BRIEF REPORT: COVID-19 EPIDEMIC TRENDS
AND PROJECTIONS IN OREGON
Results as of 6/10/2020, 6pm

ACKNOWLEDGEMENTS
This is a brief update to the Institute for Disease Modeling’s (IDM’s) previous reports. IDM
developed the Covasim software, provided Oregon Health Authority (OHA) with programming
scripts for the models, and provided extensive support and technical assistance to OHA. OHA
especially wishes to thank Cliff Kerr, Katherine Rosenfeld, Brittany Hagedorn, Dina Mistry,
Daniel Klein, Assaf Oron, Prashanth Selvaraj, Jen Schripsema, and Roy Burstein at IDM for
their support (Contact: covid@idmod.org).

RESULTS SUBJECT TO CHANGE
Please note that the data reported here are continually being updated. For daily up-to-date
information visit the OHA COVID-19 web page. The results in this brief are subject to change
as more data become available, the science to inform the model assumptions expands, and
modeling methods continue to be refined. While these results can be used to understand the
potential effects of different scenarios, it is important to note that the 80% forecast intervals for
these predictions are wide, so point estimates should be interpreted with caution.

1

 KEY FINDINGS
Success of Oregon’s interventions
•

The aggressive community interventions in Oregon were effective in dramatically reducing
COVID-19 transmission.

Changes after Oregon has begun to reopen
•
•

Recent data and model calibration provide evidence that transmission has increased since
reopening began on May 15th.
It is too early to accurately estimate the effects of phased reopening on changes in COVID19 trends. Because of this, we made three different assumptions about the recent COVID19 trends for our projections:
•

Most optimistic assumption: We assumed a 10 percentage point increase in
transmission after re-opening on May 15th. This model fit the hospitalization trend in late
May well and assumed that trend would continue. It assumed the more pronounced
increase in hospitalizations in the last few days of data (June 2nd-4th) was only a
temporary increase.
o Under this scenario, the number of new infections per day will remain relatively
stable over the next month.

•

Less optimistic assumption: We assumed a slightly larger increase in transmission (15
percentage points) after May 15th, which fit the recent observed hospitalization and
diagnoses trends better.
o Under this scenario, the number of new infections per day will gradually increase
over the next month.

•

More pessimistic assumption: We assumed a 15 percentage point increase in
transmission after May 15th, as in the “less optimistic” scenario, plus an additional 10
percentage point increase in transmission after May 25th. By doing so, this model
assumed that the recent increase in the number of new diagnoses is indicative of
increased transmission, rather than largely due to increased testing.
o Under this scenario, the number of new infections per day will increase more
dramatically. Compared to the most optimistic scenario, this model projects about
14,000 more cumulative infections (35,400 vs. 21,400), 925 more new infections
per day (1,040 vs. 115), and 17 more new severe cases per day (22 vs. 5) by
July 3rd.

2

 Conclusions
The results suggest that the epidemic has slowed in Oregon since its beginning, but that
transmission appears to be increasing since reopening. In our most optimistic scenario, the
effective reproduction number is estimated to be currently at about 1. The other scenarios
demonstrate that even small increases in transmission levels could lead to increased cases.

PURPOSE OF THIS REPORT
To describe epidemic trends in COVID-19 since Oregon began to re-open, and project trends
over the next month assuming different scenarios.

METHODS
This brief report presents analyses done using methods consistent with the previous May 29,
2020 report, with some key updates:
-

-

-

Newer data from Orpheus on COVID-19 cases (Orpheus description) were used. The
Orpheus data file was obtained on June 8th, but data after June 5th were considered
incomplete because of lags in reporting and not used.
Parameter assumptions were modified to vary susceptibility by age, such that the age
distribution of infected individuals in the model follows that of recently diagnosed cases
in Oregon.
We included a comparison of our results to a few other published models.
We moved the model simulation start date and number of starting infections forward (to
45 infections on February 15th) in order to reduce forecast interval uncertainty.

More information about the methods is in Appendix 1.

