The current study analyzed data from the Japan COVID-19 and Society Internet Survey (JACSIS), a prospective, Internet-based, self-reported questionnaire survey conducted by a major Internet research agency (Rakuten Insight, INC) from 2020 to 2023. The JACSIS was designed to assess the impact of the COVID-19 pandemic on the health and social conditions in Japan. Questionnaires used for the 2020 baseline survey were distributed from August 25 to September 30, 2020, until the target sample numbers by sex, age, and prefecture category were achieved. Data were collected from a total of 28,000 respondents aged 15–79 years out of 2.2 million participants registered with Rakuten Insight, Inc. The research agency sent survey invitations to 224,389 candidates from these 2.2 million participants. A computer algorithm was used to facilitate the random sampling of participants for this survey. We set a target of 28,000 respondents to ensure enough participants in each age and sex group for accurate estimates while staying within our survey budget. The sample reflected the official demographic composition of Japan as of October 1, 2019, by age, sex, and prefecture categories. The 2021–2023 surveys were similarly conducted between September and October, respectively. We first cross-sectionally tested the association between income and the proportion of participants refraining from regular medical visits in 2020. Then, the change in the proportion of those refraining from regular medical visits after 2020 was assessed prospectively.

All procedures were conducted in accordance with the ethical standards of the Declaration of Helsinki. The study was reviewed and approved by the Research Ethics Committee of the Tohoku University Graduate School of Medicine (approved on June 27, 2024; Approval No. 2024-1-231) and Tohoku Medical and Pharmaceutical University (approved on June 6, 2023; Approval No. 2023-010).

Of the 28,000 respondents, we excluded 2518 participants who provided invalid responses. These exclusions were performed to validate data quality and have been described in previous studies [10, 18,19,20]. From these, we selected 3414 participants who were currently attending medical visits for hypertension treatment. Additionally, 10 participants aged < 20 years at baseline were excluded because they were more likely to be financially dependent on their parents and the total number was small. We also excluded 572 participants who never participated in any of the surveys from 2021 to 2023. Finally, 2832 participants were included in the present analysis.

Information on participants’ annual household income was collected through the JACSIS questionnaires. We used equivalized annual household income in 2020 as the exposure variable. Equivalized annual household income was calculated by dividing annual household income by the square root of the number of household members to adjust for differences in household size [9, 10, 21,22,23]. We divided the participants into three groups based on the median equivalized annual household income: a lower-income group, below 3,182,000 yen (~31,820 US dollars in 2020); a higher-income group, at or above 3,182,000 yen; and an unknown-income group [10].

The definition of hypertensive patients attending medical visits was based on the question “Do you have a history of hypertension? (never/currently not, but had in the past/yes, currently (attending medical visits)/yes, currently (not attending medical visits))” with the response “yes, currently (attending medical visits).”

Refraining from regular medical visits for hypertensive patients in each survey from 2020 to 2023 was used as the outcome.

The outcome in 2020 was obtained from the question, “Did you refrain from a regular medical visit between April and May 2020?” The corresponding question between 2021 and 2023 was, “Have you refrained from regular medical visits for hypertension in the past two months?” The difference in the question sentences between 2020 and 2021–2023 occurred because the first emergency declaration in Japan was made in April 2020. Respondents who answered “Yes” to this question were treated as “refraining from regular medical visits for hypertension.”

Several covariates were selected based on previous studies and clinical findings [2, 8,9,10,11, 18, 20, 22, 24,25,26,27]. The covariates included in the analysis were sex (male or female), age, marital status (married or unmarried), the number of household members, i.e., household size (1 or 2, ≥3), body mass index (BMI) (<18.5, 18.5–24.9, 25.0–29.9, ≥30.0), current alcohol intake (yes or no), current smoking (yes or no), history of diabetes mellitus (yes or no), and history of cardiovascular disease (yes or no). The SES factors other than income, including educational attainment (college or higher, other) and employment status (employer or self-employed or regular employee, non-regular employee, or unemployed), and fear of COVID-19 (low or high), which was considered important during the COVID-19 pandemic, were also included as covariates. We used the Fear of COVID-19 scale (FCV-19S) to assess anxiety and fear of COVID-19 [28, 29] and consists of seven questions, with higher total scores indicating greater fear of COVID-19. Total scores of 7–19 were classified as low COVID-19 fear and 20–35 as high COVID-19 fear, based on the median score of the study participants [20]. We used the covariates measured in 2020. There were no missing values due to the survey design (if any item was not responded to, the survey could not be completed). Additionally, there were no “unknown” options for variables other than annual household income.

To correct the selectivity of Internet-based samples, we used an inverse probability weighting method throughout the analysis (IPW-weighting) [30]. Weights (the inverse of the propensity score, which represents the estimated probability of participating in the survey) were calculated by fitting a logistic regression model using sociodemographic and health-related characteristics to adjust for differences in respondents between this Internet survey and a widely used nationally representative survey from the 2016 Comprehensive Survey of Living Conditions of People on Health and Welfare [31]. The details of the calculation for IPW are described in previous studies [19, 32].

Sociodemographic and health-related characteristics of study participants were calculated according to equivalized annual household income in 2020. The corresponding values after IPW-weighting were also calculated. Differences between the groups were compared by assessing the standardized mean differences (SMD) which assess the standardized absolute difference by considering standard deviations [33]. An SMD of ≥|0.25| was set as the cut-off point determining the large group difference in this study [33].

An IPW-weighted multivariate robust Poisson regression analysis was performed to estimate the proportion of those refraining from regular medical visits in 2020 for each equivalized annual household income, and their proportion ratios (PRs) with the higher-income group as the reference. Model 1 was adjusted for sex, age, marital status, household size, BMI, current smoking, current alcohol intake, history of diabetes mellitus and cardiovascular disease, and fear of COVID-19. Model 2 was additionally adjusted for educational attainment and employment as other SES factors. In addition, to explore whether the association between equivalized annual household income and refraining from regular medical visits varied by respondent characteristics, subgroup analyses by sex, age (<60 years and ≥60 years), household size (1 or 2 and ≥3), and employment status (employed and unemployed) were conducted as sensitivity analyses. Interactions between equivalized annual household income and these factors were tested, excluding the unknown income group. Furthermore, in the subgroup with a pronounced association between equivalized annual household income and refraining from regular medical visits, we used the quartiles of equivalized annual household income as the exposure variable to examine the detailed association. Then, we calculated the P for trend, excluding the unknown income group.

We then assessed the time series change in the proportion of those refraining from regular medical visits using a Poisson generalized linear mixed model for repeated measurements with autoregressive order 1 (AR (1)) correlation structure. AR (1) is a standard method for assessing the covariance matrix in mixed model analyses of longitudinal data [15, 34]. The results were also adjusted for the same covariates used in Model 2, as previously mentioned. The proportion of those refraining from regular medical visits was compared to that of other groups in each year, using the respective higher-income group as a reference.

All data were analyzed using SAS software (version 9.4 M4; SAS Institute, Cary, North Carolina, USA). Statistical significance was set at p < 0.05. Bonferroni correction was applied to adjust for multiple comparisons when assessing changes in the proportion of refraining from regular medical visits during 2020–2023, stratified by equivalized annual household income. Continuous variables are expressed as the mean ± standard deviation. Absolute numbers and percentages were used to present the categorical variables.