1 Clare Brown, Sean G. Young, Mick Tilford, Jenil Patel, Suman Maity, Jaimi Allen, Jyotishka Datta, Benjamin C. Amick III, Mark L. Williams* *corresponding author November 6, 2020
1
Clare Brown, Sean G. Young, Mick Tilford, Jenil Patel, Suman
Maity, Jaimi Allen, Jyotishka Datta, Benjamin C. Amick III,
Mark L. Williams*
*corresponding author
November 6, 2020
2
COVID-19 Forecasts, Projections, and Impact Assessments
The University of Arkansas for Medical Sciences (UAMS) Fay W. Boozman College of
Public Health (COPH) faculty conducted three types of assessments for this bi-weekly report: 1)
short-term forecasts of confirmed and probable cases, hospitalizations, and deaths, 2) long-term
projections of infections and hospitalizations; and 3) findings from the Arkansas Pandemic Poll.
All forecasts and projections were developed using COVID-19 data from the Arkansas
Department of Health through Nov. 1. All findings related to the Arkansas Pandemic Poll come
from data collected by the COPH from Oct. 3 through Oct. 17.
Summary points are:
15-day models continue to predict increasing numbers of daily cases, hospitalizations, and deaths due to COVID-19. The 15-day model forecasts 112,101 cumulative
confirmed COVID-19 cases in Arkansas by Nov. 16. Including confirmed and
probable cases, the 15-day model forecasts 121,627 cases by Nov. 16.
Fifteen-day models continue to show Arkansans between 35 and 59 will have the highest number of COVID-19 cases. Young adults 18 to 34 will have the second
highest number of cases. These two age groups will make up around 68% of the
COVID-19 caseload.
All counties in Arkansas reported new COVID-19 cases in the past two weeks. Two counties had two-week rates of change greater than 100%, and 12 counties had rates
of change greater than 50%.
The 15-day models are forecasting 7,893 cumulative hospitalizations and 2,627 cumulative intensive care patients by Nov. 16.
The trend for greatest number of hospitalizations continues to be in adults 60 to 74 years, who surpass the previously highest group of adults 35 to 59. Children younger
than 17 continue to have the fewest number of hospitalizations.
The mid-term model is forecasting hospitalizations by Dec. 30 will, if the forecast holds true, increase by 2,443 over hospitalizations on Nov. 1.
The 15-day model is forecasting 2,202 cumulative deaths by Nov. 16.
The long-term eSIR model suggests the pandemic will peak in March or April 2021 with between 20,000 and 63,000 active infections.
There are strong differences in willingness to accept a COVID-19 vaccine by race/ethnicity. Blacks have lower COVID-19 vaccine acceptance than Hispanics and
Whites. Differences by race/ethnicity appear to be centered on perceived vaccine
safety.
There are clear differences in willingness to accept a COVD-19 vaccine by acceptance of infection mitigation behaviors. Stronger beliefs about the necessity and
effectiveness of COVID-19 mitigation behaviors are strongly correlated with greater
willingness to accept a COVID-19 vaccine. This suggests addressing acceptance of
mitigation behaviors is likely to positively impact vaccine acceptance.
Despite differences by race/ethnicity and beliefs about mitigation behaviors, Arkansans in general appear to be tepid toward accepting a COVID-19 vaccine.
.
3
COVID-19 Cases and Infection
15-day forecast of confirmed COVID-19 cases in Arkansas. Figure 1 shows actual and
forecast COVID-19 cases in Arkansas. The model forecasts Arkansas will reach a cumulative
112,101 confirmed
COVID-19 cases by
Nov. 16, an increase of
7,738 confirmed cases.
As shown in Figure
1, confirmed COVID-19
cases continue to
increase, with little
change in the rate of the
growth curve since early
July. The model is
forecasting a steady
increase in cases over the
next 15-days.
The 15-day forecast
in the last report was
101,780 cumulative
confirmed cases by Nov.
1, around 2.4% less than the 104,239 confirmed cases reported on that date by the Arkansas
Health Department.
In addition to the 15-
day forecast of
confirmed cases, we
forecast future cases
using both confirmed
and probable cases. The
15-day forecast of
cumulative confirmed
and probable cases is
121,627 by Nov. 16, as
seen in Figure 2. As this
is the first time we are
providing this forecast,
we cannot assess its
accuracy. We will
include this assessment
in our next report.
Figure 1 Forecast confirmed COVID-19 cases through Nov. 16
Figure 2 Forecast confirmed plus probable COVID-19 cases through Nov. 16
121,627
4
Confirmed cases are those identified using the PCR test. The Department of Health
distinguishes between confirmed and probable cases. Probable cases are diagnosed using an
antigen test, which is considered less reliable than the more commonly used PCR test. The
Department of Health
provides the number of total
cases in Arkansas by adding
the number of confirmed
and probable tests together.
Because the Department of
Health was not including
positive antigen tests until
Sept. 2, we included only
positive antigen tests on or
after Sept. 2.
Forecast confirmed
cases by age and race. The
greatest growth in cases will
continue to be in adults 18
to 59, as shown in Figure 3.
