Professional Documents
Culture Documents
Modelling The Effect of Changes To The COVID 19 IsolationPolicy For Release
Modelling The Effect of Changes To The COVID 19 IsolationPolicy For Release
Provided to NZ Ministry of Health: 22 March 2023; This version updated 27 Mar 2023 following
internal peer review.
Executive Summary
● We used a mathematical model to investigate the potential impact of ending
mandatory COVID-19 case isolation requirements on the number of infections,
hospital admissions, and deaths over the next 6 months.
● In the model, ending mandatory case isolation caused a wave of infections in the
subsequent 1-2 months, although after 4-6 months infections settled to a level
that was only slightly higher than if mandatory isolation was maintained.
● Under our best estimate, ending mandatory case isolation led to a 13-25%
increase in the total number of COVID-19 admissions and deaths in the
subsequent 6 months.
● However, there is a lot of uncertainty about the exact impact of ending
mandatory isolation, because firstly the current behaviour of people with
COVID-19 is not well known and secondly the change in behaviour in response to
the policy change is difficult to predict. Not everyone follows the current
requirements, while others will continue to follow isolation guidelines even if they
are no longer mandatory.
● To account for some of this uncertainty, we ran alternative model scenarios. In a
scenario with a smaller transmission increase, which could occur if many people
continue to test and isolate voluntarily, the increase in hospital admissions and
deaths could be as small as 6%. In a scenario with a larger transmission increase
where more people leave their dwelling while infectious, the increase in
admissions and deaths could be over 35%.
● The model results only relate to total national numbers. It is likely that groups at
higher risk of severe COVID-19, including Māori and Pacific people, and shielding
individuals will be disproportionately affected by an increase in transmission.
● The model results suggest that, beyond the 4-6 months following a policy change,
long-term outcomes are relatively insensitive to the timing of any decision to end
mandatory isolation.
● Irrespective of legal requirements, staying at home when sick is an important
public health message for reducing disease transmission.
1
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Background
We have previously modelled the dynamics of SARS-CoV-2 in New Zealand using a
compartment-based ordinary differential equation (ODE) model (Lustig et al., 2023;
Vattiato et al., 2023). This age-structured model includes waning of vaccine-derived and
infection-derived immunity, immune evasion of new Omicron sub-variants,
age-dependent hospitalisation and death rates, and changes in transmission resulting
from behavioural and policy changes. The model is calibrated to data on COVID-19
cases, hospitalisations and deaths using an approximate Bayesian computation (ABC)
method.
Anyone testing positive for COVID-19 in New Zealand is currently legally required to
self-isolate until 7 days after their positive test result or onset of COVID-19 symptoms,
whichever is earlier (Parliamentary Counsel Office, 2022). Previous work from COVID-19
Modelling Aotearoa has estimated the effect of different isolation periods on
transmission risk (Harvey et al. 2022a). This report presents results from the ODE model
(Lustig et al., 2023) to support the 16 March 2023 Public Health Risk Assessment (PHRA)
to assess whether the current public health risk of COVID-19 justifies retaining mandatory
isolation. The Ministry of Health is providing information to the PHRA related to risk of
reinfections, long covid, and key surveillance trends showing the current impact/risk
profile of COVID-19.
Model scenarios
We ran a baseline version of the model fitted to data up to 25 February 2023. We then
compared this baseline model against three scenarios in which there is an instantaneous
increase in the effective reproduction number on 21 March 2023 of 5%, 10% and 15%. This
can equivalently be thought of as a 5%, 10% or 15% increase in the average number of
contacts during an individual’s infectious period. The 10% increase in transmission is our
central estimate of the impact of ending mandatory COVID-19 isolation requirements
and shifting to guidance only, with the 5% and 15% scenarios capturing some of the
uncertainty about the effect size. We present results showing the change in the number
of SARS-CoV-2 infections, COVID-19 hospitalisations and COVID-19 deaths in each of
these scenarios. We do not explicitly show results for cases because the effect of a
policy change on testing and reporting rates is unknown.
