ArticlePDF Available

How do patients with chronic illnesses respond to a public health crisis? Evidence from diabetic patients in Japan during the COVID-19 pandemic

Authors:

Abstract and Figures

How do people change their healthcare behavior when a public health crisis occurs? Within a year of its emergence, coronavirus disease 2019 (COVID-19) has gradually infiltrated our lives and altered our lifestyles, including our healthcare behaviors. In Japan, which faces China across the East China Sea and accepted 924,800 Chinese tourists in January 2020, the emergence and spread of COVID-19 provides a unique opportunity to study people's reactions and adaptations to a pandemic. Patients with chronic illnesses who require regular doctor visits are particularly affected by such crises. We focused on diabetic patients whose delay in routine healthcare invites life-threatening complications and examined how their patterns of doctor visits changed and how demographic, socioeconomic, and vital factors disparately affected this process. We relied on the insurance claims data of a health insurance association in Tokyo. By using panel data of diabetic patients from April 2018 to September 2020, we performed visual investigations and conditional logistic regressions controlling for all time-invariant individual characteristics. Contrary to the general notion that the change in healthcare behavior correlates with the actual spread of the pandemic, the graphical and statistical results both showed that diabetic patients started reducing their doctor visits during the early stage of the pandemic. Furthermore, a substantial decrease in doctor visits was observed in women, and large to moderate reductions were seen in patients who take insulin and are of advanced age, who are at high risk of developing severe COVID-19. By contrast, no differentiated effect was found in terms of income status. We further investigated why a change in pattern occurred for each subgroup. The patterns of routine healthcare revealed by this study can contribute to the improvement of communication with the target population, the delivery of necessary healthcare resources, and the provision of appropriate responses to future pandemics. (299 words).
Content may be subject to copyright.
Since January 2020 Elsevier has created a COVID-19 resource centre with
free information in English and Mandarin on the novel coronavirus COVID-
19. The COVID-19 resource centre is hosted on Elsevier Connect, the
company's public news and information website.
Elsevier hereby grants permission to make all its COVID-19-related
research that is available on the COVID-19 resource centre - including this
research content - immediately available in PubMed Central and other
publicly funded repositories, such as the WHO COVID database with rights
for unrestricted research re-use and analyses in any form or by any means
with acknowledgement of the original source. These permissions are
granted for free by Elsevier for as long as the COVID-19 resource centre
remains active.
SSM - Population Health 16 (2021) 100961
Available online 15 November 2021
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
How do patients with chronic illnesses respond to a public health crisis?
Evidence from diabetic patients in Japan during the COVID-19 pandemic
Masataka Harada
a
,
*
, Takumi Nishi
b
, Toshiki Maeda
c
, Kozo Tanno
d
,
e
, Naoyuki Nishiya
e
,
f
,
Hisatomi Arima
c
a
Department of Industrial Economics, Faculty of Economics, Fukuoka University, Fukuoka, Japan
b
Department of Research Planning and Information Management, Fukuoka Institute of Health and Environmental Sciences, Fukuoka, Japan
c
Department of Preventive Medicine and Public Health, Faculty of Medicine, Fukuoka University, Fukuoka, Japan
d
Department of Hygiene and Preventive Medicine, Iwate Medical University, Iwate, Japan
e
Iwate Tohoku Medical Megabank Organization, Iwate Medical University, Iwate, Japan
f
Division of Integrated Information for Pharmaceutical Sciences, Department of Clinical Pharmacy, Iwate Medical University School of Pharmacy, Iwate, Japan
ARTICLE INFO
Keywords:
COVID-19
Diabetes
Doctor visits
Japan
Socioeconomic disparities
ABSTRACT:
How do people change their healthcare behavior when a public health crisis occurs? Within a year of its
emergence, coronavirus disease 2019 (COVID-19) has gradually inltrated our lives and altered our lifestyles,
including our healthcare behaviors. In Japan, which faces China across the East China Sea and accepted 924,800
Chinese tourists in January 2020, the emergence and spread of COVID-19 provides a unique opportunity to study
peoples reactions and adaptations to a pandemic.
Patients with chronic illnesses who require regular doctor visits are particularly affected by such crises. We
focused on diabetic patients whose delay in routine healthcare invites life-threatening complications and
examined how their patterns of doctor visits changed and how demographic, socioeconomic, and vital factors
disparately affected this process. We relied on the insurance claims data of a health insurance association in
Tokyo. By using panel data of diabetic patients from April 2018 to September 2020, we performed visual in-
vestigations and conditional logistic regressions controlling for all time-invariant individual characteristics.
Contrary to the general notion that the change in healthcare behavior correlates with the actual spread of the
pandemic, the graphical and statistical results both showed that diabetic patients started reducing their doctor
visits during the early stage of the pandemic. Furthermore, a substantial decrease in doctor visits was observed in
women, and large to moderate reductions were seen in patients who take insulin and are of advanced age, who
are at high risk of developing severe COVID-19. By contrast, no differentiated effect was found in terms of income
status. We further investigated why a change in pattern occurred for each subgroup.
The patterns of routine healthcare revealed by this study can contribute to the improvement of communication
with the target population, the delivery of necessary healthcare resources, and the provision of appropriate
responses to future pandemics. (299 words).
1. Introduction
Since coronavirus disease 2019 (COVID-19) was rst reported in
Wuhan, the infection has damaged and ruined peoples lives worldwide.
Despite the efforts of many countries to contain this contagious disease,
no promising treatment has been found, and COVID-19 is still spreading
throughout the world. Most young, healthy people infected with the new
coronavirus experience only a mild to moderate respiratory illness and
will recover without any special treatment (WHO n.d.). However, the
elderly and people with underlying medical problems, such as cardio-
vascular disease, diabetes, chronic respiratory disease, and malignancy,
are more likely to develop severe pneumonia (WHO n.d.; Onder et al.,
2020).
Although chronic diseases require regular visits, monitoring, and
care to control, there is growing concern that COVID-19 might invite
care avoidance or temporary disruptions in routine and non-emergency
medical care access(Czeisler et al., 2020; Moroni et al., 2020). A study
reported that 31.5% of US adults avoided routine medical care during
* Corresponding author. 8-19-1 Nanakuma, Jonan-ku, Fukuoka City, Fukuoka, 814-0180, Japan.
E-mail address: masatakaharada@fukuoka-u.ac.jp (M. Harada).
Contents lists available at ScienceDirect
SSM - Population Health
journal homepage: www.elsevier.com/locate/ssmph
https://doi.org/10.1016/j.ssmph.2021.100961
Received 12 August 2021; Received in revised form 6 November 2021; Accepted 7 November 2021
SSM - Population Health 16 (2021) 100961
2
the pandemic because of COVID-19 (Czeisler et al., 2020). In addition,
people with underlying medical conditions are more likely to avoid
routine care visits (Czeisler et al., 2020). Therefore, medical care
avoidance triggered by COVID-19 can exacerbate chronic diseases and
other related symptoms.
At the same time, how severely a dropout from routine medical care
affects ones state of the disease varies by the type of chronic illness.
Therefore, this study focused on the healthcare behaviors of diabetic
patients. Diabetes is one of the most common life-threatening chronic
diseases and one of the primary causes of serious complications such as
atherosclerotic cardiovascular disease and microvascular disease (i.e.,
retinopathy, nephropathy, and neuropathy), particularly when patients
miss adequate care management (American Diabetes Association 2021a,
2021b). Diabetic patients are also susceptible and vulnerable to
COVID-19 (Fang et al., 2020) and may withhold or reduce their visits for
routine healthcare. However, studies examining the healthcare behavior
of diabetic patients are still scarce. Therefore, it is worth exploring their
changing patterns of routine medical care during the COVID-19
pandemic.