The 15-day model is forecasting 39,846 confirmed cases in adults 35 to 59 by Nov. 16. The
second highest growth will be in young adults 18 to 34. The model is forecasting 35,960
cumulative confirmed cases in young adults 18 to 34 by Nov. 16. Together, these two age groups
will account for 68% of the COVID-19 caseload in the state. The least growth will be in adults
older than 75, with a projected 8,265 confirmed cases by Nov. 16.
As shown in Figure 4,
the 15-day model is
forecasting increasing
cases in all racial/ethnic
groups in Arkansas.
However, the model
forecasts the increase will
be greater in Whites.
Indeed, if we compare the
slopes for Blacks and
Hispanics to that of
Whites, the slopes are
increasing far more
modestly in Blacks and
Hispanics. The 15-day
model is forecasting
60,826 cumulative
confirmed COVID-19 cases among Whites by Nov. 16, an increase of approximately 10,000
cases over the number reported through Nov. 1st.
As shown in Table 1, we assessed the relative change in case rates by age and race/ethnicity
from Oct. 19 to Nov. 1. The greatest relative change was among adults older than 75 and those
60 to 74. The relative rate of increase was 16% and 13% respectfully. Those 17 or younger and
Figure 3 Forecast COVID-19 cases through Nov. 16 by age
Figure 4 Forecast COVID-19 cases through Nov. 16 by race
5
adults 35 to 59 both had relative rates of increase of 10% and 8%. Arkansans 18 to 34 had the
lowest rate of change in the past two weeks.
Table 1: Relative change in confirmed cases by race/ethnicity and age
Increase Relative change
Race/ethnicity
White 4,871 11%
Black 1,576 8%
Hispanic 658 4%
Age
< 17
18 to 34
35 to 59
60 to 74
> 75
1,162
2,515
3,272
1,470
981
10%
8%
9%
12%
16%
Map 1 Relative change in reported cases by County
6
The relative change in cases from Oct. 19 to Nov. 1 was greatest among Whites, 11%. Blacks
had a relative change of 8%, and Hispanics 4%.
Relative change in community COVID-19 cases. Map 1, on the previous page, shows the
relative change in each county’s case rate in the last two weeks. The relative change is
determined by calculating the percent change between case rates from the most recent two-week
period, Oct. 19 to Nov. 1, with the case rates from the prior two-week period, Oct. 5 to Oct 18.
Counties in red had the greatest relative change.
Fewer counties are showing large change rates in this report than in the previous one.
Counties with positive change rates greater than 50%, shown in red, continue to be concentrated
in rural counties in Arkansas. In the current report, we identified 12 counties with a positive rate
of change over 50%, compared to 17 counties in the previous report. Additionally, two counties
had change rates greater than 100%. Cleveland County had a change rate of 483% and Sevier
County 130%. Thirty-eight of the 75 counties in Arkansas had no change or negative change
rates.
It is important to note that the rate of change, if viewed without knowing underlying rates,
may not tell the full story of a county’s COVID-19 burden. A steady rate of 250 cases per 10,000
from one reporting period to the next would have a rate of change of 0%, even though the
disease rate is
high. There are
a number of
factors to
consider
beyond the
underlying
rate. Changes
in rates can be
affected by
recent events.
For example,
ADH recently
hosted two
free mass
testing events
in Cleveland
and Scott
counties,
which likely
contributed to
the high
relative
increases in
those counties.
While what can be said about a county’s COVID-19 burden is limited when assessing change
rates alone, change rates are useful when combined with other data. Map 2 shows the number of
COVID-19 cases per capita for each county. Per capita rates can be used along with data shown
on Map 1 to draw conclusions about how COVID-19 is spreading throughout the state. Per capita
Map 2 COVID-19 cases per 10,000 population, Oct. 19 to Nov. 1
7
rates are concerning when high. No county had a per capita rate in the last two weeks that
exceeded 100 per 10,000. The three counties with the highest per capita rates in the last two
weeks were Poinsett, 81, Craighead, 63, and Greene, 59.
When data are combined to form an overall picture, what we can conclude, for example, is
that Poinsett County has a high and slowly increasing caseload, compared to Cleveland County,
which has a more modest, but rapidly increasing caseload. If we look at data from the two maps
together, we can conclude that the pandemic has now firmly established itself in rural areas of
the state, primarily in the northeastern and western counties of the state, with comparatively high
case and relative change rates. We can also conclude the state would benefit from more
widespread testing to better describe the scope and magnitude of the pandemic.
COVID-19 positivity rates. Broadly defined, the COVID-19 positivity rate is the number of
people who test positive for COVID-19 as a proportion of the number of people who have been
tested. The positivity rate is an indicator of COVID-19 transmission in the state. A lower
positivity rate is indicative of less transmission, and a higher rate is indicative of greater COVID-
19 transmission. The positivity rate is dependent on the number of tests conducted. The positivity
rate has also taken on greater significance as part of CDC guidelines for local schools having in-
person classes. According to guidelines, the ideal positivity rate is less than 5%, but for practical
purposes less than 10% is acceptable.