2
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
The network contagion model results estimated that the complete removal of case
isolation behaviours could lead to a 15% increase in transmission, relative to a baseline
with no household quarantine and only symptomatic testing but high compliance with
case isolation (Harvey et al., 2022b). There is significant uncertainty in this estimate, in
particular due to the difficulty of estimating the proportion of people currently testing
and following case isolation requirements, the number of people who would follow
“guidance to isolate”, and other unmodelled changes in behaviour that might
accompany such a policy change. If case isolation behaviour has already decreased, or
if people follow the isolation guidance after mandatory isolation has ended, the
modelled increase would be less than 15%. However, model results following the end of
the Covid-19 Protection Framework in September 2022 showed that subsequent
epidemiological data were consistent with a larger than expected increase in
transmission (Vattiato et al., 2023). This was potentially the result of unmodelled
behaviour changes, which could also occur following an end to mandatory Covid-19
isolation.
The impact of shifting to guidance will also depend on what the recommended isolation
guidance is. For example, voluntary test-to-release for when to end isolation would
reduce transmission, but advice to only isolate while symptomatic would increase risk. In
previous work, COVID-19 Modelling Aotearoa (Harvey et al. 2022a) calculated that the
impact of the current 7 day policy of mandatory isolation of confirmed COVID-19 cases
results in an average of 9 hours infectious in the community after release1 and 83 hours
of excess isolation2 per confirmed case. The same report calculated that a
“test-to-release” policy with a 7 day maximum, but with possible early release after day 5
upon returning 1 (respectively 2) negative results on an antigen test would result in only
a small increase in hours infectious in the community after release (12 (resp. 10) hours)
but a large reduction in hours of excess isolation (51 (resp. 57) hours).
Factors that may shift outcomes towards the lower transmission increase scenario
include strong, clear public health messages on the importance of following the
1
An estimated 15% of individuals are still infectious at the end of their mandatory 7 day isolation
period. The average hours infectious in the community is the mean number of hours that
individuals remain infectious for, after the end of mandatory isolation, averaged across the whole
population of confirmed cases including those cases whose infectivity resolves before the end of
mandatory isolation and who contribute zero hours to the mean.
2
The estimated 85% of individuals whose infectious period ends before the end of their
mandatory isolation period will be required to spend some time in isolation in excess of the end of
their infectious period. Average hours of excess isolation is the mean number of hours that
individuals spend in isolation when no longer infectious, averaged across the whole population,
including those cases whose infectivity resolves after the end of their isolation period and who
contribute zero hours to the mean.
3
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
recommended isolation guidelines and what the guidelines are, availability of and access
to rapid antigen tests, and providing financial support for people to isolate.
Results
Table 1 shows the change in infections, hospitalisations and deaths under each scenario,
relative to the baseline model in which there is no policy or behavioural change. The
Table shows results for a 7 week period and for a 26 week period following the policy
change. Figure 1 shows the relative change in cumulative infections, hospitalisations and
deaths over time compared to the baseline model. See Supplementary Material for
additional results broken down by age group.
4
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Table 1. Model results for the short-term and long-term impact of ending mandatory COVID-19 isolation requirements. Differences
in cumulative infections, COVID-19 hospital admissions, and COVID-19 deaths, in the 7 weeks and 26 weeks following the policy change,
and peak hospital occupancy during the 26 weeks following the policy change, under three model scenarios (+5%, +10% and +15% change
in transmission on 21 March 202). All results are relative to the baseline model with no policy change. In each table cell, the first line shows
change in absolute numbers and the second line shows relative (percentage) change compared to baseline. Values in brackets represent
the 95% confidence intervals on these differences.