An important aspect of this pandemic is the dynamic and gradual
nature of its inltration into our lives. It took more than two months for
pneumonia of unknown cause in Wuhan, which was rst reported in
detail by WHO on January 5, 2020 (World Health Organization, 2020),
to be declared as a pandemic on March 11, 2020. Meanwhile, people
started altering their healthcare behaviorfrom those who perceived
greater risks from COVID-19 to those perceiving lesser risks. Our risk
calculus is also driven in part by emotion and not just by the death toll
from COVID-19. A celebritys death sometimes has a signicant impact
on human behavior (Toriumi et al., 2020). Each key event has gradually
changed peoples perceptions and behaviors toward COVID-19. Japan,
which faces China across the East China Sea and accepted 924,800
visitors from China in January 2020 (partly owing to the Chinese New
Year on January 25) (Japan National Tourism Organization 2020), has
been involved in this pandemic since the beginning. Therefore, Japan is
an ideal place to study how diabetic patients have reacted and adapted
themselves to the spread of COVID-19 since the beginning of the
pandemic.
To provide further background during the pandemic, Japans rst
patient of COVID-19 was reported on January 16, and the rst large-
scale infection with SARS-CoV-2, or cluster, was conrmed in the Dia-
mond Princess (a cruise ship) on February 1. On February 27, all the
Japanese public schools were closed. In response to the growing COVID-
19 infection, a state of emergency was declared to Tokyo and several
prefectures on April 7, was extended to the remaining prefectures in the
following week and was lifted on May 25. Note, however, that the state
of emergency only asks for voluntary restraint on nonessential outings
and does not have coercive power to limit peoples behavior, unlike
lockdown. Japan marked the rst-wave record of the daily COVID-19
deaths of 31 on May 2, and the second wave peaked at the end of
July. Though the second wave was larger than the rst wave, the gov-
ernment did not release the state of emergency (Nippon dot com n.d.).
The number of deaths due to COVID-19 had been maintained at a very
low level compared to many other countries for the rst several months
of the pandemic. The death toll due to COVID-19 in Tokyo and its sur-
rounding three prefectures (Saitama, Chiba, and Yokohama) as of
September 30, 2020 (when our data coverage ends) was only 719, or the
death rate of 0.0019% (Nippon H¯
os¯
o Ky¯
okai n.d.; Statistics Bureau,
Ministry of Internal Affairs and Communications 2021, Calculated by
the authors.) In summary, it is unlikely that a government policy or an
overwhelmed healthcare system coercively limits access to healthcare,
although patients might have voluntarily refrained from doctor visits
given their surroundings.
To investigate when the decline in routine medical care started, we
rst tested the hypothesis that H1: doctor visits among diabetic patients
started declining since January 2020, shortly after COVID-19 was rst re-
ported by observing the trend of doctor visits of all diabetic patients.
Thereafter, in terms of routine visits as a product of an individuals
decision-making process, people with different backgrounds would react
differently to COVID-19. Therefore, the second hypothesis posits that
H2: patients who are at a higher risk of severe COVID-19 were more likely to
reduce the frequency of doctor visits. Our data contain demographic, so-
cioeconomic, and vital variables, each of which is more or less associ-
ated with the risk factors of COVID-19. We utilized these variables as
touchstones to uncover the potential mechanisms underlying the dif-
ferential behaviors.
2. Data and methods
2.1. Data
For our analysis, we rely on the data offered by a health insurance
society in Tokyo, where primary policyholders work in the ground
transportation industry. Japans universal healthcare is provided by
several systems, and one of these systems allows a group of companies
(usually in the same industry) to become a joint insurer.
Specically, we obtained insurance claims data from October 2016
to September 2020. Considering that insurance claims are reported
monthly in Japan, we created a monthly panel dataset. For the unit of
observation, we selected individuals diagnosed with diabetes by using
diagnosis codes from 84,907 members whose enrollment status was
conrmed as of March 31, 2020. There were 1110 and 107 individuals
diagnosed with diabetes before and after April 1, 2018, respectively.
Using the information of these 1217 patients, we created a monthly
panel dataset ranging from April 2018 to September 2020. The 107 in-
dividuals diagnosed with diabetes after April 2018 were included in the
sample after their rst diagnosis, which resulted in 35,043 monthsunit
observations.
Among these patients, 201 patients who visited a doctor every month
during the period of analysis and 37 patients with missing income values
were excluded from the regressions with unit xed effects, thus resulting
in 979 patients and 28,401 monthunit observations in Table 1. To
adjust the difference in the number of days across months in the sta-
tistical analysis, we once transform monthly data to daily data and hy-
pothetically set all visits that were made in the middle of the month.
Finally, a couple of supplemental remarks on this dataset. First, in-
surance claims are submitted by healthcare providers in Japan. Second,
we do not have the records of the treatments for COVID-19, which are
100% covered by the government. However, COVID infection does not
Table 1
Descriptive statistics.
Mean SD Min Max
Doctor Visits within a Month 0.677 0.468 0.000 1.000
MoCount 6.069 8.771 21.017 8.450
AfterMoCount 1.354 2.496 0.000 8.450
Women 0.209 0.407 0.000 1.000
Dependent 0.144 0.351 0.000 1.000
Age as of March 31, 2019 57.5 8.46 24.0 73.0
Monthly Salary as of September
30, 2020
375,497 189,390 58,000 1,390,000
Dipeptidyl-peptidase IV inhibitor 0.353 0.478 0.000 1.000
Alpha-glucosidase inhibitors 0.086 0.281 0.000 1.000
Sulfonylurea 0.178 0.382 0.000 1.000
Biguanide 0.280 0.449 0.000 1.000
Thiazolidine 0.064 0.244 0.000 1.000
Glinide 0.025 0.155 0.000 1.000
Glucagon-like peptide-1 receptor
agonists
0.025 0.158 0.000 1.000
Insulin 0.065 0.247 0.000 1.000
Sodium-glucose cotransporter-2
inhibitors
0.196 0.397 0.000 1.000
Compounding Agents 0.237 0.425 0.000 1.000
N of monthunit observations 28,401
Data covered the period from April 2018 to September 2020.
M. Harada et al.
SSM - Population Health 16 (2021) 100961
3
affect the membership status of the health insurance. On the other hand,
our dataset does not distinguish those who dropped out from routine
diabetic care and those who died of COVID-19 during April 2020 and
September 2020. However, Japans death rate due to COVID-19 was
very low (0.002%) during this period. Therefore, a simple calculation
in Appendix C shows that the expected number of deaths in our sample is
only about 0.14, which is negligible.
2.2. Method for graphical analysis
To visually inspect the dynamic change in doctor visits after the
emergence of COVID-19, we utilized the local mean smoothing tech-
nique with a triangular kernel and used a bandwidth of four. In other
words, the weights for neighboring months linearly decrease and
become zero before or after four months. The selection of the kernel does
not affect the shape of local mean smoothing curves. The bandwidth was
set at four because it is rare in Japan that the length of a prescription
exceeds three months. The outcome variable, namely Visitit, is an indi-
cator of whether a patient i visited a doctor in a given monthyear t. We
code Visitit =1 when at least one drug for diabetes was prescribed to the
patient. Thus, we counted all visits made for diabetes treatment, but this
does not exclude the possibility that the patient consulted a doctor for
another illness at the same time.
2.3. Estimation
Regression analyses were performed to conrm the results of the
graphical analyses controlling for various confounding factors. We rst
model the overall pattern of doctor visits with the full sample to test the
rst hypothesis that the doctor visits among diabetic patients started
declining at the beginning of the COVID-19 pandemic by estimating the
following logistic model:
where the upper equation shows the relationship between the outcome
and covariates, and the lower equation shows the content of the cova-
riates. The outcome variable Visitit is dened in the previous subsection.
MoCountt refers to the number of days as of the reference date, January
6, 2020, when the Ministry of Health, Labour and Welfare (henceforth,
MHLW) of Japan rst reported pneumonia of unknown cause (Ministry
of Health, Labour and Welfare, 2020). Note that this value takes nega-
tive values before the reference date and is divided by 30 to interpret the
results with respect to a month. We expect the doctor visits to decrease
in this variable throughout the observed period because of the dropouts
from regular outpatient care, most of which occurs within a few visits
after the rst diagnosis (Masuda et al., 2006). We also expect that the
drop size decreases because more and more individuals in the sample
become either regular outpatients or constant dropouts. Also, the coef-
cient of MoCountt, or β1, represents the average slope calculated dur-
ing the pre-COVID period (Apr. 2018Dec. 2019) after control.