Testing for
COVID-19 in
the state is on
par with the
national average
(2.7/1,000
versus
3.9/1,000).
Figure 5 shows
the seven-day
moving average
of the positivity
rates for
Arkansas and
the United
States.
Following the
second week in
May, the
positivity rate in
Arkansas increased and remained above the national average, except for two drops below the
national average for short periods in August and September. The October positivity rate declined
compared to rates in August and September. As of Oct. 31, the positivity rate in Arkansas was
11%, higher than the national rate of 6.8%.
Mid-term forecast of COVID-19 cases. The mid-term forecast provides a look at what
might happen between the end of November to the beginning of 2021. We used a SEIR model to
predict a seven-day rolling average.
Figure 5 COVID-19 positivity rate for Arkansas and the U.S.
8
As shown in
Figure 6, the
model forecasts
150,777
cumulative
cases on Dec.
31. The growth
rate will be
2.6% per week.
If the forecast
holds true,
Arkansas will
add 50,000 new
cases over the
number
reported to the
Department of Health on Oct. 31. If we include estimated active cases not reported, largely
because they are asymptomatic, we can expect an additional 20,000 active cases.
Long-term projection of active cases. As shown above in Figure 7, the eSIR model is
projecting the peak of the epidemic in Arkansas will be in late March or early April, with a mean
prediction of 35,718 active infections. The light-pink shaded region in Figure 7 shows the
uncertainty in the model (90% confidence interval), while the red line shows the mean estimate.
Summary. The short-term forecast describes significant continued growth in COVID-19
cases over the next 15 days. A plausible reason for this outcome is that portions of the
community do not see themselves at high risk of infection and are behaving accordingly. A
second plausible reason is pandemic fatigue. While not measured, it is described by many as
people and households simply tired of following CDC and state guidelines. The greatest number
Figure 7 Projected active COVID-19 infections
Figure 6 Projected COVID-19 cases through Dec. 31
9
of cases are in adults between the ages of 18 and 59. Adults younger than 60 may have
developed the impression that COVID-19 is not a significant risk for them or if infected they will
not develop serious disease.
For the long-term projections the timing and number of cases at the peak of the pandemic has
not changed substantially from the previous reports, although the lower bound of the confidence
interval has expanded downward. What this suggests is a greater amount of uncertainty in the
model.
10
COVID-19 Hospitalizations and ICU Admissions
Short-term forecasts of hospitalizations. Figure 8 shows the 15-day forecast for COVID-19
hospitalizations in the state on Nov. 16. The estimated trend in hospitalizations is consistent with
the increasing trend in confirmed cases. The 15-day model forecasts there will be 7,893
cumulative hospitalizations in Arkansas by Nov. 16, an increase of 801 or 11% in
hospitalizations over Nov. 1.
Figure 9 below shows a similar growth pattern for patients needing intensive care. The 15-
day model is forecasting 2,627 COVID-19 cumulative intensive care patients by Nov. 16, an
increase of 197 or 8% over Nov. 1.
In our last report, forecast
hospitalizations and patients needing
intensive care were very close to actual
numbers, within 1% and 2%
respectively. The model forecast 7,109
cumulative hospitalizations by Nov. 1
while the actual number was 7,092, a
difference of 17 hospitalizations.
Cumulative intensive care patients were
forecasted to be 2,401, less than 30
fewer than the actual number of
intensive care patients on Nov. 1 of
2,430.
Similar to the previous report, the
forecast of hospitalizations by age,
shown in Figure 10 on the next page,
presents a similar growth pattern
compared to the growth pattern for cases
shown in Figure 3, and emphasizes the
reasons why mitigation of COVID-19 is
important, especially for older adults.
The current report forecast the greatest
number of hospitalizations will be in
adults 60 to 74. Adults 60 to 74 are
forecast to have 2,545 cumulative
hospitalizations by Nov. 16, increasing
by 290 hospitalization. This compares
to 2,454 hospitalizations in adults 35 to
59, the second highest number.
The hospitalization rate of adults 60
to 74 diagnosed with COVID-19 is 18%,
almost three times higher than the
hospitalization rate of 6% among adults
Figure 8 Forecast hospitalizations
Figure 9. Forecast intensive care
11
35 to 59.
Hospitalizations in
adults 35 to 59 are
forecast to increase
by 138. The group
with the third highest
number of
hospitalizations are
adults over 75.
Almost one quarter of
adults over 75
diagnosed with
COVID-19 will be
hospitalized.
The groups with
the fewest
hospitalizations are young adults between 18 and 34 and children 17 or younger. Young adults
have a relatively low rate of growth in hospitalizations compared to older adults. Nonetheless,
the number of actual hospitalizations is not trivial. Young adults 18 to 34 are forecast to have
719 hospitalizations by Nov. 16, an increase of 42. The relative change in hospitalizations by
race/ethnicity and age are shown in Table 2.
Children younger than 17 are forecast to have approximately 163 cumulative hospitalizations
by Nov. 16. While the number of hospitalizations is small compared to other age groups, this
group continues to have growing hospitalization numbers. The number of actual hospitalizations
in children under 17 was 147 on Nov. 1. An increase of 16 hospitalizations represents an increase
of 11% in just two weeks.