INTERVENTIONS
Oregon implemented numerous measures to slow the transmission of COVID-19, including:
•
•
•

•

On March 8, 2020: Governor Brown declared an emergency due to the public health
threat.
On March 12, 2020: A large number of measures were put in place, such as bans on
gatherings of more than 25 people, as detailed here.
On March 16, 2020: Schools were closed statewide, as detailed here. Further measures
were put in place on March 16th and 17th, including the closure of restaurants and bars
and gatherings of more than 25 people, as detailed here.
On March 19, 2020: Non-urgent health care procedures were suspended to conserve
personal protective equipment and hospital beds.
3

 •
•

•

On March 23, 2020: Aggressive interventions, namely the “Stay Home, Save Lives”
recommendations, were put in place.
On April 22, 2020: Testing guidelines were revised to allow for expanded testing,
including testing of people who are asymptomatic and work in care settings or are in
congregate settings; they were refined on May 1, 2020 (Revised testing guidelines).
Since the beginning of the epidemic in Oregon: Public health staff have routinely
investigated diagnosed cases, asked those cases to identify their close contacts, and
then notified those contacts of their exposure (i.e., contact tracing). Because of limited
public health resources in Oregon, public health staff had only been able to actively
follow up with contacts in households and congregate settings. Contacts have been
asked to voluntarily stay in quarantine for 14 days after their last known exposure. Any
diagnosed cases have been asked to voluntarily stay isolated for at least 72 hours after
their symptoms resolve (i.e., quarantine). Contact tracing efforts have recently started to
expand, as mentioned below (see also May 12 weekly report).

REOPENING
On May 1, 2020, Oregon announced plans for phased relaxation of community mitigation
strategies, with additional expansion of testing and contact tracing to keep transmission low
(Reopening Plans May 1, 2020). Some key changes have included:
•
•
•
•

•

On May 1, 2020: Certain elective and non-urgent medical procedures resumed (Medical
Procedures May 1, 2020).
On May 2, 2020: The widespread use of face coverings was encouraged (Face
Coverings May 2, 2020).
On May 5, 2020: Some parks, outdoor recreation facilities, and areas across Oregon
were opened for day use (Parks May 5, 2020).
On May 7, 2020: Governor Brown published detailed guidance on reopening. This
included requirements for counties to reopen, such as having sufficient capacity for
testing and contact tracing. The guidance also called for the widespread public use of
face coverings, maintaining physical distance of six feet between individuals as much as
possible, and following good hygiene and disinfection practices (Reopening Guidance
May 7, 2020).
On May 15, 2020: Some counties began to reopen, and certain restrictions were eased
statewide, such as allowing social gatherings of under 10 people and cultural/civic/faith
gatherings of up to 25 people with physical distancing, as detailed here and here).
Briefly:
o On May 15th, 31 of the 36 counties in Oregon had been approved for Phase 1 of
reopening.
o By June 1st, 35 counties were approved for Phase 1 reopening. The most
populous county (Multnomah) has applied for Phase 1 reopening on June 12th.

4

 o On June 5th and 6th, 28 counties were approved for Phase 2 reopening, as well
as one more on June 8th.

RESULTS
As with previous modeling reports, the results in this brief report are subject to change as more
data become available, the science to inform the model assumptions expands, and modeling
methods continue to be refined (see Appendix 2 for information on the limitations). The models
simulate the spread of COVID-19 in Oregon statewide under different scenarios. They do not
take into account the complex disease spread or intervention effectiveness within and between
specific populations over time, such as for communities of color, workers in certain
occupations, or people in congregate settings. They are using average transmission levels;
hence they do not, for example, model the recent outbreaks in work settings differently than
other types of transmission.
Epidemic trends to date
The model was calibrated by modifying the assumptions from the literature to best fit data from
Orpheus on confirmed positive COVID-19 diagnoses, number of tests completed,
hospitalizations (referred to as severe cases below), and deaths for Oregon. The dates on
which model transmission levels change were selected based on key policy enactment dates,
with the exception of 3/31/20 (which was based on data observation). The degree of changes
in transmission were informed by hospitalization and diagnoses data (i.e., not by the assumed
effect of any policy). The model was run 11 times in calibration.
The calibration provides evidence that Oregon’s interventions -- combined with increased
hygiene and other measures that appear to have begun earlier -- dramatically reduced the
burden of COVID-19 in Oregon (Figure 1).
•