By contrast, AftertMoCountt is our primary explanatory variable and
the interaction term of between Aftert and MoCountt, where Aftert is the
dummy variable of whether the doctor visits were made after the rst
report of COVID-19 on January 6, 2020. Given that the rst ve days in
2020 include New Years holidays and weekends, we coded Aftert=1
for all doctor visits made in January 2020. Therefore, the coefcient of
this variable indicates how much additional decrease in the rate of
doctor visits since the emergence of COVID-19. The rst hypothesis is
supported if the coefcient is negative and statistically signicant (i.e.
β2<0 or ORβ2<1).
Dit is a set of 10 indicators of whether a patient i was prescribed a
given diabetes drug in the last visit. These 10 drugs include alpha-
glucosidase inhibitors, biguanide, dipeptidyl-peptidase IV inhibitors,
glinide, glucagon-like peptide-1 receptor agonists, insulin, sodium-
glucose cotransporter-2 inhibitors, sulfonylurea, thiazolidine, and
compounding agents. The values of these indicators were updated dur-
ing the next visit. Therefore, Dit represents the type of drug prescribed.
We regard patients prescribed with insulin as being in a relatively
advanced condition (American Diabetes Association, 2021c).
Finally,
α
i and φt are the xed effects for patients and months,
respectively. The month xed effects (φt) enable us to make seasonal
adjustments to doctor visits: human factors such as life and business
cycles of the insured and environmental factors such as temperature,
humidity, and pollution or allergen level in the air. The patient xed
effects (
α
i) control for the effects of all observed and unobserved indi-
vidual characteristics that are constant during the period of analysis (i.
e., October 2018September 2020). Such characteristics include genes,
socioeconomic environments determined prior to October 2018, and
pre-existing illnesses. Given that logistic regression with the unit xed
effects yields biases with the ordinary maximum likelihood estimation,
the models are estimated using a conditional likelihood approach.
To test the second hypothesis that patients at higher risk of severe
COVID-19 were more likely to reduce the frequency of doctor visits, we
analyzed the differential effects according to the following criteria: (1)
men or women, (2) younger or older (dened as below or above the
median), (3) higher or lower income (dened as below or above the
median), and (4) never prescribed insulin or ever prescribed insulin. We
regard those who fall in the second category (i.e., women, older patients,
lower-income patients, and those prescribed insulin) as the target sub-
sample and those who fall in the rst category as the control subsample.
As a proxy of patientsincome, we use standard monthly remuneration,
which is calculated from policyholdersearnings and is used to deter-
mine the premium for the beneciaries. Age and income were measured
as of March 31, 2019, and September 30, 2020. To prioritize compati-
bility with visual evidence, age and income were dichotomized at their
median values. Thus, we included the interaction term between each of
the aforementioned four dummy variables and two other variables,
MoCountt and AftertMoCountt, in Eq. (1) in the model:
P(Visitit =1|xit) = exp(f(xit ))
1+exp(f(xit)) f(xit ) = β1MoCountt+β2Aftert⋅MoCountt+δDit +
α
i+Φt,(1)
f(xit) = β1Ctrli+β2Ctrli⋅MoCountt+β3Ctrli⋅Aftert⋅MoCountt+β4Trgti+β5Trgti⋅MoCountt+β6Trgti⋅Aftert⋅MoCountt+δDit +
α
i+Φt,(2)
M. Harada et al.
SSM - Population Health 16 (2021) 100961
4
where Ctrli and Trgti are the dummy indicators of whether a patient
belongs to the control subsample and the target subsample, respectively.
The subscript t is added to these indicators when the sample is divided
on the basis of insulin prescription because some patients started taking
insulin during the observed period. The coefcients of the triple inter-
action terms, namely, β3 and β6, represent the additional change in the
rate of doctor visits after the emergence of COVID-19 for each of the
control and target groups, which are the subsample equivalent of β2 in
Eq. (1).
3. Results
Table 1 presents the descriptive statistics of all variables used in our
regression analysis. The rst row shows that the mean of doctor visits
within a month is 0.677, thus suggesting that two-thirds of patients saw
a doctor at least once per month. The minimum and maximum of
MoCount in the second row roughly correspond to the middle of April
2018 and September 2020, respectively. As discussed earlier, the drug
variables starting from the eighth row refer to the type of drug pre-
scribed in the last visit, and they show a low proportion of insulin and
high proportions of sulfonylurea, biguanide, sodium-glucose cotrans-
porter-2 inhibitor, and compounding agents.
3.1. Full-sample results
Fig. 1 shows how the emergence of COVID-19 caused behavioral
changes among diabetic patients in Japan; the vertical axis represents
the average proportion of patients with diabetes who visited a doctors
ofce at least once in a given month. The curve of local mean smoothing
started declining after the emergence of COVID-19, thus indicating that
patients visited doctors less frequently after the emergence of the
disease.
Note that different mechanisms caused the decline in doctor visits
during the rst several months and the second one in 2020. The former
was caused by the dropouts of newly diagnosed diabetic patients and the
hot summer in 2018. Indeed, a study with diabetic patients in Japan
showed that only one in ten individuals diagnosed with diabetes
routinely visited a doctor and most patients dropped out within three or
fewer visits (Masuda et al., 2006). We discuss this point in more detail in
Appendix B.
We performed regression analysis to conrm this visual pattern
while statistically controlling for various potential confounders. Table 2
reports the results of the logistic analysis with patient xed effects with
different sets of control variables (see Table A1 for the same table in
coefcients and standard errors). The upper rows show the odds ratios,
and the lower rows show the 95% condence intervals. The odds ratios
in the rst and third rows in all columns show that the odds ratios
remain almost unchanged regardless of whether we control for monthly
dummies and last-prescribed drugs.
The odds ratios for β1 and β2 are both less than one and statistically
signicant across all models, thus suggesting that patients became less
likely to see a doctor as time passes. The onset of the COVID-19
pandemic further reinforced this tendency. Note that although Fig. 1
shows the at curve several months before the emergence of COVID-19,
our estimate for β1 represents the average slope until December 2019
after controlling for other covariates such as month and unit xed ef-
fects. Therefore, the predicted downward trend of β2 is not an artifact of
this at slope. According to model (3) with the full set of control vari-
ables, the odds ratios for MoCountt and AftertMoCountt are 0.982 and
0.975, respectively, thus implying that the average doctor visits for
patients one month after COVID-19 emergence is 4.3 (={(1 0.982) +
(1 0.975)} ×100) percentage points(p.p.) lower than the outcome
value predicted from the model in the month immediately before the
emergence, and 2.5 p.p. is accounted for by the COVID-19 pandemic.
3.2. Subsample results
We now focus on the differences of the trends between subsamples
divided by sex, age, income, and progression of diabetes. We rst
analyzed the local mean smoothing plots to identify visual clues and
then relied on regression analysis to test the hypotheses statistically.
Fig. 2 shows the local mean smoothing plots of the differential effects of
four key variables. The red dashed lines represent the local means of the
target subsamples, whereas the blue solid lines refer to the counterparts
of the control subsamples. The gure shows that the most signicant
reductions in doctor visits occurred in female patients (upper left) and
those who were prescribed insulin (upper right). The older patients
(lower left) show a moderate decline compared with the younger sub-
population, whereas the division according to income shows no differ-
ential effects on doctor visits.
Table 3 examines these visual relationships statistically (see Table A2
for the same table in coefcients and standard errors). The differential
effects of the emergence of COVID-19 based on sex, age, income, and
diabetes progression are reported in the rst to fourth columns. Overall,
the results of the statistical tests correspond to the visual evidence.
Specically, H0:β3=0 was not rejected in all models at 5% level, thus
Fig. 1. The change in the proportion of doctor visits using the full sample of
people diagnosed with diabetes. The curve is drawn with the local mean
smoothing of doctor visits from April 2018 to September 2020. The red vertical
line indicates January 1, 2020. (For interpretation of the references to colour in
this gure legend, the reader is referred to the Web version of this article.)