Table 2: Relative change in hospitalizations by race/ethnicity and age
Increase Relative change
Race/ethnicity
White 283 8%
Black 65 4%
Hispanic 15 2%
Age
< 17*
18 to 24
35 to 59
60 to 74
> 75
*
30
94
143
135
*
5%
4%
7%
9%
*number of cases too small to report
Figure 10 Forecast COVID-19 hospitalizations by age
12
Our forecasts of hospitalizations by age groups for Nov. 1 were fairly accurate and close to
actual hospitalizations observed for all groups, with differences less than 11% between actual
and forecasted hospitalizations across all age groups.
We also forecasted
hospitalizations by race as
shown in Figure 11. All races
are forecast to show steady
increases in hospitalizations.
As expected, the majority of
hospitalizations were noted in
Whites. By Nov. 16, we
expect hospitalizations to rise
from 3,822 to 4,545 among
Whites, 1,919 to 2,331 among
Blacks, and 744 to 925 among
Hispanics.
Figure 11
Forecast COVID-19 hospitalizations by race
13
Hospitalizations by county. Evaluating the distribution of hospitalizations across the state
can help understand the impact COVID-19 may have on regional and state health system
resources. We created two graphics related to county-level hospitalization. For privacy reasons,
three counties with fewer than 10 hospitalizations were excluded from the analyses.
Map 3 provides the hospitalization rates per 100,000 residents. Make note that these are per
100,000, rather than per 10,000 like the maps related to positive cases. Sixty-seven of the 75
counties in Arkansas have hospitalization rates per 100,000 that are greater than 100. This means
that one out of every 1,000 people has been hospitalized for COVID-19 in nearly every Arkansas
county. The counties with the highest per capita hospitalization rates are Lee, 614.0, Chicot,
584.7, and Hempstead, 557.0. For these three counties, one of every 200 residents have been
hospitalized for COVID-19.
Map 3 Hospitalization Rates per 100,000 Residents by County of Residence
14
Similarly, understanding the percent of COVID-19 positive patients who have been
hospitalized can be an important measure of disease spread and an indicator of future
hospitalizations when combined with the number of new local cases. Map 4 provides the percent
of confirmed COVID-19 cases hospitalized. For example, a value of 5% means that 5 out of 100
of COVID-19 cases from that county were hospitalized. Fourteen counties have rates above 10,
which means that one of every 10 COVID-19 cases in those 14 counties were hospitalized. There
are nine counties with less than 5% hospitalization rates. The counties with the highest rates are
Cleburne, 12.4%, Lawrence, 12.3%, and Sharp, 12.3%. These are rural counties.
Mid-term hospitalization projections through Dec. 31. We introduce, for the first time,
mid-term projections for hospitalizations. Based on the SEIR prediction of the total cases, we
factored in the hospitalization rate, estimated over 15-day average to adjust for higher variability,
to predict the total number of cases requiring hospitalization at any given time. As shown in
Figure 12, the number of hospitalizations will continue to increase through the end of December.
Map 4 Percent of Positive Cases that were Hospitalized by County of Residence
15
By Dec. 31, hospitalizations are forecast to reach 9,537 cumulative hospitalizations, an increase
of 2,443 over Nov. 1. These results are based on limited data and as we receive more data we
will have greater confidence in our projection.
Long-Term Projections. Table 3, below, shows the long-term projections for
hospitalizations. By April 7, it is expected there will be 857 individuals hospitalized for COVID-
19 disease based on over 35,000 active infections. Of these hospitalizations, 299 will require
intensive care. We also consider a worst-case scenario. The worst-case scenario projects 1,426
hospitalizations based on almost 60,000 active infections on March 30. If the projected number
of hospitalizations holds true, the number of patients requiring intensive care would be 499.
Summary. The 15-day models are forecasting the greatest number of hospitalizations due to
COVID-19 will be in adults 60 to 74. COVID-19 disease is more severe in older people. The
rapid increase in the growth trend, with a 17% hospitalization rate in this group, highlights the
growing impact COVID-19 will have on the state’s hospitals. The forecast cases and
hospitalizations continue to be worrisome for older Arkansans and emphasizes the need for
continuing mitigation practices by all age groups. Almost a fourth of adults 60 and over who test
Table 3: Long-term projections of active infections, hospitalizations, intensive care, and
ventilations in Arkansas
Mean-Case Estimates Worst-Case Estimates
Peak Date April 7 March 30
Active Infections 35,718 59,421
Hospitalizations 857 1,426
Intensive Care 299 499
Ventilations 104 174
Figure 12 Projected hospitalizations by Dec. 31
16
positive for COVID-19 are hospitalized. Currently, growth in COVID-19 cases among these age
groups is relatively slow. But, with family holidays approaching and anticipated mixing of
family members and friends of all ages, infection rates in older adults could markedly increase. If
this were to happen, hospitalizations will also dramatically increase.
