COVID-19 forecast models are changing, and for some pretty good reasons. First, we are learning much more about the disease – for example, how quickly it spreads in different populations and the high proportion of asymptomatic cases. Second, we are also learning the disease was spreading in the U.S. and elsewhere much earlier than the start of official case counts and that testing by-in-large is still not sufficiently broad in most countries or U.S. states to establish “real” population estimates of incidence or prevalence.
Initial forecasts In the U.S. were generally based on observed growth rates of symptomatic disease; as a result, these early estimates tended to have a much faster observed doubling rate than what was occurring in the population. Modelers (ourselves included) took a combination of rapidly expanding testing and a very quickly spreading disease and misinterpreted this as being an incredibly quickly spreading disease. This produced generally pessimistic projections of the virus’ spread.
Other related important developments have included:
- The portion of potentially asymptomatic cases may be very high. This affects how we think about risk of hospitalization and other outcomes of exposed and infected persons.
- The maximum attack rate may be around 73% (assuming prisons represent optimal conditions for pandemic growth rates. )
What does this all mean?
The first COVID-19 forecast models tended to predict a nearly apocalyptically bad outbreak – outputs of epidemic growth rate models are extremely sensitive to assumptions and incomplete and confusing early data may have led to preliminary estimates that were off by orders of magnitude. As we started to learn more, model outputs reduced to merely catastrophic. As models have begun to incorporate social distancing as limiting factors and incorporated other behaviors the forecasted infection rates and hospital demand in many models was dramatically reduced – the question is whether modeling is tending towards optimism, potentially over-reacting to poor initial predictions. As social distancing takes effect, an important policy concern is that our models may be confusing an impermanent but potentially temporarily effective public health intervention (lockdowns and social distancing) for a permanent recession of the disease.
Our big caution at this stage is that testing is still key, and that high-intensity contact tracing and traditional disease control efforts need to be stepped up in most states to prevent resurgence. The states and populations that are experimenting in relaxing social distancing early on do so at risk to us all, particularly as poor testing may allow leakage of the disease into dense and still-vulnerable population areas elsewhere. While costly, many important lessons will be gained from the “natural experiment” the U.S. is about to undertake, assuming that we are able to sufficiently measure the outcomes of these experiments.
Alex Hartzman wishes to acknowledge the guidance and review from Al Dobson, Ph.D. and Joan DaVanzo, Ph.D. Dobson | DaVanzo is currently engaged with the Connecticut Hospital Association to provide COVID modeling.