Table 2
Estimated effects of COVID-19 on doctor visits from April 2018 to September
2020 using the full sample of people diagnosed with diabetes.
Outcome: doctor
visits
(1) (2) (3)
exp(β
1
): MoCount 0.984***
[0.978,0.990]
0.983***
[0.977,0.990]
0.982***
[0.976,0.989]
exp(β
2
):
AfterMoCount
0.969***
[0.950,0.987]
0.976*
[0.956,0.997]
0.975*
[0.954,0.996]
H
0
: β
2
=0 p =.001 p =.024 p =.018
Patient xed effects
Monthly dummies
Last-prescribed
drugs
N of units 979 979 979
N of monthunit
observations
28,401 28,401 28,401
Conditional logistic regressions with heteroscedasticity robust standard errors
clustered by patient ID. Top rows: odds ratios; bottom rows: 95% condence
intervals.
* p < .05, ** p < .01, *** p < .001 with a two-tailed t-test.
M. Harada et al.
SSM - Population Health 16 (2021) 100961
5
suggesting that the additional decrease in medical care due to the
emergence of COVID-19 was not conrmed in a statistical sense for the
control subsamples. This point becomes more apparent when the local
mean smoothing is drawn with the bandwidth selected according to the
Rule of Thumb estimator (Silverman, 1986) to avoid over-tting.
Figure A1 shows that the curves of all control subsamples are drawn
as a straight line.
On the other hand, H0:β6=0 was rejected in three models with
β6<0 or ORβ6<1, thus indicating that those who belong to the target
subsamples further reduced their doctor visits after the emergence of
COVID-19. More specically, the average visit one month later is 6.9 p.
p. lower for female patients, 4.0 p.p. lower for older patients, and 7.7 p.
p. lower for those who were ever prescribed insulin, excluding the
impact of β5, or the temporal attrition of the doctor visits, than the
average doctor visits before the emergence of COVID-19.
Finally, we tested whether the control and target subsamples
reduced routine medical care at different paces after the emergence of
COVID-19 by testing H0:β2+β3=β5+β6. The sixth row from the
bottom of Table 3 shows that the smallest p-value (p =.028) is in the rst
column, which examines the differential effects of sex. This is followed
by insulin prescription (p =.040) and age (p =.094), although the
difference in age was not statistically signicant. Income does not have a
differential effect (p =.939).
4. Discussion
4.1. Substantive impact of the emergence of COVID-19
This study investigated the relationship between the emergence of
COVID-19 and the frequency of doctor visits among diabetic patients
with the following two hypotheses: (1) diabetic patients started reducing
routine medical care after the emergence of COVID-19, and (2) this
tendency was more prominent in the subgroups that were susceptible to
COVID-19.
We rst estimated the relationships between the emergence of
COVID-19 and doctor visits by using the entire sample to test the rst
hypothesis. The visual and statistical evidence both indicated that dia-
betic patients started decreasing their doctor visits after January 2020.
The magnitude of the impact is not negligible. The odds ratio for the
additional reduction in doctor visits (AftertMoCountt) was 0.975, thus
Fig. 2. The change in the proportion of doctor visits using the subsamples. The subsamples were divided by sex (upper left), age (lower left), income (lower right),
and diabetes progression (upper right). The curve is drawn with the local mean smoothing of doctor visits between April 2018 and September 2020. The red vertical
line indicates January 1, 2020. (For interpretation of the references to colour in this gure legend, the reader is referred to the Web version of this article.)
M. Harada et al.
SSM - Population Health 16 (2021) 100961
6
implying that the average rate of doctor visits decreases by 2.5 p.p. every
month because of the spread of COVID-19.
4.2. Findings as expected
We then examined the differential effects of the emergence of
COVID-19 to test the second hypothesis. Our ndings contain expected
and somewhat surprising results. Previous studies have shown that the
elderly and diabetic patients are at a higher risk of severe COVID-19
(Zhou et al., 2020). Not surprisingly, the graphical and statistical
pieces of evidence consistently showed that older patients and patients
prescribed with insulin reduced their doctor visits at a higher pace than
the younger patients and patients who were never prescribed insulin,
although the impact on older patients is somewhat weak.
The reduction in doctor visits requires close attention if it implies the
medically undesirable discontinuation of routine care, but it is a prac-
tical compromise if the interval of regular visits was widened following
the doctors advice. Fig. 3 was used to distinguish between these two
possibilities. The left and right panels show the subsample smoothing
plots for the insulin-prescribed patients and older patients respectively.
Table 3
Estimated effects of COVID-19 on doctor visits between April 2018 to September 2020 using the subsamples divided by sex, age, income, and diabetes progression.
Outcome: doctor visits (1) (2) (3) (4)
Control Subsample: Men Age <50% Income 50% Never Insulin
Target Subsample: Women Age 50% Age 50% Ever Insulin
exp(β
1
): Ctrl Subsumed Subsumed Subsumed 0.898
[0.641,1.256]
exp(β
2
): CtrlMoCount 0.980*** 0.980*** 0.979*** 0.981***
[0.972,0.987] [0.970,0.990] [0.970,0.988] [0.974,0.988]
exp(β
3
): CtrlAfterMoCount 0.987 0.991 0.978 0.981
[0.964,1.010] [0.962,1.021] [0.951,1.005] [0.959,1.003]
exp(β
4
): Trgt Subsumed Subsumed Subsumed Subsumed
exp(β
5
): TrgtMoCount 0.992 0.984*** 0.986** 0.993
[0.979,1.005] [0.976,0.992] [0.977,0.995] [0.978,1.008]
exp(β
6
): TrgtAfterMoCount 0.931** 0.960** 0.972 0.923**
[0.892,0.972] [0.934,0.987] [0.943,1.001] [0.876,0.972]
H
0
: β
3
=0 p =.266 p =.549 p =.112 p =.091
H
0
: β
6
=0 p =.001 p =.004 p =.057 p =.002
H
0
: β
2
+β
3
=β
5
+β
6
p =.028 p =.094 p =.939 p =.040
Patient xed effects
Monthly dummies
Last-prescribed drugs
N of units 979 979 979 979
N of monthunit observations 28,401 28,401 28,401 28,401
Conditional logistic regressions with heteroscedasticity robust standard errors clustered by patient ID. Top rows: odds ratios; bottom rows: 95% condence intervals.
* p < .05, ** p < .01, *** p < .001 with a two-tailed t-test.
Fig. 3. The change in the proportion of doctor visits for patients prescribed with insulin (left) and above median in age (right). The outcome variables take the value
of 1if a patient saw a doctor in a given month (red short-dashed line), within the past two months (pink dashed line), within the past four months (purple long-
dashed line), and within the past six months (blue solid line). (For interpretation of the references to colour in this gure legend, the reader is referred to the Web
version of this article.)
M. Harada et al.
SSM - Population Health 16 (2021) 100961
7
The red short-dashed lines represent the reproduction in Fig. 2. The pink
dashed lines, purple long-dashed lines, and blue solid lines are created
by changing the outcome variable to whether a patient saw a doctor at
least once in the past two, four, or six months, respectively.
Both panels show that the fraction of those who saw a doctor at least
once in two or more months was quite high. However, we observed a
similar reduction in the outcome variable after January 2020 in both
panels when the outcome variables include whether a patient visited a
doctor at least once in the past two or four months (i.e., the two dashed
lines in the middle), thus implying that some patients were not able to
visit a doctor once in four months. The sharp bends after the emergence
of COVID-19 became visually negligible when the outcome variable was
whether a patient visited a doctor at least once in the past six months.
This indicates that patients with a high-risk prole for COVID-19 seemed
to adapt to this crisis by setting an unconventionally long interval for
doctor visits. For example, patients who require insulin probably con-
sulted a doctor to obtain multi-month prescriptions because their health
condition would be hard to sustain without insulin. These results imply
that among the high-risk group, it is unlikely that the reduction in doctor
visits caused an increase in mortality rates, which is also consistent with
the nding that the dropout rate for diabetic patients who require in-
sulin injection was much lower compared to diabetic patients with
milder symptoms (Masuda et al., 2006).