The second highest hospitalizations were noted in adults 35 to 59. The high growth in
COVID-19 cases and hospitalization among adults 18 to 65 is important. The vast majority of the
workforce is between the ages of 18 and 65. Even if not hospitalized with COVID-19 disease,
these adults will likely be out of the workforce for extended periods of time in isolation. Isolating
persons with COVID-19 will have a ripple effect, as persons in close contact with infected
persons are quarantined. As the pandemic in Arkansas continues to increase, isolating infected
persons and quarantining their contacts will result in significant numbers of people unable to
work.
17
COVID-19 Deaths
15-day forecast of COVID-19 deaths. The 15-day model is forecasting 2,202 deaths by
Nov. 16, as shown in Figure 13. The forecast is an increase of 357 or 16% over deaths reported
on Nov. 1. Our previous forecast of COVID-19 deaths was within 5% of actual numbers. The
model forecast 1,925 deaths by Nov. 1. The actual number was 1,845, a difference of 80.
Mid-term projections
of COVID-19 deaths. Mid-
term projections provide a
look at what might happen
between the end of
November and late-
December. As has been
stressed previously, the
farther out in time a model
projects, the less confidence
we have in model outcomes.
We use a SEIR model to
predict a seven-day rolling
average.
As shown in Figure 14,
the seven-day rolling
average forecast of
cumulative deaths in the
state on Dec. 31 is 2,887.
This is an increase of 1,042
deaths compared to actual
deaths on Nov. 1. The
weekly growth rate of deaths
through the end of the year
is forecast to be 2.4%.
To assess changes in the
number of deaths since the
last report, we measured the
relative change in the
number of cumulative deaths
by age and by race/ethnicity
from Oct. 19 to Nov. 1. The
growth rates in deaths align
with the growth rates in
hospitalizations, with the
exception of the younger 3
age categories. The younger
age categories show lower growth rates in deaths. This is consistent with the expectation that
COVID-19 will cause milder disease in children and young adults compared to older adults. The
relative change in deaths by race/ethnicity and age are shown in Table 4.
Figure 13 Projected COVID-19 deaths through Nov. 16
Figure 14 Projected COVID-19 deaths through Dec. 31
18
Summary. Consistent with increasing cases and hospitalizations over the last two weeks,
deaths rates continued to increase and are expected to do so for the foreseeable future. The recent
spikes in cases and hospitalizations may be reflected in an increasing death rate in the upcoming
forecasts. As the slope of the 15-day forecast suggests, the number of deaths is likely to increase
at a high rate in the next two weeks. If we compare the 15-day and mid-term forecasts,
November and December are likely to see 1,000 deaths, which is half again as many COVID-19
deaths in Arkansas as between March and October. Unfortunately, given current data, we cannot
take the Thanksgiving holiday into account. But, with the gathering of families over the holidays,
we are likely to see a sharp increase in cases, hospitalizations, and deaths similar to previous
holidays.
Table 4: Relative change in deaths by race/ethnicity and age
Increase Relative change
Race/ethnicity
White 96 9%
Black 65 4%
Hispanic* - 1%
Age
< 17*
18 to 24*
35 to 59*
60 to 74
> 75
-
-
-
33
79
-
5%
3%
7%
8%
*number of cases too small to report
19
Arkansas Pandemic Poll
The Fay W. Boozman College of Public Health at the University of Arkansas for Medical
Sciences (COPH) instituted a random digit dial (RDD) telephone poll to assess Arkansan’s views
on the COVID-19 pandemic. The pulse poll captures a random sample of adults in Arkansas
using random digit dialing. However, to ensure the results reflect the adult population of
Arkansas as a whole, we weighted survey results based on age and gender. To date, almost 9,000
Arkansans have completed the survey.
In this report, we examine vaccine acceptance among Arkansans. Vaccine acceptance is the
willingness to take a vaccine. We used a recently validated measure for vaccine acceptance
designed for the general population. We asked the respondent to focus on COVID-19 when
answering questions (see Methodological Notes). Data on vaccine acceptance was collected
between Oct. 3 and Oct. 17. The sample used for this assessment was 1,100.
Vaccine Acceptance in Arkansas
Figure 14 shows the mean levels of vaccine acceptance using a 10-item scale. In addition to
the overall measure, the scale has five subscales. The total scale score is the average score on all
10 items. Responses on a single item ranged from 1 to 7, with 4 signifying neither agreement nor
disagreement. The higher the
overall score, the more a person
is willing to accept a vaccine.
The overall vaccine acceptance
score of the 1,100 Arkansas who
participated in the poll is 5.0.
High vaccine acceptance would
be 6.0 or greater, so 5.0 is
slightly better than neither agree
nor disagree.
As shown in Figure 14,
‘safety’ refers to the subscale
measuring perceived vaccine
safety, and had a mean value of 4.7. ‘Effect’ refers to perceived vaccine effectiveness and the
belief that vaccines are necessary. This subscale had a mean value of 5. ‘Accept’ refers to
subscale measuring acceptance of vaccine selection and scheduling. This subscale had a mean of
5.1. “Value’ measures perceived value and effect of a vaccine. This subscale had a mean of 5.3.
Perceived legitimacy of authorities to require vaccinations is measured by the subscale ‘Legit.” It
had a sample mean of 4.9. Overall, no scale or subscale mean was significantly greater than 4.
