Arizona: Second Wave Infection Analysis
[08/02/20 Update: This analysis has been refreshed. See our recent post: Arizona: Summer Analysis Update]
[07/31/20 Update: Our analysis for Arizona will be refreshed with our improved algorithm and latest data. Our estimate of increased activity in June was low and as a result we underestimated the current spike. Case 4 cumulative deaths below are likely consistent, but the timing is off.]
Current trends in Arizona show a significant increase in infections. There was a bump in hospital admissions for COVID-19 in early June followed by a bump in deaths occurring 10 days later. Much of the media is reacting with hysteria concerning this second wave. We take close look at the data and evaluate what the real risks are going forward using our COVID Decision Model, which is well suited for modeling dynamic heterogeneous effects.
We reference the Arizona Department of Public Health COVID-19 Dashboard for statistics. We present the model results first for four scenarios and then review the Arizona Department of Public Health data. The details of each case study are presented at the bottom of this post.
Our analysis shows that the best path forward is to loosen up by age demographic where risks are low, protect the vulnerable, understand and monitor trends and adjust behavior recommendations appropriately. This approach provides the best balance of economic recovery and public health balance, a data driven response based a clear understand of infection dynamics so clear and common sense policy adjustments can be implemented. In this analysis we see the current infection spike decaying rapidly if people continue common sense social distancing norms (hand washing and the use of masks indoors in high density situations).
The Arizona Second Wave Model
There are two trends that seem to be somewhat independent: increased infections peaking in late June and a peak in hospital admissions for COVID-19 in early June and deaths in mid-June.
First, we have an increase in infections that are likely resulting from an overall increase in connectivity proportional to age (younger people dropping social distancing norms). The loosening of stay at home orders in mid May and widespread civil protests after Memorial Day both contributed to this increased connectivity. This increased social contact for younger demographics has caused a large spike in infections. Death rates resulting from these infections will be very low.
Second, there was a peak in hospitalizations on June 5th and deaths on June 15th. This spike in deaths in June is more likely due to penetration of the virus into vulnerable Native American populations and surge in hospitalizations and resultant deaths for sick individuals in the Mexican border regions seeking medical care in the US.
Let's take a look at the model results for Arizona calibrated to current trends. We have modeled a heterogeneous shift in behavior by age group, proportional to actual individual risk. We assess four scenarios to understand the nature of this second infection wave in Arizona and a range of outcomes over time.
- Case 1 assumes a more optimistic shift in behavior by age to pre-COVID-19 levels. Younger people are more active, older people are less so. Vulnerable people are protected as well folks over 70 years.
- Case 2 assumes a more conservative shift in behavior. Vulnerable people and folks over 70 years old have additional protection. This closely matches current trends.
- Case 3 starts with Case 2 assumptions, but then increases lower age connectivity while protecting the vulnerable population, similar to Case 1. This shows a third wave similar to the second wave and the total fatalities are similar to Case 2.
- Case 4 starts with Case 2 assumptions, but goes to full connectivity for all age brackets while continuing to protect the vulnerable.
The cumulative deaths for each case are shown below. Actual death data from Arizona department of health (date of death) is plotted for reference. Note that due to the random nature of the Monte Carlo simulation, there are ebbs and flows in the model of the virus spread, much like real life.
The reproduction factor over time is still very low as a starting baseline, with R0 in the range of 1.5 if we ignore simulation noise spikes. Arizona is different from high connectivity demographic areas like NYC where the R0 was closer to 4 during the unmitigated initial spread. You can see the dynamics of the loosening in Case 3 and Case 4. In all cases, the virus runs it's course without a vaccine if age appropriate social distancing norms are kept in place. Again, Case 2 is the most likely scenario for a managed recovery.
The percent of the population infected (active infections + dead + recovered) is presented below for all four cases. In May the number crosses the 3% range, consistent with antibody testing results. Currently we estimate that about 10% of the population has been infected. The head of the CDC recently reported that we are only tracking about 10% of the total cases [Market Watch Article]. Currently there are about 70k cases reported in Arizona. If the CDC is right, we would expect about 700k actual cases or about 10% of the Arizona population. Our model is consistent with that assessment.
Active Infections are shown below. Note that the second wave hump for Case 1 exceeds the ratios of positive tests that we see in the Arizona state data. Case 2 more accurately shows the ~2X ratio between April and June infection positive rates. Active Infectious (those who are infectious and not quarantined) are a subset of those who are actively infected.
Hospitalization admissions and critical care occupancy are shown below. Note that Case 1 and Case 4 put the highest load but are the lease likely scenarios. Case 2 and Case 3 are peak in mid June consistent with current Arizona data.
Arizona Infection Trends
Arizona is increasing molecular PCR (Polymerase Chain Reaction) tests which verify the presence of an active viral infections. Tests results are showing increased percentage positives. The current percentage of positives for testing for June is 16%, in April it averaged 10%. The trend for June is upward showing a peak close to 20% for the 3rd week in June. We can deduce the relative growth in infections if we assume a uniform increase in the sample window. Since the percentage positive has increased, the overall infection rate now is higher than in May by a factor of 2x to 2.5x. Serology testing shows a slow and steady increase in overall percentage positive as would be expected as the infection spreads over time.
The demographics of COVID-19 victims shows a heavy infection rate in the 20-45 age bracket.
Death and Hospitalization Rates
The death rate is showing a dip in mid May and second hump around June 15th. New hospitalizations were relatively flat in April, dipped in May and peaked again around June 5th and are now trailing off. Peak deaths will typically trail about 10 days from peak hospital admissions and this is what the data shows. Deaths should continue to decline and track the hospital admission rate by 10 days. Deaths are aligned with the senior age brackets. The 65+ age group accounts for 12% of the infections and 74% of the deaths. The 20-45 age group accounts for 49% of the infections and only 5% of the deaths.
Why Declining Hospitalizations but Increasing Cases Positive Percentage?
We know that the mobility data being tracked by the IHME dashboard for Arizona shows an increase in mobility but we also know that people are more aware of the need to protect vulnerable people. Common sense would indicate that as more people are out and about, more infections will occur. These infections will be predominately in the age groups where the risk is lowest. Those at risk and with risk conditions would tend to continue to isolate and protect themselves. This would explain the disconnect between the growing infect rate and the declining hospitalization rate.
IHME social distance tracking data for Arizona based on cell phone mobility shows increased connectivity over time. This trend does show an increase in activity from early April through June.
The CDC now has a tracking dashboard that includes mobility metrics. Keep in mind that these metrics don't include social distancing physical contact reduction, hand washing and mask usage. The trends are similar.
Early on the proportional correlation between death rates and actual infections was relatively constant yielding a fairly consistent Infection Fatality Rate (IFR). Now that we have figured out that elderly people and those with existing co-morbidity conditions are vulnerable, the behavior has shifted, and the IFR is dropping. The general population is now interacting at higher rates and the virus is working through these populations with death rates no greater than those of typical flu seasons.
Detail Arizona Model Results by Case