These results imply that some diabetic patients probably took pre-
cautionary actions from an early stage of COVID-19 and relied on
available information at that time rather than just waiting for the clin-
ical risk evaluations of COVID-19 to become available around April
2020 (Fang et al., 2020; Guo et al., 2020). Indeed, information on the
types of people who develop severe COVID-19 started circulating when
Japans rst case was conrmed on January 16 (Ministry of Health,
Labour and Welfare, 2020). The news was covered by a variety of media
groups with background knowledge of past epidemic diseases. For
example, a news article covering the rst Japanese case already reported
that the infections of those who have chronic diseases and the elderly
may become severe (Yomiuri Shimbun, 2020).The rst government
expert meeting on February 16 attracted considerable media coverage,
where advanced age and chronic illnesses, including diabetes, were
clearly mentioned as the risk factors for COVID-19 (Asahi Shimbun,
2020; Yomiuri Shimbun, 2020).
It also comes as no surprise that virtually no differential effects were
found between higher- and lower-income patients because income is not
related to the exacerbation of COVID-19. At the same time, this nding
highlights the importance of available healthcare options during a state
of emergency to sustain routine care. If the universal healthcare system
was not introduced in Japan, we might have observed differential effects
between the rich and the poor because the availability and quality of
healthcare are known to affect access to prompt doctor visits (Czeisler
et al., 2020).
4.3. An unexpected nding and exploration
However, the analysis of the differential effects of sex revealed
somewhat unexpected ndings. From a medical perspective, the con-
ditions of coronavirus-infected men are more likely to deteriorate than
their counterparts (Gebhard et al., 2020; Scully et al., 2020). Therefore,
male diabetic patients have more reasons to avoid routine medical care
than women. However, our analysis revealed that the subsamples of
women reduced doctor visits at a higher pace than their counterpart.
Therefore, the differential effects between men and women stemmed
from causes other than medical ones. What were they?
Although it is beyond the scope of this paper to test all potential
causes, we propose one testable hypothesis: increased exposure to media
coverage of COVID-19 discouraged female patients from visiting a
doctor. During the pandemic, the show business industry was particu-
larly hardhit. Lesser entertainment news became available, and news
coverage of COVID-19 lled the gap. As Americans watched Dr. Fauci
every day during the pandemic, several specialists appeared in Japanese
TV shows every day to disseminate information about this disease.
Moreover, the governments expert meeting asked citizens to refrain
from going out for non-essential businesses as early as February 17,
2020 (Asahi Shimbun Feb. 17, 2020), thus further boosting the view-
ership of these programs.
If the increased news intake is associated with less frequent doctor
visits, female patients with a dependent status would have reduced their
doctor visits more because they had more time to watch these news
shows than female workers. Therefore, we analyzed the differential ef-
fects between policyholders and dependents within the subsample of
women. Fig. 4 shows that the local mean smoothing curves look similar
between these two groups, thus indicating the difference in news intake
does not explain the womens reduced visits after the emergence of
COVID-19. When we used Eq. (2) with the subsample of women and
tested whether the size of the slope after the emergence of COVID-19
differs between female policyholders and female dependents, the null
hypothesis H0:β2+β3=β5+β6 cannot be rejected with p =.939.
The result dees our tentative proposition but agrees with a recent
study that found no signicant direct association between social media
consumption and preventive behavior (Liu, 2021). Research on mass
media also suggests that increased exposure to a particular media
coverage cannot cause conforming behavior in its audience but may
have limited yet diverse effects (Perse & Lambe,2016). It can be inferred
that the reduction in doctor visits among female patients is related to
their sex. A study reported that higher rates of medication non-
adherence due to costsare seen among female diabetic patients in the
United States (Bhuyan et al., 2018). However, an internet survey of US
adults (Czeisler et al., 2020) and our study show that income inequality
Fig. 4. The differential effects of the emergence of COVID-19 on doctor visits
with the subsamples of women. The subsamples are divided by membership
status (policy-holder/dependent). The red vertical line indicates January 1,
2020. See the footnote of the graph for the bandwidths used to draw each curve.
(For interpretation of the references to colour in this gure legend, the reader is
referred to the Web version of this article.)
M. Harada et al.
SSM - Population Health 16 (2021) 100961
8
within female patients is unlikely to be a driver of reduced routine care.
On the other hand, several studies show that women are more risk averse
than men during this pandemic (Haischer et al., 2020) and in general
(Jianakoplos & Bernasek,1998; Rosen et al., 2003). Given that
pre-existing conditions often exacerbate COVID-19, it is not surprising
that female diabetic patients responded to this pandemic in a highly
risk-averse manner.
This study has several limitations. First, the exact dates of doctor
visits and prescription lengths are missing to protect the privacy of the
insured. This limitation may result in less accurate measurements. Sec-
ond, readers may need to carefully evaluate the external validity of our
ndings. Our data come from a single joint insurer, where all members
were under 75 years old and either they or their dependents worked in
the ground shipping industry. Although we believe that this limitation
might make some of the differentiating effects more conservative and
would not change our substantive ndings, the generalizability of the
results might be compromised. Similarly, in the time horizon, it covers
only the rst several months after the emergence of COVID-19. Thus, we
cannot examine the trajectories of their reduced doctor visits in the
middle and long terms. Lastly, we cannot necessarily judge in this study
whether fewer doctor visits reduce healthcare quality. The reduction in
doctor visits among the older and insulin-prescribed patients due to the
increased risk of infection was inevitable in this pandemic but was surely
alarming because they require careful monitoring. By contrast, as sug-
gested in Fig. 3, some patients and doctors might have arranged a
reduced format of routine care without critically compromising
healthcare quality.
5. Conclusion
In conclusion, our study contributes to the existing literature in two
ways. First, our study lls the gap between the actual spread of COVID-
19 and the healthcare behaviors of people. We show a dynamic reduc-
tion in routine medical care among diabetic patients after the emergence
of COVID-19. Second, our study explains the factors that play an
essential role in healthcare avoidance by using health insurance claims,
which are behavioral-based measurements and augment the ndings via
opinion surveys (Czeisler et al., 2020). We found that lifestyle and
economic factors have no signicant impact, patients with high-risk
factors have large behavioral changes, and sex-related factors play a
crucial role. These ndings facilitate a deeper understanding of human
behaviors in response to this public health crisis, thus contributing to the
improvement of communications with the target population, the de-
livery of necessary healthcare resources, and the provision of appro-
priate responses to future pandemics.
CRediT authorship contribution statement
Masataka Harada: Conceptualization, Methodology, Software,
Writing original draft, Writing review & editing, Visualization.
Takumi Nishi: Software, Data curation, Writing review & editing.
Toshiki Maeda: Methodology, Writing review & editing, Project
administration. Kozo Tanno: Investigation, Resources, Writing review
& editing. Naoyuki Nishiya: Investigation, Resources, Writing review
& editing, and. Hisatomi Arima: Writing review & editing, Supervi-
sion, Project administration.
Declaration of competing interest
None.
Acknowledgements
We thank Tokyo Kamotsu-Unso Kenko-Hoken-Kumiai for providing
us with the valuable data, and Kei Asayama, Nagako Okuda, Daisuke
Sugiyama, Hiroshi Yatsuya, and Akira Okayama for their invaluable
support and contributions to this study. We also thank Elsevier Language
Editing Services for editing a draft of this manuscript. This work was
supported by JSPS KAKENHI Grant Numbers 20K20674 and 21K17317.
Appendix A. Appendix Figures and Tables
Table A.1
Replication of Table 2 with regression coefcients and standard errors.
Outcome: doctor visits (1) (2) (3)
Estimation Methods Cond.FE Cond.FE Cond.FE
β
1
: MoCount 0.016*** 0.017*** 0.018***
(0.003) (0.003) (0.003)
β
2
: AfterMoCount 0.032*** 0.024* 0.025*
(0.010) (0.011) (0.011)
H
0
: β
2
=0 p =.001 p =.024 p =.018
Patient xed effects
Monthly dummies
Last-prescribed drugs
N of units 979 979 979
N of monthunit observations 28,401 28,401 28,401
Conditional logistic regressions with heteroscedasticity robust standard errors clustered by patient ID. Top rows:
coefcients, bottom rows: standard errors.