This suggests Arkansans are very tepid when it comes to accepting a coronavirus vaccine.
3
3.5
4
4.5
5
5.5
6
Total Score Safety Effect Accept Value Legit
Figure 14 Mean Values for Vaccince Accpetance and the
Subscales in Arkansas
20
Variation in vaccine acceptance by
race/ethnicity. There were no differences in
vaccine acceptance by sex or age. However,
racial/ethnic differences were observed, as
shown in Figure 15. Acceptance was highest
among Hispanics (5.3) and lowest among
Blacks (4.5). The mean acceptance score for
Whites was 5.1. The differences across
race/ethnicity were relatively large and
statistically significant.
There were also differences across all
acceptance subscales by race/ethnicity, as
shown in Figure 16. Blacks score lowest on
all subscales, but especially on perceived
vaccine safety. Blacks reported their highest
scores on believing vaccines have positive value, 4.7 compared to 5.6 for Hispanics, and 5.2 for
Whites. On the subscale, selection of vaccines and their scheduling, the mean for Blacks is 4.8
compared to 5.6 for Hispanics and 5.2 for
Whites.
Overall, Blacks were least willing to
accept vaccines. This finding suggests the
state should make special efforts to
understand how Blacks perceive the
acceptability of vaccines differently than
Whites and Hispanics, and develop
communication and distribution programs
accordingly. However, the results also
argue for a primary need to move the
whole Arkansas population higher in
terms of vaccine acceptance.
Vaccine acceptance by pandemic
characteristics. We considered vaccine acceptance by beliefs about other pandemic mitigation
behaviors: wearing a mask regularly in public, agree wearing a mask helps stop the spread of
COVID-19, perceived chance of getting COVID, agree with the decision to allow large groups to
gather, and past two weeks attended church, temple, or other religious gathering. Other questions
were explored but the pattern is the same.
Chances of getting COVID-19. We asked respondents what they thought the chances are of
getting the coronavirus. There were differences across all vaccine acceptance subscales. As
shown in Figure 17 on the next page, respondents who said there was no chance they would get
infected had a mean vaccine acceptance score of 4.6 compared to a mean of 5.1 for those who
felt there was some chance they would get infected. There were similar differences across
vaccine effectiveness (5.1 vs. 4.6) and legitimacy of government to require vaccines (5.0 vs. 4.5).
Vaccine safety had the lowest values for both any chance (4.9) and no chance (4.2) of getting
3
3.5
4
4.5
5
5.5
6
Safety Effect Accept Value Legit
Figure 16 Vaccine Acceptance Subscales by
Race/Ethnicity
White Black Hispanic
3
3.5
4
4.5
5
5.5
6
White Black Hispanic
Figure 15 Overall Vaccine Acceptance Scale by
Race/Ethnicity
21
COVID-19. The acceptance of scheduling of vaccines (5.2 vs. 4.8) and vaccines have a positive
value (5.4 vs. 4.9) had the highest mean scores.
Mask Wearing.
We asked three
questions about
mask wearing. First,
we asked if a
respondent had
regularly worn a
mask in the past
two weeks. Second,
we asked if a
respondent thought
the mask helped
protect him/her
from the getting the
coronavirus. Third,
we asked if a
respondent believed
a state order requiring citizens to wear a face mask in public is needed.
Figure 18 shows the results for persons who said they regularly wear masks in the past two
weeks. For this group, the overall vaccine acceptance score was 5.1 compared to 4.1 for those
who did not
regularly wear a
mask. Differences
across the
subscales are most
striking for vaccine
safety (4.7 vs. 3.7)
and legitimacy of
the government to
require vaccines
(5.0 vs. 3.6).
Differences
between regular
mask wearers and
non-regular mask
wearers in vaccine
acceptance was less stark across the subscales of vaccines have a positive value (5.4 vs. 4.6),
vaccines are effective (5.1 vs. 4.2), and acceptance of selection and scheduling of vaccines (5.1
vs. 4.3).
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 18 Vaccine Accpetance by Regularly Wears Mask
Yes No
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 17 Vaccine Acceptance by Chance of Getting Coronavirus
No chance Any chance
22
Next, we examined if
respondents believe masks
helped stop the spread of
the coronavirus. Overall, if
the respondent believes
wearing masks helps stop
the spread of the virus, the
total vaccine acceptance
measure was higher, 5.2,
than if the respondent does
not believe masks help
stop viral transmission,
4.5. Large differences were
found for perceived safety
(4.9 vs. 4.0) and legitimacy
of government to require
vaccines (5.2 vs. 4.3). Positive value of the vaccine had the highest values (5.4 vs. 5.0) with the
smallest difference. Both effectiveness of the vaccine (5.2 vs.4.5) and acceptance of selection
and scheduling the vaccine (5.2 vs. 4.7) showed large mean differences.
Finally, we assessed if respondents believed the state order to wear a mask was needed was
related to vaccine acceptance. As Figure 20 shows, the difference in mean scores was not large.
Those who felt a state order was needed had a mean score of 5.1 compared to those who felt it
was not needed, 4.8.
