interaction terms between each income level and the March 2020 indicator.
To measure the differences in changes in health care utilization after the national emergency declaration based on the percent of the zip code that is non-white, we estimated a regression model of the form = + 1 ℎ2020 + 2 1 + 3 2 + 4 ℎ2020 × 1 + 5 ℎ2020 × 2 + 1 + 2 + 3 + 4 + 5 ℎ + In this model, 1 represents zip codes 80% or more non-white residents and 2 represents zip codes with between 79% and 21% non-white residents. The omitted category is zip codes with 80% or more white residents. The regression model includes indicator controls for the race category, plus interaction terms between each race category and the March 2020 indicator.

eFigure 1: Unadjusted Utilization of Preventive, Non-elective, Elective, and Pharmaceutical Services in January/February and March/April 2020 Compared to 2018
Note: Colonoscopy population limited to ages 46-64, mammogram population limited to women ages 46-64, vaccine population limited to children ages 0-2, and labor and delivery population limited to women ages 19-45.  We also examined the industry distribution of the Castlight population, across 30 industries. As presented in eTable 2, most enrollees receive insurance through an employer in the telecommunications industry. 4119.7% This table shows regression-adjusted differences in use of office visits and telemedicine in March and April 2020, relative to the 2018 to 2020 time period. The dependent variable in each column is the monthly number of persons per 10,000 eligible persons with the respective procedure. Regression models include fixed effect controls for year and month, state, patient gender, and age category (categorized as 0-2, 3-18, 19-26, 27-45, 46-64). 95% Confidence intervals in parentheses. ** p<0.01, * p<0.05