Lower +83,000 [+32,000, +95,000] +500 [+200, +600] +23 [+12, +40] +81,000 [+59,000, +88,000] +700 [+400, +800] +73 [+63, +135] +103 [+30, +130]
(+5% on +27% [+25%, +29%] +25% [+21%, +26%] +15% [+12%, +16%] +6% [+5%, +9%] +7% [+6%, +11%] +8% [+7%, +13%] +24% [+12%, +26%]
21Mar23)
Central +179,000 [+73,000, +200,000] +1,000 [+400, +1,300] +51 [+27, +88] +164,000 [+117,000, +176,000] +1,400 [+700, +1,500] +148 [+124, +269] +233 [+95, +287]
(+10% on +57% [+52%, +65%] +55% [+48%, +57%] +34% [+26%, +36%] +12% [+11%, +17%] +15% [+13%, +21%] +17% [+15%, +25%] +54% [+37%, +57%]
21Mar23)
Higher +282,000 [+123,000, +308,000] +1,700 [+700, +2,100] +85 [+45, +145] +247,000 [+174,000, +264,000] +2,100 [+1,100, +2,300] +225 [+185, +406] +382 [+179, +463]
(+15% on +91% [+80%, +108%] +90% [+82%, +96%] +56% [+43%, +59%] +18% [+16%, +24%] +23% [+20%, +31%] +25% [+23%, +37%] +88% [+70%, +92%]
21Mar23)
5
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure 1. Model results for the relative (percentage) increase in cumulative infections (left), cumulative COVID-19 hospital admissions
(middle) and cumulative COVID-19 deaths (right), over one year following an increase in transmission of 5% (blue), 10% (green), 15%
(purple) compared to the baseline model (0% change in red). Results are aggregated across all ages. Solid curves show the difference in
best-fit model trajectory with and without the transmission increase, shaded bands contain the difference for paired trajectories (same
set of parameters) with and without the transmission increase for 95% of accepted model realisations.
6
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
7
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Acknowledgements
The authors acknowledge the role of the New Zealand Ministry of Health, StatsNZ, and
the Institute of Environmental Science and Research in supplying data in support of this
work. The authors acknowledge the role of the COVID-19 Modelling Government Steering
Group in designing the modelling questions to support the Public Health Risk
Assessment. The authors are grateful to Nigel French, Amanda Kvalsvig, Markus
Luczak-Roesch, Melissa McLeod, L. Martis, Ben Ritchie, Matt Parry and Patricia Priest for
feedback on an earlier version of this report. This work was funded by the New Zealand
Department of Prime Minister and Cabinet and Ministry of Health.
8
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Author contributions
● SD: prepared document, including model output manipulation and visualisations;
contributed to ODE model.
● JG: contributed to the design and building of the interaction network that the
Network Contagion Model runs on.
● EH: design, building, and running the Network Contagion Model which feeds
parameters into the ODE model, contributed to ODE model design, assisted in
writing the document.
● DO: led design and building of the Network Contagion Model which feeds
parameters into the ODE model, contributed to ODE model design, assisted in
writing the document.
● FPE: involved in running the Network Contagion Model which feeds parameters
into the ODE model, and interpreting results.
● EPF: involved in running the Network Contagion Model which feeds parameters
into the ODE model, and interpreting results, assisted in writing the document.
● MP: prepared document, involved in building the ODE model.
● OM: contributed to the design of the contagion code for the Network Contagion
Model, involved in building the ODE model, developed/implemented the
fitting/credible band construction methods, assisted in writing the document.
● ST: contributed to the design and building of the interaction network that the
Network Contagion Model runs on.
● GV: prepared document, involved in building the ODE model, ran model
simulations presented.
● DW: developed the contagion code codebase for the Network Contagion Model.
References
1. Harvey, E.P., Looker, J., O’Neale, D.R.J., Plank, M.J., Priest Forsyth, E., Trent, J.
(2022a). Quantifying the impact of isolation period and the use of rapid antigen
tests for confirmed COVID-19 cases. COVID-19 Modelling Aotearoa pre-print,
available at:
https://www.covid19modelling.ac.nz/quantifying-the-impact-of-isolation-period/
2. Harvey, E.P., O’Neale, D.R.J., Patten-Elliot, F., and Priest Forsyth, E. (2022b).
Estimating the effect of changes in case isolation on the effective reproduction
number of COVID-19 in Aotearoa. Advice sent to the COVID-19 Modelling
Government Steering Group 15/09/2022.