The mean difference
across the vaccine
acceptance subscale
was largest for
legitimacy of
government to
require vaccines (5.1
vs. 4.6) and smallest
for the belief that
vaccines have a
positive value (5.4 vs.
5.2). Perceived safety
of the vaccine was
low for those who felt
the order was
warranted (4.8) and those who felt it was not warranted (4.4). For vaccine effectiveness (5.1 vs.
4.8) and acceptance of the scheduling of vaccines (5.2 vs. 4.9) the results were similar to the
overall vaccine acceptance scale scores.
Social Gathering. Several questions in the Pandemic Poll ask about social gatherings. One
asks about whether a respondent agrees or disagrees with the decision to allow large social
gatherings. A second question asks the respondent if he/she attended church, temple or other
religious event in person in the past two weeks.
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 19 Vaccine Acceptance by Masks Helps Stop the Spread of
the Coronavirus
Yes No
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 20 Vaccince Acceptance by Statewide Mask order Needed
Yes No
23
As shown in Figure 21, the mean difference on the vaccine acceptance scale and subscales
between those who believe large gatherings should be allowed and those who do not are small
(5.1 vs. 4.8). Furthermore, there was no real mean difference between the groups on the subscale
measuring the value
of vaccines.
Generally, if a
person agreed with
the decision to
allow large group
gatherings, they
were less likely to
accept vaccines
than those who
disagree with the
decision to allow
large gatherings.
As shown in
Figure 22, there
were small
differences in the
overall measure of
vaccine acceptance
by the measure of
church attendance,
4.8 vs. 5.1. And,
differences were
small across all the
sub-scales. Overall,
greater church
attendance was
associated with
lower vaccine
acceptance, but the
difference was not
large.
Summary. As Arkansas works towards developing a COVID-19 vaccination program, it is
important to understand variation in vaccine acceptance across the state. We used a valid and
reliable tool designed to assess vaccine acceptance in the general population. The tool was
designed to assess multiple components of vaccine acceptance. Essentially, the measure assumes
a person’s decision to accept a vaccine is the product of an evaluation of vaccine safety,
effectiveness, scheduling, beliefs about the positive value the vaccine has for self and society,
and legitimacy of the government to require vaccines. We found important differences for
vaccine acceptance and its subscale components by race/ethnicity and beliefs about pandemic
mitigation efforts. Asking a simple question about taking a COVID-19 vaccine misses important
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 21 Vaccince Acceptance and Agree with Decision to Allow Large
Social Gatherings
Agree Disagree
3
3.5
4
4.5
5
5.5
6
Total Safety Effect Accept Value Legit
Figure 22 Vaccine Accpetance by Whether Attended Church, Temple of
Other Religious Gathering in person in Past Two Weeks
Yes No
24
opportunities to target different approaches to different groups to improve vaccine acceptance.
What is clear is: Blacks have lower vaccine acceptance than Hispanics and Whites and
differences may center on perceived vaccine safety. It is also clear that the more a person agrees
with and practices COVID-19 mitigation practices, the more likely he/she is to accept a vaccine.
25
Methodological Notes
Short-term forecasts. Time series forecasting is a method that uses observed data to predict
future values. The purpose of the models is to fit the best curve to data and extend the curve into
the future. To forecast aspects of the pandemic in Arkansans, the models used COVID-19 cases,
hospitalizations, ICU admissions, and death data reported to the Arkansas Department of Health.
It should be noted the report defines a “case” as a COVID-19 test result reported and posted by
the Department of Health. As indicated by recent research, the number of undiagnosed COVID-
19 infections in the community may be higher by 40 to 50%. We cannot provide a precise
number of undiagnosed infections in the community, as an antibody seroprevalence study has not
yet been completed in the state.
Mid-term Projections. The SEIR model projects COVID-19 cases and deaths using the
same basic parameters — susceptible (S), exposed (E), infected (I), and recovered (R), that have
been widely used to model epidemics since the 1920s. In addition, SEIR models account for the
changing social conditions, such as the face mask order and opening schools, changing infection
probabilities, and symptomatic and asymptomatic spread of cases. To arrive at the best model fit
for mid-term projections of COVID-19, we first used a SEIR model (Exposed (E)) to model
existing cases. The resulting fit was very good, but required a second step to project cases out to
predicted date. The difficulty with SEIR-like models is that actual COVID-19 cases may not
accurately represent viral spread. This can occur for a number of reasons, including variation in
rates of testing and limited knowledge of the contribution of asymptomatic infections to viral
spread. To extend our SEIR model projections, we calculated a seven-day rolling average model
using the number of cases to date. Results between the SEIR and seven-day rolling average
estimates were consistent, with a fit coefficient above 75%.
Long-term projections. The eSIR model is based on the extended state-space SIR (eSIR)
model. A standard SIR model has three components: susceptible (S), infected (I), and removed
(R), including both recoveries and deaths. The proportion of the population falling into each
mutually exclusive category is assumed to vary over time, creating the standard epidemic curve.
