Preplanned Studies: COVID-19 Stay-At-Home Orders and Older Adults’ Cognitive Health — United States, June 2018–February 2022

The global prevalence of cognitive impairment is estimated to be 19% (1), with more than 55 million people living with dementia worldwide (2) and 6.5 million in the U.S. (3). While lack of social activities is known to negatively impact cognitive functioning and increase risk of cognitive impairment including dementia among older adults (4), little is known on the effect of coronavirus disease 2019 (COVID-19) related stay-at-home orders on older adults’ cognitive health. This study examined the impact of the COVID-19 stay-at-home orders implemented in certain states of the U.S. and the cognitive health of older adults. Stay-at-home orders requested residents to stay at home as much as possible, and many public shops and venues to close down temporarily (5). This research used data from the U.S. Understanding America Study (UAS), a longitudinal internet survey representative of individuals aged 50 and above in the U.S., and the COVID-19 State Policies (CUSP) database. A difference-in-differences (DID) approach was used to compare trajectories of four cognitive scores before and after state-specific stay-at-home orders were implemented between states with and without COVID-19 stay-at-home orders. This study found no significant relationship between state-specific stay-at-home policies and cognitive health of U.S. older adults.

This study conducted an observational retrospective cohort study. The study population included U.S. adults aged 50 and older who participated in UAS and answered survey questions relevant to cognitive health between June 2018 and February 2022. UAS is a nationally representative longitudinal internet panel survey of more than 9,000 adults older than 18 years. Our outcome variables were four cognitive scores from numbers, picture vocabulary, verbal analogies, and serial seven subtraction tests. The tests were designed to measure the respondent’s quantitative reasoning and lexical knowledge according to the Woodcock-Johnson Tests of Cognitive Abilities (6). Survey questionnaires containing these tests were fielded in two waves, one in June 2018, and the other in July 2020. The respondents could take a test in each wave any time after it became available; only those who participated in the first wave were eligible to participate in the second wave. We included in our analyses all individuals who participated in at least one wave to improve precision of estimates, although only those who participated in both waves contributed to DID coefficients. The independent variable was implementation of stay-at-home orders from the CUSP database. Implementation was treated as a binary variable that equals 1 if the state issued a stay-at-home order (treatment group) and 0 if the state did not issue any order or issued but did not specifically restrict movement of the general public during the study period (control group). Our control variables included gender, age, immigration status, marital status, education, ethnicity, race, presence of other household members, employment status, and household income. These variables were also obtained from UAS.

This research examined summary statistics of key variables, and t-tests for continuous variables and Chi-squared tests for categorical variables were used to compare means between treatment and control groups. Our DID model specification was as follows:

where the dependent variable $y $ is a cognitive test score for individual $ i $, state $ s $, and quarter $ q $. $ {d}_{s} $ is an indicator for whether state $ s $ implemented strict stay-at-home orders. $\; {p}_{q} $ is an indicator for whether quarter $ q $ is after the second quarter of 2020, since most states implemented stay-at-home orders between March and April 2020. $ {X}_{i,q} $ contains individual-level sociodemographic control variables in quarter $ q $. $ {\delta }_{q} $ and $ {\alpha }_{s} $ are quarter and state-fixed effects. ${\varepsilon}_{i,q,s}$ is the error term and is clustered at the state level. We conducted secondary analyses on subsamples stratified by age (65 or over vs. under 65), gender and whether the individual lived alone. All analyses were performed using Stata BE 17.0 (StataCorp, College Station, TX, U.S.A.).

Table 1 shows the numbers of state and person-wave observations for each of the four cognitive tests and the average test scores in states with stay-at-home orders (treatment group) and without stay-at-home orders (control group). Forty states out of 51 issued a stay-at-home order between March and April 2020. During the study period, the total number of observations per panel was about 8,000, of which over 80% were from the treatment group, with an average of approximately 170 observations per treatment state compared to approximately 100 per control state. The average cognitive test score was slightly higher in states with stay-at-home orders than in states without stay-at-home orders in all panels. Relative to the control states, treatment states had higher proportions of immigrants and unmarried people and higher average education levels and household incomes (Table 2). To assess selective attrition between Wave 1 and Wave 2, we compared summary statistics by the number of waves the respondent participated in (Supplementary Table S1). Although we did not find large differences between those who participated in only one wave of cognitive assessment versus two, the former had somewhat higher socioeconomic status than the latter in terms of education and income. Table 3 shows the differential change in cognitive scores in states with stay-at-home orders relative to states without stay-at-home orders. None of the associations was statistically significant. In terms of coefficient magnitudes, stay-at-home orders were associated with lower numbers test and serial seven subtraction test scores compared to states without stay-at-home orders, by 0.184 points and 0.041 points, respectively. In contrast, stay-at-home orders were associated with increased picture vocabulary test and verbal analogies test scores by 0.221 and 0.757 points, respectively. Subgroup analyses yielded consistent results (not shown).