•

The data are consistent with a stepped reduction in transmission in Oregon, beginning
with a 5% decrease in transmission by March 8th, up to a brief 80% decrease in
transmission after March 23rd. Indeed, while the interventions before March 23rd
appeared to have slowed epidemic growth, the additional aggressive measures
implemented on March 23rd (i.e., “Stay Home, Save Lives”) appear to have curtailed that
growth. The reductions were likely due to people spending more time at home, as well
as an increase in hygiene and disinfection practices, wearing of facial coverings, and
physical distancing outside the home, but we do not have the data to determine the
relative contribution of each change.
The data suggest that these dramatic reductions in transmission waned somewhat after
March 31st, but the number of new daily infections was still declining through mid-May.

The calibration also provides evidence that transmission has increased since reopening began
on May 15th. Based on daily hospitalizations flattening out after a lengthy decline, transmission
appears to have increased by at least 10 percentage points since May 15th, but it is too early to
5

 determine the full extent of these changes. Hospitalizations typically follow new infections by
about 12 days. Therefore, the effects of 31 counties reopening on May 15th would be expected
to start appearing after May 27th. In addition, the three most populated counties (Clackamas,
Washington, Multnomah) opened later (May 23rd, June 1st, and anticipated on June 12th), so
we would not expect to see most of the effect of those counties reopening on hospitalizations
yet. A recent increase in diagnosed cases (June 1st-5th) suggests a potential further increase in
transmission, but was due at least in-part to recent workplace outbreaks and associated
testing (OHA Weekly COVID-19 Report). It is unclear if this increased level of transmission will
be sustained. Given this uncertainty, we calibrated the model three different ways:
•

•

•

The most optimistic calibration assumed that the hospitalization trend in late May would
continue, but that the more pronounced increase in the last few days of data (June 2nd4th) would not continue. That is, we assumed that it was a temporary increase due to
random variability. This corresponds to a ten percentage point increase in transmission
after May 15th (i.e., from a 70% baseline reduction baseline to a 60% reduction baseline
reduction).
A less optimistic calibration assumed a slightly larger increase in transmission (15
percentage points) after May 15th, which fit the recent observed hospitalization and
diagnoses trends better.
A more pessimistic calibration used the recent data on new diagnoses to further modify
transmission estimates. Specifically, we assumed that the recent increase in new
diagnoses (June 1st-5th) was indicative of increased transmission, rather than being due
to more widespread testing, increased contact tracing, and/or active monitoring of close
contacts of cases (OHA Weekly COVID-19 Report) 1. Following a 15 percentage point
increase in transmission after May 15th, we assumed an additional 10 percentage point
increase after May 25th.

In Figure 1, we present the most optimistic calibration. Appendix 3 contains a comparison of
the three calibrations.
Of note, it is too early to assess the effects of the recent marches for Black Lives Matter, which
began in Oregon on May 28th and greatly expanded around May 31st.2 The COVID-19 data
used for this report includes only 5 days since May 31st.

1

Because testing guidelines were updated April 22nd to expand testing and presumptive cases were recently
added, the number of new diagnoses in early May increased temporarily and then decreased. The recent
increases appeared after that decrease.
2 https://en.wikipedia.org/wiki/George_Floyd_protests_in_Portland,_Oregon#May_28

6

 Figure 1: Model calibration with Oregon case data. Dotted single vertical lines correspond, from left to
right, to simulation start date (February 15th); estimated reductions in transmission relative to baseline
of 5% (March 8th), 10% (March 12th), 50% (March 16th), 80% (March 23rd), 70% (March 31st), and an
early-estimate change to 60% after reopening (May 15th). The double vertical line at April 27th-28th
indicates a change in the portion of tests allocated to symptomatic people. Raw data are presented as
squares; estimates from the calibration are presented as lines. Note: The estimated reductions in
transmission are imprecise, especially given some are based on few data points. The shaded areas
represent variability among the calibration runs.