* p < .05, ** p < .01, *** p < .001 with a two-tailed t-test.
M. Harada et al.
SSM - Population Health 16 (2021) 100961
9
Fig. A.1. The change in the proportion of doctor visits using the subsamples with the Rule of Thumb bandwidth (Silverman, 1986). The subsamples are divided by
sex(upper left), age(lower left), income(lower right), and severity of diabetes(upper right). The curve is drawn with local mean smoothing of doctor visits between
April 2018 to September 2020. The red vertical line indicates January 1, 2020. See the footnote of the graph for the bandwidths used to draw each curve.
Table A.2
Replication of Table 3 with regression coefcients and standard errors.
Outcome: doctor visits (1) (2) (3) (4)
Estimation Methods Cond.FE Cond.FE Cond.FE Cond.FE
Control Subsample: Men Age <50% Income 50% Never Insulin
Target Subsample: Women Age 50% Income <50% Ever Insulin
β
1
: Ctrl Subsumed Subsumed Subsumed 0.108
(0.171)
β
2
: CtrlMoCount 0.021*** 0.020*** 0.022*** 0.019***
(0.004) (0.005) (0.005) (0.004)
β
3
: CtrlAfterMoCount 0.013 0.009 0.022 0.019
(0.012) (0.015) (0.014) (0.011)
β
4
: Trgt Subsumed Subsumed Subsumed Subsumed
β
5
: TrgtMoCount 0.008 0.016*** 0.014** 0.007
(0.006) (0.004) (0.005) (0.008)
β
6
: TrgtAfterMoCount 0.071** 0.040** 0.029 0.081**
(0.022) (0.014) (0.015) (0.027)
H
0
: β
3
=0 p =.266 p =.549 p =.112 p =.091
H
0
: β
6
=0 p =.001 p =.004 p =.057 p =.002
H
0
: β
2
+β
3
=β
5
+β
6
p =.028 p =.094 p =.939 p =.040
Patient xed effects
(continued on next page)
M. Harada et al.
SSM - Population Health 16 (2021) 100961
10
Table A.2 (continued )
Outcome: doctor visits (1) (2) (3) (4)
Estimation Methods Cond.FE Cond.FE Cond.FE Cond.FE
Monthly dummies
Last-prescribed drugs
N of units 979 979 979 979
N of monthunit observations 28,401 28,401 28,401 28,401
Conditional logistic regressions with heteroscedasticity robust standard errors clustered by patient ID. Top rows: coefcients, bottom rows: standard
errors.
* p < .05, ** p < .01, *** p < .001 with a two-tailed t-test.
Appendix B. Explanation for the negative slope in 2018
In this appendix, we explain the reason why we observe the negative slope in Fig. 1. In short, this drop is explained by the combination of the
dropouts from regular diabetic care and very high summer temperatures in 2018.
To illustrate these points, we drew Figure B1 below. This gure is different from Fig. 1 in that the sample is limited to the individuals who joined the
sample in October 2016, and the starting date of the observation is now October 2016. One caveat is that the steep decline in the rst several months
(Oct.2016 to Jan. 2017) is caused by the coding of the outcome variable and should be ignored. Whenever individuals joined the sample, they were
diagnosed with diabetes in doctor visits. Therefore, the outcome value of the rst observation is always coded as one.
Fig. B.1. The change in the proportion of doctor visits using the individuals who joined the sample in October 2016 with the extended period of observation. The
curve is drawn with the local mean smoothing of doctor visits from October 2016 to September 2020. The red vertical line indicates January 1, 2020.
To look at the gure, although this sample primarily consists of regular outpatients or dropouts, we still observe a moderate decline over months.
However, the slope in the pre-pandemic period is less steep compared with the post-pandemic counterpart, as well as the pre-pandemic slope in Fig. 1.
This indicates that some amount of the decline in the pre-pandemic period can be explained by the dropout patients (Masuda et al., 2006). As we
reported in Subsection 2.1, roughly 10% of the individuals joined the sample sometime after April 2018, and most of them became dropouts within
several months. Therefore, as time passed, the proportion of dropouts (and a small number of regular outpatients) increased, so the curve also
approached at.
Second, a closer look at the slope suggests that the sharp declines occurred in July 2017 and 2018, not in July 2019 and 2020. We argue that this
variation was likely to be caused by the difference in weather conditions in July. As we see in Figure B2, July temperatures in 2017 and 2018 were
exceptionally high compared to 2019 and 2020. In Tokyo, where humidity is quite high in summer, individuals with diabetes may well skip their
routine healthcare, especially if the condition is still not life-threatening. Indeed, the upper right panel of Fig. 2 shows no drop in July 2018 for the
subsample for the patients who required insulin. These seasonal variations are statistically controlled for in the regression analysis using monthly
dummy variables.
M. Harada et al.
SSM - Population Health 16 (2021) 100961
11
Fig. B.2. The bar charts of Julys average temperatures in Tokyo from 2017 to 2020.
Appendix C. Estimation of COVID-19 deaths during April 1, 2020 and September 30, 2020
In this appendix, we calculate the expected number of death due to COVID-19 during the period when our data cannot identify whether the patients
died of COVID-19 or did not see a doctor, namely from April 1, 2020, to September 30, 2020. The following calculation shows the estimated number of
death due to COVID-19 is only 0.14, which is sufciently small to ignore this issue in our inference.
In the following calculation, we relied on the following data sources: the COVID-19 data archive of NHK (Japans national broadcasting company)
for the death toll, the census for population, and the report by Ministry of Health, Labour and Welfare that was based on Japans 322,007 COVID-19
positive patients for the death rates of COVID-19 positive people with and without diabetes (Nippon H¯
os¯
o Ky¯
okai n.d.; Statistics Bureau, Ministry of
Internal Affairs and Communications n.d.; Ministry of Health, Labour and Welfare 2021).
Several caveats must be noted. First, our estimate is based on the statistics during the period mentioned above, and the area consists of Tokyo and
its surrounding three prefectures (Saitama, Chiba, and Kanagawa), where almost all members of the health insurance society live. Second, the number
of Japans viral tests has been very low, so we relied on the number of COVID-19 death rather than the number of positive cases, which is likely to be
underestimated. Third, our data cannot identify the death toll for other causes, but we statistically control for its impact, assuming the death rate of
other causes is constant and not considered in this section. Finally, we assume that the risks of getting COVID-19 are the same regardless of whether
one has diabetes or not.
The expected number of death due to COVID-19 was thus calculated as follows:
E(Death) = (Number of the Sample)
×(COVID19 Death Rate for Diabetic Patients)
= (Number of the Sample)
×(COVID19 Death Rate for the Population)
×(Multiplier of COVID19 Death Rate for Diabetic Patients)
= (Number of the Sample)
×Dearth Toll due to COVID19
Population ×Dearth rate of COVID19 positive People with Diabetes
Dearth rate of COVID19 positive People without Diabetes
=979 ×719
36,938,977 ×0.0476
0.0065
0.1395.
What this result tells us is, no matter how high the mortality risk due to COVID-19 is, our sample size is only 979, and only 719 people were dead
due to COVID-19 by the end of September 2020 in the area of study. Therefore, the resulting estimate for the death toll is also very small.
References
American Diabetes Association. (2021). 10. cardiovascular disease and risk management:
Standards of medical care in diabetes2021. Diabetes Care, 44(Supplement 1),
S125S150. https://doi.org/10.2337/dc21-S010
American Diabetes Association. (2021). 11. microvascular complications and foot care:
Standards of medical care in diabetes2021. Diabetes Care, 44(Supplement 1),
S151S167. https://doi.org/10.2337/dc21-S011
American Diabetes Association. (2021). 9. pharmacologic approaches to glycemic
treatment: Standards of medical care in diabetes2021. Diabetes Care, 44
(Supplement 1), S111S124. https://doi.org/10.2337/dc21-S009
Asahi Shimbun. (Feb. 17, 2020). Kansen shin-kyokumen seihu keikai. New phase of
infection: The Government Heightened Vigilance. http://database.asahi.com/index.sh
tml. (Accessed 11 November 2021).