The model creates projections of active infections, including mild and asymptomatic infections,
over time. Active infections are not cumulative infections from the beginning of the pandemic,
nor are they restricted to new cases on a given day. Rather, the model estimates the proportion of
the population with an unresolved infection at a given point in time.
Changing model assumptions and their impact on projections. Since the last report, the
model’s assumption regarding the likelihood of transmission has been adjusted slightly upward
to better match Arkansas data. The model was also extended an additional six months into the
future to better observe the predicted post-peak dynamics. The eSIR model was originally
developed using assumptions based on data from China, such as the R0 estimate. R0 (pronounced
R-naught) is a measure of how many people one infected person can infect. The model learns
and improves over time by adjusting internal assumptions as more Arkansas-specific data
become available. For example, the R0 changed in the model over time from 3.15 to 1.39. Earlier
versions of the model, working with less Arkansas data, relied more heavily on the assumptions
derived from Chinese studies. Consequently, in the beginning, the model predicted a more
aggressive epidemic than we have observed in Arkansas. As more Arkansas data have become
available, the model has adjusted itself to better reflect the more extended epidemic curve we
now observe.
26
Comparison to other models. Curve fitting models, like the widely cited University of
Washington IHME model, tend to make strong assumptions, which are unlikely to hold as more
data become available. In addition, curve fitting models cannot account for epidemic dynamics.
This often results in severe reductions in predictive strength beyond short-term windows.
SIR/eSIR models, like we use in this report, have a stronger theoretical basis for long-term
projections. Regarding the eSIR model’s relatively late date for a peak, this is in line with other
long-term projection models, such as the CIDRAP Viewpoint, which predicts the COVID-19
pandemic will last 18 to 24 months. Furthermore, reports from week to week cannot be
compared to each other. As more data are added to a model, differences reflect new Arkansas-
specific data. Therefore, the results reported above should not be compared to the previous
reports. However, the eSIR model may be suggesting the COVID-19 growth curve may be
leveling off.
Arkansas Pandemic Poll. One the challenges of looking at responses by race/ethnicity is
that some racial/ethnic groups in the poll have too few numbers to be included in the analyses.
For example, we do not have enough respondents who are Marshallese to include them in the
analyses. However, the Marshallese are an important racial/ethnic group in the state with respect
to the COVID-19 pandemic.
The percentage of individuals who felt that they have a low chance of getting infected in the
last two weeks. Response categories vary from 1 (no chance) to 5 (high chance). Responses 1, 2,
and 3 were recoded as low chance and 4 and 5 are recoded as high chance of contracting
COVID-19.
Vaccine Acceptance Measure
Original source: A survey instrument for measuring vaccine acceptance. Sarathchandra, D,
Navin. C, Largent, M, McCright, A. Preventive Medicine 109:1-7, 2020.
Instructions for the interviewer to read to the respondent. The following questions ask your
views on vaccines particularly potential COVID-19 vaccines. Please indicate how much you
agree or disagree with the following. Interviewer, read response categories following each
question. (Note the underlined text in the instructions is the only change from the original
instrument.) The response categories are: strongly disagree, moderately disagree, slightly
disagree, I am not sure, slightly agree, moderately agree, strongly agree. The questions are :
vaccines are safe; vaccines contain dangerous ingredients; some vaccines are unnecessary since
they target relatively harmless diseases; vaccines are effective in preventing diseases; we give
children the right number of vaccines; we give children too many vaccines, vaccines conflict
with my beliefs that children should use natural products and avoid toxins; vaccines are a major
advance for humanity; the government should not force children to get vaccinated to attend
school; to protect public health, we should follow government guidelines about vaccines.
The overall vaccine acceptance scale is the sum of 10 items. Subscales are the following
sums: 1&2 perceived safety of vaccines; 3&4 perceived effectiveness and necessity of vaccines;
5&6 acceptance of vaccine selection and scheduling; 7&8 perceives the value of vaccines; 9&10
perceived legitimacy of authorities to require vaccinations.
The Cronbach’s alpha for overall scale is 0.83.
27
Glossary of Terms
Active infection = a positive infection, with or without a COVID-19 test, that has not yet
recovered or died
Case = a positive COVID-19 test result reported to the Arkansas Department of Health
Community = population not in a prison or population not in a prison or nursing home
Cumulative = total number of a given outcome (e.g., cases) up to date
Extended state-space SIR (eSIR) model = a model based on three components: susceptible
(S), infected (I), and removed (R, including both recoveries and deaths)
Susceptible-Exposed-Infected-Recovered model (SEIR) = another variant of standard
epidemiological model considering exposure as another factor controlling for disease dynamics
Hospitalization = a positive infection or case that was admitted to the hospital
ICU = intensive care unit admission
Infection = a COVID-19 infection, with or without a test and regardless of having recovered
or died
Non-incarcerated (NI) = representative of an individual who is not in a jail or in a
correctional facility
Positivity Rate = The number of people who test positive for covid-19 as a proportion of
people have been tested
Projections = long-term predictions
Recovered = a positive infection that is no longer symptomatic or shedding virus
Susceptible = an individual who can be infected with the disease of interest
Time series forecast = short-term forecast of events through a sequence of time