7

 Scenario projections
We modeled three future scenarios through July 3rd based on the three different calibration
assumptions described above. For all scenarios, we assumed 2,000 tests per day to
conservatively3 reflect current testing levels (OHA COVID-19 Testing). We ran the forecast
model 11 times to simulate the epidemic and produce forecast intervals.
Most optimistic scenario: We assumed a 10 percentage point increase in transmission after
May 15th.
•

If this trend continues over the next month, the number of new infections per day is
projected to remain relatively stable (Figure 2). The effective reproduction number
(Re) – the expected number of secondary cases that a single case generates – is
estimated to remain at about 1 (Figure 4).

Less optimistic scenario: We assumed a 15 percentage point increase in transmission after
May 15th.
o If this trend continues over the next month, the number of new infections per day is
projected to slowly increase (Figure 2). Specifically, the model projects about 3,600
more cumulative infections (25,000 vs. 21,400), 170 more new infections per day
(270 vs. 100), and 4 more new severe cases per day (9 vs. 5) by July 3rd than the
most optimistic scenario. 1 The Re is estimated to increase to about 1.2 (Figure 4).

More pessimistic scenario: We assumed a 15 percentage point increase in transmission after
May 15th and an additional 10 percentage point increase in transmission after May 25th.
o If this trend continues over the next month, the number of new infections per day is
projected to increase more dramatically (Figure 3).4 Specifically, the model projects
about 14,000 more cumulative infections (35,400 vs. 21,400), 900 more new
infections per day (1,000 vs. 100), and 17 more new severe cases per day (22 vs. 5)
by July 3rd than the most optimistic scenario. 1 The Re is estimated to increase to
about 1.6 (Figure 4).

3

Higher total levels of testing are anticipated, but increases are partially driven by additional virus surveillance
and study-related testing. Since the nature of how tests relate to diagnoses, and how diagnoses related to
transmission may be different for additional, non-clinical tests, we conservatively assumed 2,000 tests / day in the
model.
4 Figure 3 projections for the “more optimistic” and “less optimistic” scenarios are identical to those in Figure 3;
these scenarios were first presented separately to show differences at the smaller y-axis scale.

8

 Summary
While these results can be used to understand the potential trends in COVID-19 under
different scenarios, it is important to note that the 80% forecast intervals for these predictions
are wide, reflecting their uncertainty.5
Nevertheless, modeling results suggest that transmission, after over a month of decline,
appears to be increasing since reopening: the Re is estimated to be currently at about 1 in our
most optimistic scenario. The other scenarios demonstrate that even small increases in
transmission -- whether due to changes in policies, workplace practices, or public adherence to
physical distancing, hygiene, and facial covering recommendations -- could lead to increased
cases. Indeed, other models have suggested that physical distancing measures and/or other
precautions (e.g., wearing facial masks, hand washing) will likely need to continue to some
extent for COVID-19 even with extensive testing and contact tracing (e.g., Giordano et al.,
2020; Kretzschmar et al., 2020; Rosenfeld et al., 2020).

Comparison with other model results
Results from the most optimistic calibration in this report are consistent6 with the results from
Imperial College London and with RT Live , which estimate the Re for Oregon is currently very
close to 1. Columbia University’s projections for Oregon are more similar to this report’s
projections for the less optimistic scenario, with increasing number of new infections.

Trends in mobility measures
For context, we examined trends in physical distancing measures based on smartphone mobility
data (Appendix 4). From these data, physical distancing appears have reached a peak in the
first week of April and has slowly decreased since then. That decrease has continued since reopening began May 15th. However, these measures are based on mobility data, and do not
measure personal practices related to hygiene, wearing of facial coverings, maintenance of sixfoot physical distance from others outside the home, or workplace practices.

“the forecast intervals used correspond to the 10th and 90th percentiles of the simulated trajectories. Although these forecast
intervals bear some similarities to confidence or credible intervals, since they are typically produced through a combination of
stochastic variability and parameter uncertainty, they do not have a rigorous statistical interpretation.” (p 18 of IDM report)
6 Imperial College, RT Live, and Columbia University reports / projections accessed 6/10/2020.
5

9

 Figure 2: Model projections for the next 4 weeks, assuming that after May 15th: 1) transmission
increased by 10 percentage points (blue line), and 2) transmission increased by 15 percentage points
(red line). The lighter shaded areas correspond to 80% forecast intervals (i.e., 10th and 90th percentiles
of the projection).