Bhuyan, S. S., Shiyanbola, O., Deka, P., Isehunwa, O. O., Chandak, A., Huang, S.,
Wang, Y., Bhatt, J., Ning, L., Lin, W. J., et al. (2018). The role of gender in cost-
related medication nonadherence among patients with diabetes. The Journal of the
American Board of Family Medicine, 31(5), 743751. https://doi.org/10.3122/
jabfm.2018.05.180039
Czeisler, M.´
E., Marynak, K., Clarke, K. E., Salah, Z., Shakya, I., Thierry, J. M., ,
Weaver, M. D., et al. (2020). Delay or avoidance of medical care because of covid-
19related concernsUnited States. Morbidity & Mortality Weekly Report, 69(36),
1250. https://doi.org/10.15585/mmwr.mm6936a4. June 2020.
Fang, L., Karakiulakis, G., & Roth, M. (2020). Are patients with hypertension and
diabetes mellitus at increased risk for covid-19 infection? The Lancet. Respiratory
Medicine, 8(4), e21. https://doi.org/10.1016/S2213-2600(20)30116-8
Gebhard, C., Regitz-Zagrosek, V., Neuhauser, H. K., Morgan, R., & Klein, S. L. (2020).
Impact of sex and gender on covid-19 outcomes in Europe. Biology of Sex Differences,
11, 113. https://doi.org/10.1186/s13293-020-00304-9
Guo, W., Li, M., Dong, Y., Zhou, H., Zhang, Z., Tian, C., Qin, R., Wang, H., Shen, Y.,
Du, K., Zhao, L., Fan, H., Luo, S., & Hu, D. (2020). Diabetes is a risk factor for the
progression and prognosis of covid-19. Diabetes, 36(7), e3319. https://doi.org/
10.1002/dmrr.3319
M. Harada et al.
SSM - Population Health 16 (2021) 100961
12
Haischer, M. H., Beilfuss, R., Hart, M. R., Opielinski, L., Wrucke, D., Zirgaitis, G.,
Uhrich, T. D., & Hunter, S. K. (2020). Who is wearing a mask? Gender-, age-, and
location-related differences during the covid-19 pandemic. PLoS One, 15(10), Article
e0240785.
Japan National Tourism Organization. (2020). Honichi gaikyaku su (2020 nen 1 gatsu
suikei-chi)( Estimated number of the foreign tourists in January, 2020.). https://www.
jnto.go.jp/jpn/statistics/data_info_listing/pdf/200219_monthly.pdf [Last accessed
on 11 November 2021].
Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse? Economic
Inquiry, 36(4), 620630. https://doi.org/10.1111/j.1465-7295.1998.tb01740.x
Liu, P. L. (2021). Covid-19 information on social media and preventive behaviors:
Managing the pandemic through personal responsibility. Social Science & Medicine,
277, 113928. https://doi.org/10.1016/j.socscimed.2021.113928
Masuda, Y., Kubo, A., Kokaze, A., Yoshida, M., Sekiguchi, K., Fukuhara, N., &
Takashima, Y. (2006). Personal features and dropout from diabetic care.
Environmental Health and Preventive Medicine, 11(3), 115119.
Ministry of Health, Labour and Welfare. (2021). Shiryo 4-3: shingata korona uirusu kansen
y¯
osei-sha no j¯
ush¯
oka risuku inshi heno taiou-tou (Material 4-3: Response to the patients
who have the aggrevating factors of COVID-19). https://www.mhlw.go.jp/content/109
00000/000823697.pdf [Last accessed on 11 November 2021].
Ministry of Health, Labour and Welfare. (2020). Shingata korona uirusu ni kanren-shita
haien no kanja no hassei ni tsuite (1st case) (On the occurrence of the rst pneumonia
patient related to the New Coronavirus). https://www.mhlw.go.jp/stf/newpage_0890
6.html [Last accessed on 11 November 2021].
Ministry of Health, Labour and Welfare. (2020). Chuka jinmin kyowakoku kohoku-sho
bukan-shi ni okeru genin humei haien no hassei ni tsuite (On the occurrence of pneumonia
of unknown cause in Wuhan, Hubei, China). https://www.mhlw.go.jp/stf/newpage_0
8767.html [Last accessed on 11 November 2021].
Moroni, F., Gramegna, M., Ajello, S., Beneduce, A., Baldetti, L., Vilca, L. M.,
Cappelletti, A., Scandroglio, A. M., & Azzalini, L. (2020). Collateral damage. Journal
of the American College of Cardiology: Case Reports, 2(10), 16201624. https://doi.
org/10.1016/j.jaccas.2020.04.010
Nippon dot com (n.d.). Shingata korona uirusu kansen-sho ryuko deno kensaku kekka
(Search results of COVID-19 pandemic). https://www.nippon.com/ja/search.html
[Last accessed on 11 November 2021].
Nippon H¯
os¯
o Ky¯
okai. (n.d.). Shingata korona uirusu todouhuken no kansensha-su (The
number of cases of coronavirus by prefecture). https://www3.nhk.or.jp/news/specia
l/coronavirus/data/ [Last accessed on 11 November 2021].
Onder, G., Rezza, G., & Brusaferro, S. (2020). Case-fatality rate and characteristics of
patients dying in relation to COVID-19 in Italy. Journal of the American Medical
Association, 323(18), 17751776. https://doi.org/10.1001/jama.2020.4683
Perse, E. M., & Lambe, J. (2016). Media effects and society. Routledge.
Rosen, A. B., Tsai, J. S., & Downs, S. M. (2003). Variations in risk attitude across race,
gender, and education. Medical Decision Making, 23(6), 511517. https://doi.org/
10.1177/0272989X03258431
Scully, E. P., Havereld, J., Ursin, R. L., Tannenbaum, C., & Klein, S. L. (2020).
Considering how biological sex impacts immune responses and covid-19 outcomes.
Nature Reviews Immunology, 16. https://doi.org/10.1038/s41577-020-0348-8
Silverman, B. W. (1986). Density estimation for statistics and data analysis. CRC Press.
Statistics Bureau, Ministry of Internal Affairs and Communications. (2021). Kokusei ch¯
osa
jink¯
o sokuh¯
o sh¯
ukei (reiwa ni-nendo) (The preliminary report of the census: population in
2020). https://www.e-stat.go.jp/ [Last accessed on 11 November 2021].
Toriumi, F., Sakaki, T., & Yoshida, M. (2020). Social emotions under the spread of covid-
19 using social media. Transactions of the Japanese Society for Articial Intelligence, 35
(4). https://doi.org/10.1527/tjsai.F-K45. FK45.
World Health Organization (n.d.). Coronavirus. https://www.who.int/health-topics/co
ronavirus [Last accessed on 11 November 2021].
World Health Organization. (2020). Pneumonia of unknown cause china. https://www.
who.int/csr/don/05-january-2020-pneumonia-of-unkown-cause-china/en/ [Last
accessed on 11 November 2021].
Yomiuri Shimbun. (Feb. 17, 2020). korei” “jibyoni juten senmonka kaigi shingata haien
chinseika he giron. Expert meeting discussed a new coronavirus to curb its infection and
put emphasis on advanced ageand chronic diseases. https://database.yomiuri.co.
jp/rekishikan/. (Accessed 11 November 2021).
Yomiuri Shimbun. (Jan. 17, 2020). Shingata haien kokunai hatsu no kanja. Japans First
Patient of the New Pneumonia. https://database.yomiuri.co.jp/rekishikan/. (Accessed
11 November 2021).
Zhou, F., Yu, T., Du, R., Fan, G., Liu, Y., Liu, Z., Xiang, J., Wang, Y., Song, B., Gu, X.,
Guan, L., Wei, Y., Li, H., Wu, X., Xu, J., Tu, S., Zhang, Y., Chen, H., & Cao, B. (2020).
