A mix of policy and behavior helped Colorado dodge a bullet late last year, contributing to a cumulative 2020 COVID-19 death count far below some of the worst-case scenarios predicted by the state's modeling team in the fall, one of its members says.
But were other factors at play? Was the modeling off? Did Colorado buckle down on preventative measures in the nick of time? Was it sheer luck — or perhaps a mix of factors?
It's impossible to say precisely, experts contend. But one thing is certain: The job of those who attempt to forecast the pandemic hasn't gotten any easier in the new year, given nascent variables such as the vaccine rollout and the development and spread of mutations.
Infectious disease epidemiologists are "trying to change their shoes while riding a bicycle in the middle of a hurricane," said Dr. Elizabeth Carlton, assistant professor at the Colorado School of Public Health and member of the state's COVID-19 modeling team, a multidisciplinary team comprised of experts at colleges and universities throughout the state.
Says Dr. Phoebe Lostroh, a Colorado College microbiology professor with a history of highly accurate El Paso County virus projections: "It's just as complicated as forecasting the weather, if not more so."
Wildcard: Human behavior
COVID directly caused 4,215 deaths in Colorado last year — less than a quarter of one of the worst-case scenarios proffered by the state's coronavirus modeling team last fall, according to preliminary data released Friday by the state that is expected to be finalized later this spring.
A Dec. 4 report from the state's COVID-19 modeling group warned that deaths could reach as high as 7,650 by the end of the year if the state saw an 11% reduction in transmission control — a measure that monitors how well Coloradans are adhering to pandemic-related behavior and policy changes — and a 30% drop in social distancing due to the holidays. At the time, it was on track for 5,600 deaths.
An Oct. 28 report, just 15 days ahead of the state's all-time high in new daily diagnoses, warned that Colorado's death toll could reach as high as 17,500 by the end of 2020 with a decrease in transmission control due to the holidays. At the time it was on track for 7,600 deaths.
Explanations for the lower-than-expected death toll are complex, involving a mix of factors like policy, behavior, politics — and perhaps even factors like climate and the health of Coloradans prior to the pandemic.
"We don't generate forecasts, we generate projections," Carlton said — "what if" scenarios for a multitude of factors including how quickly the virus is spreading and how well Coloradans are complying with transmission-control measures.
As to why the state's COVID death toll came in lower than anticipated, "the week before Thanksgiving we saw this dramatic reduction in the amount of population mixing and the contacts people where having," she said.
Consequently, hospitalizations began to fall by early December, and deaths followed suit, with the usual lag of a couple of weeks.
"It was some combination of policy and behavior that started sometime in late November, and we're in a much better place because of it," she said.
Lostroh's take: "The worst-case scenarios took into account the worst possible choices, and I don't think we made all of those," she said, speaking of Coloradans on a whole.
The models being produced on a state level are "so much more sophisticated than what I'm doing," Lostroh said, adding that longer-term forecasting, versus the week-by-week forecasting she does, is "inevitably more inaccurate."
The Harvard grad's modeling is based off of two "very simple assumptions": One, that the "virus is spreading exponentially," and two, that "every infected person infects, on average, more than one other person."
Assuming human behavior isn't changing, "when these two things are true, my forecast has been accurate, for weeks and weeks," Lostroh said. "When either is not true, it's no longer as accurate."
The biggest wild card in modeling, Carlton contends, is humans behavior.
"I think the hardest thing to predict in any model is how people are going to behave," Carlton said. "It's much easier to predict how the virus is going to behave."
A race between vaccine, variants
To make matters more complicated, there are new variables that weren't present a couple of months ago. For one, there's the vaccine rollout, contributing to a rising herd immunity. And then there are variants like the highly transmissible and potentially more lethal B.1.1.7, which triggered alarm when announced by U.K. officials in December and caused strict lockdown measures in southern England. There is also L452R, first seen in Denmark last spring and recently linked to several large-scale outbreaks in California.
Both have been found in Colorado, though only 33 variant cases had been identified in the state, with 13 cases under investigation. Only 30% of the state's positive cases are being screened for the B.1.1.7 variant, and genome sequencing to confirm it is only performed on approximately 3% of positive tests statewide, state health officials cautioned in a Friday news release.
"One of the challenges for any modeling team is responding in real time" to changing factors, Carlton said. Are COVID patients spending less time in the hospital? Are deaths dropping due to advancements in treatment? What mutations are present in Colorado, and how rapidly are they spreading?
The state's COVID modeling team is now building vaccination scenarios into its models, she said, as well as variant scenarios. But both are a "moving target."
For instance, it's unknown how rapidly the B.1.1.7 variant is spreading in the state.
"Right now it looks like in Colorado it's not spreading very rapidly," she said, adding that in the U.K., the variant "held at low levels for several months and then increased very rapidly."
Why would it spread so rapidly there and not here?
"I've been banging my head against the wall, trying to figure it out," she said.
The further ahead one tries to forecast, "the more uncertainty you have and the more you have to make educated guesses," Lostroh said.
But one thing is fairly certain.
"I think it's a race between the spread of that variant and vaccinating people," she said.