10

 Figure 3: Model projections for the next 4 weeks, assuming that after May 15th: 1) transmission
increased by 10 percentage points (blue line), 2) transmission increased by 15 percentage points (red
line), and 3) transmission increased by 15 percentage points, then an additional 10 percentage points
on May 25th (green line). The lighter shaded areas correspond to 80% forecast intervals (i.e., 10th and
90th percentiles of the projection).

11

 Figure 4: Projected effective reproduction number (Re) through June 30th, assuming that starting May
15th: 1) transmission increased by 10 percentage points (blue line), 2) transmission increased by 15
percentage points (red line), and 3) transmission increased by 15 percentage points, then an additional
10 percentage points on May 25th (green line). The lighter shaded areas correspond to 80% forecast
intervals (i.e., 10th and 90th percentiles of the projection). Re is the expected number of secondary
cases that a single case generates.

12

 APPENDICES
Appendix 1: Detailed transmission model methods
We applied Covasim version 1.0.2, an individual-based (i.e., “agent-based”) COVID
transmission model with parameters informed by the literature; the full source code is available
on GitHub. The methods and assumptions for Covasim are described in detail here.
The model was calibrated by modifying the assumptions to best fit data from Orpheus on
confirmed positive COVID-19 diagnoses, number of tests completed, hospitalizations (referred
to as severe cases below), and deaths for Oregon.
Our model assumed random network connections, had scenario noise set at zero, and used
default parameters from Covasim version 1.0.2, except for the following changes:
1) Population age distribution was based on American Community Survey 2018 singleyear estimates for Oregon. We used a simulation population size to 420,000 with
Covasim’s population rescaling functionality enabled.
2) The COVID-19 virus had a pre-intervention Beta value of 0.021, instead of 0.016 (based
on observed hospitalizations before interventions took effect).7
3) Disease parameters were updated to match recent CDC best estimates for pandemic
planning scenarios (CDC Planning Scenarios) for age-specific hospitalization
probabilities. Specifically, we adjusted Covasim’s more granular age-specific severe
probabilities (given infection) such that for Oregon’s population they equated to 2.6% for
ages 0-49, 6.9% for ages 50-64, and 11.4% for ages 65 and older.
4) Parameter assumptions were modified to vary susceptibility by age, such that the age
distribution of infected individuals in the model follows that of cases diagnosed between
April 16th and May 15th in Oregon.
5) The relative probability of symptomatic individuals being tested was adjusted to match
actual diagnoses counts given our inputted number of tests, with a change in relative
odds on April 28th-onward.
To match observed epidemic trends, we started the model with 45 infected individuals on Feb.
15th, 2020; this date was moved forward and the number of infections increased from prior
reports to produce narrower forecast intervals. It is not possible to calibrate the model with a
single importation event near the first diagnosis (Feb. 21, 2020), which was a community
acquired infection.

7

With an average of 20 contacts per individual per day and a mean duration of infectiousness of 8 days, this perday probability roughly translates to an R0 of 3.

13

 Appendix 2: Limitations
The results in this brief report are subject to change as more data become available, the
science to inform the model assumptions expands, and modeling methods continue to be
refined. There are limitations important to note:
-

-

-

-

-

-

-

-

The projections included in this report are based on the best available local data and
evidence as of June 8th, 2020, but the local collection of epidemiologic data on COVID19 cases may lag in ways we did not account for, and data improvement efforts are
ongoing.
After the initial imported cases, the model assumes that no additional cases were
imported from elsewhere over time. Any such cases would inflate local transmission
levels, though any actual resulting diagnoses, hospitalizations, and deaths from
imported cases are included in the data the model is calibrated to.
For simplicity, we assumed random network connections and a combined effect of
various interventions for the future scenarios (e.g., physical distancing, expanded
testing and contact tracing) on overall transmission, but Covasim does have the ability
to incorporate more complex network dynamics and specific intervention effects (as
described here). We will explore those and other modeling options in the future.
The model produces diagnoses largely among the symptomatic population, whereas in
actuality an increasing number of high risk but asymptomatic individuals are being
diagnosed through contact tracing and testing in workplaces and congregate settings.
Estimated reductions in transmission over time are imprecise and not necessarily due to
any particular policy, especially given some are based on few data points.
We assumed that individuals who were diagnosed subsequently reduced their
transmission by 80%, but this reduction may vary as social norms change.
Although our model was calibrated to track actual testing and diagnoses counts, it did
not explicitly account for reduced transmission from individuals who are not tested but
undergo quarantine due to contact tracing efforts.
Given the fairly low number of cases in Oregon, trends in cases and their age
distribution (and therefore prognosis) are sensitive to a single outbreak or super
spreader event, such as the recent Pacific Seafood outbreak with over 100 cases.
These models simulated the spread of COVID-19 in Oregon statewide under different
scenarios. They did not take into account the complex disease spread or intervention
effectiveness within and between specific populations over time, such as for
communities of color, workers in certain occupations, or people in congregate settings.