Clinical course and risk factors for mortality of adult inpatients with covid-19 in
wuhan, China: A retrospective cohort study. The Lancet, 395(10229), 10541062.
https://doi.org/10.1016/S0140-6736(20)30566-3
M. Harada et al.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Masks are an effective tool in combatting the spread of COVID-19, but some people still resist wearing them and mask-wearing behavior has not been experimentally studied in the United States. To understand the demographics of mask wearers and resistors, and the impact of mandates on mask-wearing behavior, we observed shoppers (n = 9935) entering retail stores during periods of June, July, and August 2020. Approximately 41% of the June sample wore a mask. At that time, the odds of an individual wearing a mask increased significantly with age and was also 1.5x greater for females than males. Additionally, the odds of observing a mask on an urban or suburban shopper were ~4x that for rural areas. Mask mandates enacted in late July and August increased mask-wearing compliance to over 90% in all groups, but a small percentage of resistors remained. Thus, gender, age, and location factor into whether shoppers in the United States wear a mask or face covering voluntarily. Additionally, mask mandates are necessary to increase mask wearing among the public to a level required to mitigate the spread of COVID-19.
Article
Full-text available
Temporary disruptions in routine and nonemergency medical care access and delivery have been observed during periods of considerable community transmission of SARS-CoV-2, the virus that causes coronavirus disease 2019 (COVID-19) (1). However, medical care delay or avoidance might increase morbidity and mortality risk associated with treatable and preventable health conditions and might contribute to reported excess deaths directly or indirectly related to COVID-19 (2). To assess delay or avoidance of urgent or emergency and routine medical care because of concerns about COVID-19, a web-based survey was administered by Qualtrics, LLC, during June 24-30, 2020, to a nationwide representative sample of U.S. adults aged ≥18 years. Overall, an estimated 40.9% of U.S. adults have avoided medical care during the pandemic because of concerns about COVID-19, including 12.0% who avoided urgent or emergency care and 31.5% who avoided routine care. The estimated prevalence of urgent or emergency care avoidance was significantly higher among the following groups: unpaid caregivers for adults* versus noncaregivers (adjusted prevalence ratio [aPR] = 2.9); persons with two or more selected underlying medical conditions† versus those without those conditions (aPR = 1.9); persons with health insurance versus those without health insurance (aPR = 1.8); non-Hispanic Black (Black) adults (aPR = 1.6) and Hispanic or Latino (Hispanic) adults (aPR = 1.5) versus non-Hispanic White (White) adults; young adults aged 18-24 years versus adults aged 25-44 years (aPR = 1.5); and persons with disabilities§ versus those without disabilities (aPR = 1.3). Given this widespread reporting of medical care avoidance because of COVID-19 concerns, especially among persons at increased risk for severe COVID-19, urgent efforts are warranted to ensure delivery of services that, if deferred, could result in patient harm. Even during the COVID-19 pandemic, persons experiencing a medical emergency should seek and be provided care without delay (3).
Article
Full-text available
Background: Emerging evidence from China suggests that coronavirus disease 2019 (COVID-19) is deadlier for infected men than women with a 2.8% fatality rate being reported in Chinese men versus 1.7% in women. Further, sex-disaggregated data for COVID-19 in several European countries show a similar number of cases between the sexes, but more severe outcomes in aged men. Case fatality is highest in men with pre-existing cardiovascular conditions. The mechanisms accounting for the reduced case fatality rate in women are currently unclear but may offer potential to develop novel risk stratification tools and therapeutic options for women and men. Content: The present review summarizes latest clinical and epidemiological evidence for gender and sex differences in COVID-19 from Europe and China. We discuss potential sex-specific mechanisms modulating the course of disease, such as hormone-regulated expression of genes encoding for the severe acute respiratory syndrome coronavirus 2 (SARS-CoV2) entry receptors angiotensin converting enzyme (ACE) 2 receptor and TMPRSS2 as well as sex hormone-driven innate and adaptive immune responses and immunoaging. Finally, we elucidate the impact of gender-specific lifestyle, health behavior, psychological stress, and socioeconomic conditions on COVID-19 and discuss sex specific aspects of antiviral therapies. Conclusion: The sex and gender disparities observed in COVID-19 vulnerability emphasize the need to better understand the impact of sex and gender on incidence and case fatality of the disease and to tailor treatment according to sex and gender. The ongoing and planned prophylactic and therapeutic treatment studies must include prospective sex- and gender-sensitive analyses.
Article
Full-text available
The COVID-19 pandemic has caused an enormous strain on healthcare systems and society on a global scale. We report a new phenomenon of medical care avoidance among patients with acute coronary syndrome, which is due to concerns about contracting SARS-CoV-2 infection during hospital stay, ultimately leading to dire clinical outcomes.
Article
Full-text available
Aim To figure out whether diabetes is a risk factor influencing the progression and prognosis of 2019 novel coronavirus disease (COVID‐19). Materials and Methods A total of 174 consecutive patients confirmed with COVID‐19 were studied. Demographic data, medical history, symptoms and signs, laboratory findings, chest computed tomography (CT) as well we treatment measures were collected and analyzed. Results We found that COVID‐19 patients without other comorbidities but with diabetes (n=24) were at higher risk of severe pneumonia, release of tissue injury‐related enzymes, excessive uncontrolled inflammation responses and hypercoagulable state associated with dysregulation of glucose metabolism. Furthermore, serum levels of inflammation related biomarkers such as IL‐6, C‐reactive protein, serum ferritine and coagulation index, D‐dimer, were significantly higher (p< 0.01) in diabetic patients compared with those without, suggesting that patients with diabetes are more susceptible to an inflammatory storm eventually leading to rapid deterioration of COVID‐19. Conclusions Our data support the notion that diabetes should be considered as a risk factor for a rapid progression and bad prognosis of COVID‐19. More intensive attention should be paid to patients with diabetes, in case of rapid deterioration. This article is protected by copyright. All rights reserved.
Article
In the face of a pandemic, social media have found to be vital information channels that might exert a positive influence on people’s preventive behaviors. However, little is known about the underlying processes that may mediate or moderate the relationship. The present study examined the relation between novel coronavirus disease (COVID-19) information consumption on social media and preventive behaviors, as well as the mediation role of personal responsibility and moderation role of health orientation. A sample of 511 online participants (mean age = 32.47 years) responded to anonymous questionnaires regarding COVID-19 information consumption on social media, health orientation, personal responsibility, and preventive behaviors. Bias-corrected bootstrap method was used to test the moderated mediation model. Results indicated that after controlling for participants’ age, gender, education, income, and insurance, personal responsibility mediated the relationship between COVID-19 information consumption on social media and preventive behaviors. Meanwhile, the direct relation between COVID-19 information consumption on social media and preventive behaviors, and the mediation effect of personal responsibility were moderated by health orientation. The present study can extend our knowledge about how risk information consumption on social media is related to one’s behavioral outcomes. Implications and limitations about the present study are also discussed.
Article
The spread of COVID-19, the so-called new coronavirus, is currently having an enormous social and economic impact on the entire world. Under such a circumstance, the spread of information about the new coronavirus on SNS is having a significant impact on economic losses and social decision-making. In this study, we investigated how the new type of coronavirus has become a social topic in Japan, and how it has been discussed. In order to determine what kind of impact it had on people, we collected and analyzed Japanese tweets containing words related to the new corona on Twitter. First, we analyzed the bias of users who tweeted. As a result, it is clear that the bias of users who tweeted about the new coronavirus almost disappeared after February 28, 2020, when the new coronavirus landed in Japan and a state of emergency was declared in Hokkaido, and the new corona became a popular topic. Second, we analyzed the emotional words included in tweets to analyze how people feel about the new coronavirus. The results show that the occurrence of a particular social event can change the emotions expressed on social media.
Article
A male bias in mortality has emerged in the COVID-19 pandemic, which is consistent with the pathogenesis of other viral infections. Biological sex differences may manifest themselves in susceptibility to infection, early pathogenesis, innate viral control, adaptive immune responses or the balance of inflammation and tissue repair in the resolution of infection. We discuss available sex-disaggregated epidemiological data from the COVID-19 pandemic, introduce sex-differential features of immunity and highlight potential sex differences underlying COVID-19 severity. We propose that sex differences in immunopathogenesis will inform mechanisms of COVID-19, identify points for therapeutic intervention and improve vaccine design and increase vaccine efficacy.