3. Horizon Research Limited (2022). Behavioural surveillance 2 survey, November
2022 Report. Prepared for: Evaluation and Behavioural Science, Intelligence,
9
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
10
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Supplementary Material
These results suggest that there are reasonably high levels of compliance with current
case isolation requirements. This also shows that although household contact quarantine
requirements were replaced with guidance in September, in November there were a
reasonable number of household contacts following the guidance. This is evidence that
there is still a significant proportion of the public who are currently taking actions
including symptomatic testing, case isolation and household contact quarantine to
reduce transmission. And therefore there is potential for an increase in transmission if
case isolation requirements are removed, and if behaviours changed around testing and
staying home if sick, or if a household contact.
The ODE model is based on numerous parameter assumptions: some of these values
are fixed, others are fitted using a naive ABC (approximate Bayesian computation)
method (see Lustig et al., 2023 for details). Using this method, the model picks random
combinations of parameter values from their 'prior' distributions (results presented here
3
In the survey, of the people who indicated that they would have difficulty self-isolating, 18% said
it was because they “don’t have space to self-isolate away from others at home”. This indicates a
potential misinterpretation of “self-isolation”, and suggests that some of those who hadn’t (or
wouldn’t) self-isolate, may still stay home, they just may not be able to stay away from their
household contacts.
11
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
are from 15,000 independent draws from the prior distribution), then outputs a sample
from an approximate posterior distribution of accepted values. The accepted values are
those resulting in the 1% best fitting trajectories (i.e. those that give the smallest value for
a distance function that measures the difference between a simulated trajectory and
observed data). The distance function is based on the number of new daily infections in
a routinely tested cohort of border workers (up to July 2022), total and age-stratified
daily reported cases of COVID-19, total and age-stratified daily hospital admissions for
COVID-19, and total daily COVID-19 deaths (excluding deaths that are classified as “Not
COVID-related”. The age-stratified daily cases and hospital admissions are included for
each age group as a proportion of the total. The accepted model runs are then used to
plot a 'best fit' line and a 95% confidence interval.
It is worth noting that multiple different combinations of parameter choices can result in
simulations that give an equally good fit to empirical data, including combinations of
parameters that may conflict with other parameter combinations. It is therefore
important to consider modelling results as an ensemble of plausible trajectories, subject
to the constraints and assumptions of the model.
Supplementary results
The results shown here are for scenarios with a specified increase in transmission on 21
March 2023, representing the potential impact of a reduction in the number of infected
individuals effectively isolating during their infectious period, as described in the Main
Text. The model was run with an increase in transmission of: 0% (henceforth referred to
as the 'baseline' model), 5%, 10% and 15%.
Figure S1 shows model results in the four scenarios for the number of daily infections,
reported cases, new daily Covid-19 hospital admissions, hospital occupancy, and daily
Covid-19 deaths. Note that for the results on cases, the model assumes that the case
ascertainment rate (i.e. proportion of infections that are reported as cases) is not
affected by any policy change. In reality, it is likely that ending mandatory
reporting/isolation will lead to a drop in the case ascertainment rate, as is thought to
have occurred in Australia for example. This means that the model may significantly
overestimate the number of reported cases following a policy change, but we note that
this does not affect the number of infections, hospitalisations or deaths in the model.
Figures S2-S7 show the increase in infections, hospital admissions and deaths in the
scenarios with an increase in transmission, compared to the baselines model. In each
graph, solid curves show the difference in the cumulative number of outcomes over time
in the best-fit model between the baseline (no change) model and the scenario with the
specified increase in transmission (5%, 10% or 15%). Shaded bands represent a 95%
confidence interval (under model assumptions) for the difference in the cumulative
number of outcomes as a result of the transmission increase.
12
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Graphs are shown both for absolute increase and for relative increase compared to
baseline. We do not plot cases here as the effect of any policy change on testing and
reporting behaviour is unknown and not included in the model (see also Main Text). Each
set of graphs shows results split into 10-year age groups and for the aggregated totals
(bottom-right plot in yellow box).