Last, there remain significant unknowns, including information about public compliance with
recommendations (e.g., hygiene, face coverings, physical distancing) and the disease
dynamics. As CDC stated (CDC Planning Scenarios) “new data on COVID-19 is available
daily; information about its biological and epidemiological characteristics remain limited, and
uncertainty remains around nearly all parameter values.”

14

 Appendix 3: Comparison of model calibrations

Number of daily new diagnoses

Figure A-1: Daily diagnoses by model calibration. Raw data are presented as squares; estimates from
the calibration are presented as lines.

15

 Number of daily new severe cases

Figure A-2: Daily severe cases (hospitalizations) by model calibration. Raw data are presented as
squares; estimates from the calibration are presented as lines. Note: hospitalization results for the 3 rd
calibration (“15% after 5/15, 25% after 5/25”) are identical to that of “15% increase after 5/15” during the
calibration timeline (the May 25th change would not affect hospitalizations until approximately June 6th).

16

 Appendix 4: Mobility Data
We examined data from the University of Maryland COVID-19 Impact Analysis Platform which
has compiled and displayed a large number of indicators related to social distancing.
Specifically, the platform has developed a Social Distancing Indexi that combines smartphone
data for the following metrics: percent staying home, reduction in all trips, work trips, non-work
trips, travel distance, and out-of-county trips. We chose this data source over previously reported
data from Safegraph because of the additional metrics included.
Social distancing sharply increased in mid-March, reaching a high of 55 during the first week of
April. It has slowly decreased since then. When some counties began to reopen on May 15 th,
the trend continued the slow decline. Specifically, the index was 40 from May 10 th to May 16th,
37 from May 17th to 23rd, and 34 from May 24th to May 30th, and 29 from May 31 to June 6.
These data reflect human mobility and as such, provide helpful information on the extent of
transmission opportunities. However they are imprecise, as they do not directly represent
physical proximity to potentially infected persons. Most of these metrics will also relax by
necessity as the state reopens. Therefore, data on personal practices, such as maintenance of
physical distance from others, wearing masks, and hand washing practices could be quite
valuable as reopening continues. In a national survey, for example, most respondents (60%)
said they are maintaining a distance of at least 6 feet from others outside their homes and half
(50%) reported wearing a mask all of the time.ii
Social Distancing Index, University of Maryland COVID-19 Impact Analysis Platform,
Oregon, January – May 2020

i

The social distancing index is computed from six mobility metrics by this equation: social distancing index = 0.8*[% staying
home + 0.01*(100 - %staying home)*(0.1*% reduction of all trips compared to pre-COVID-19 benchmark + 0.2*% reduction
of work trips + 0.4*% reduction of non-work trips + 0.3*% reduction of travel distance)] + 0.2*% reduction of out-of-county
trips. The weights are chosen based on share of residents and visitor trips (e.g., about 20% of all trips are out-of-county trips,
which led to the selection of a weight of 0.8 for resident trips and 0.2 for out-of-county trips); what trips are considered more
essential (e.g., work trips more essential than non-work trips); and the principle that higher social distancing index scores
should correspond to fewer chances for close-distance human interactions and virus transmissions
ii
Ipsos conducted this survey using the KnowledgePanel, a representative address-based panel of U.S. adults age 18 and
over. Interviews were conducted in English. Sample size was 1,009 and margin of error was +/- 3.2%. Data are from May 12
survey

17