13
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S1. Model results for new daily infections (per 100,000 people), cumulative
infections since 1 January 2023, daily reported cases, new daily Covid-19 hospital
admissions, hospital occupancy and daily Covid-19 deaths for the four model scenarios
(red = baseline, blue = 5% increase, green = 10% increase, purple = 15% increase in
transmission on 21 March 2023). Solid curves show the best-fit model trajectory, shaded
bands contain 95% of accepted model realisations, red points show observed data. Note
the model assumes that there is no major new variant of SARS-CoV-2, no change in the
intrinsic virulence of the virus, no seasonality or other changes affecting transmission
rates over the relevant time period. This means that the baseline model should not be
used as a long-term prediction for the absolute numbers of infections, admissions or
deaths, but rather a scenario of what would be likely to occur under these specific
modelling assumptions.
14
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S2. Model results for the change in cumulative infections following an increase in
transmission of 5% (blue), 10% (green), 15% (purple) compared to the baseline model.
Results are shown split into 10-year age bands, and aggregated across all ages (bottom
right graph). Solid curves show the difference in best-fit model trajectory with and
without the transmission increase, shaded bands contain the difference in trajectories
with and without the transmission increase for 95% of accepted model realisations.
15
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S3. Model results for the percentage increase in cumulative infections following an
increase in transmission of 5% (blue), 10% (green), 15% (purple) compared to the baseline
model. Results are shown split into 10-year age bands, and aggregated across all ages
(bottom right graph). Solid curves show the difference in best-fit model trajectory with
and without the transmission increase, shaded bands contain the difference in
trajectories with and without the transmission increase for 95% of accepted model
realisations.
16
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S4. Model results for the change in cumulative Covid-19 hospital admissions
following an increase in transmission of 5% (blue), 10% (green), 15% (purple) compared
to the baseline model. Results are shown split into 10-year age bands, and aggregated
across all ages (bottom right graph). Solid curves show the difference in best-fit model
trajectory with and without the transmission increase, shaded bands contain the
difference in trajectories with and without the transmission increase for 95% of accepted
model realisations.
17
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S5. Model results for the percentage increase in cumulative Covid-19 hospital
admissions following an increase in transmission of 5% (blue), 10% (green), 15% (purple)
compared to the baseline model. Results are shown split into 10-year age bands, and
aggregated across all ages (bottom right graph). Solid curves show the difference in
best-fit model trajectory with and without the transmission increase, shaded bands
contain the difference in trajectories with and without the transmission increase for 95%
of accepted model realisations.
18
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S6. Model results for the change in cumulative Covid-19 deaths following an
increase in transmission of 5% (blue), 10% (green), 15% (purple) compared to the baseline
model. Results are shown split into 10-year age bands, and aggregated across all ages
(bottom right graph). Solid curves show the difference in best-fit model trajectory with
and without the transmission increase, shaded bands contain the difference in
trajectories with and without the transmission increase for 95% of accepted model
realisations.
19
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Figure S7. Model results for the percentage increase in cumulative Covid-19 deaths
following an increase in transmission of 5% (blue), 10% (green), 15% (purple) compared
to the baseline model. Results are shown split into 10-year age bands, and aggregated
across all ages (bottom right graph). Solid curves show the difference in best-fit model
trajectory with and without the transmission increase, shaded bands contain the
difference in trajectories with and without the transmission increase for 95% of accepted
model realisations.
20
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
Infections
0% 312,000 129,000 - - - - -
379,000
0% 1,900 800 - - - - -
2,500
Covid-19 deaths
0% 151 96 - 257 - - - -
Infections
21
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
0% 1,384,000 750,000 - - - - -
1,515,000
0% 9,100 5,100 - - - - -
11,200
Covid-19 deaths
10% 1,039 760 - 1,797 148 124 - 269 16.6 14.8 - 24.9
15% 1,116 822 - 1,932 225 185 - 406 25.2 22.5 - 36.6
22
This report is a pre-print. It has been subject to internal peer review.
Embargoed until 3pm, 11th April 2023, NZ time.
15% 817 413 - 1015 382 179 - 463 87.7 69.8 - 92.1
23