Listen to the podcast:
When the coronavirus pandemic took hold, millions of nonessential workers had the benefit of shifting from the office to the safety and convenience of home. But that wasn’t a practical option for essential workers employed in jobs requiring a high degree of interaction with the public or with each other. Doctors warned that the risk for disease transmission would be greater for these vulnerable workers, and they were right. A new study has measured that risk, revealing the real threat for essential workers and the potential benefits from business closure policies. Scientists from Penn collaborated with researchers from Independence Blue Cross on the study titled, “The Impact of the Non-essential Business Closure Policy on COVID-19 Infection Rates,” which was recently published by the National Bureau of Economic Research. Hummy Song, a Wharton professor of operations, information and decisions, and Aaron Smith-McLallen, director of data science and health care analytics for Independence Blue Cross, are two of the co-authors. They joined Knowledge@Wharton to discuss the findings.
Listen to the podcast above or read an edited transcript of the conversation below.
Knowledge@Wharton: Why was it important to study this, and what did you find?
Hummy Song: We thought that this would be important to study since many state and local governments were issuing, and continue to issue, these business closure orders, and it was unclear to what extent it would have any effect at curbing infection risk. This ultimately impacts how we think about weighing the negative economic effects of closure — such as lost wages, unemployment, etc. — against the desired health effects, specifically the reduced incidence and transmission of COVID-19.
Our goal in this work was to examine the extent to which being designated as an essential worker versus as a nonessential worker by this policy impacts your risk of being positive for COVID-19. We took this one step further and also looked into how it impacts the risk of those who are living with essential versus nonessential workers.
You might wonder how we determine who is essential and who is not essential. To do that, we simply use the definitions as stated in Pennsylvania Gov. Tom Wolf’s nonessential business closure order, because our data come from Pennsylvania. Some examples of essential workers are those employed by hospitals, transportation systems, food manufacturing, etc., whereas nonessential workers include those in mining, construction, general merchandise stores, etc.
“Workers who were deemed essential have a 55% higher likelihood of being positive for COVID-19, compared to those who are classified as nonessential.” –Hummy Song
Our main finding is that workers who were deemed essential have a 55% higher likelihood of being positive for COVID-19, compared to those who are classified as nonessential. In other words, nonessential workers absolutely seem to experience a protective effect from this policy.
The other finding I wanted to highlight is that it’s not only the essential or nonessential workers who are experiencing this difference in infection risk, but also the other members in their household. Specifically, we’re finding that, compared to those who are living with a nonessential worker, dependents living with an essential worker have a 17% higher likelihood of being COVID positive, and roommates living with an essential worker experience a 38% higher likelihood of being COVID positive.
Knowledge@Wharton: Let’s talk about the data, and this is where you come in, Dr. Smith-McLallen. Your company provided this dataset, correct?
Aaron Smith-McLallen: This is another great collaboration between the University of Pennsylvania — Wharton, in particular — and Independence Blue Cross. At the beginning of the pandemic, we were thinking about how we, as a health insurer, along with our research partners could add a unique perspective to understanding the impact that COVID-19 is having on our community.
We got to thinking about Gov. Wolf’s nonessential business closure policy here in Pennsylvania, and we realized that we were in a unique position to measure the effectiveness of that policy on reducing infection rates. This is a unique analytic problem that requires us to link COVID status and employer sector, or the industry that a person works in, at the individual level. To evaluate the policy impact, we needed data from individuals who tested positive, but also from those who were not tested or tested negative, and to be able to identify which industry they worked in. So, we needed to have the numerator of the people who were positive, and the denominator of all of the other people who either didn’t get tested or who tested negative, and then link that to the industry.
Health systems and government agencies see a small portion of the population who get tested, but they don’t see who didn’t get tested. Those agencies also can’t reliably link the tests that they do see to the industry that the person works in. So, there’s really no way for any other entity other than a health insurer to provide this level of analysis about this policy and take this granular of a look at it.
I think the data that we have are really informative as to the impact of policy. Most people get their health insurance through their job. As an insurance company, we know who is associated with each company as either a policyholder or a dependent. And we know the industry they work in because every company has a NAICS (North American Industry Classification System) code associated with it. That’s the code that the Wolf administration used to identify industries as essential or nonessential. The dataset is also unique in its breadth. Independence Blue Cross provides medical insurance to more than 50% of commercially insured individuals in the greater Philadelphia area.
Knowledge@Wharton: Philadelphia is the fifth- or sixth-largest metro area in the U.S., depending on the listing. Does that mean the findings can be generalized to other metro areas? What about rural areas?
“I think the data that we have are really informative as to the impact of policy.” –Aaron Smith-McLallen
Smith-McLallen: We expect the overall takeaway to be generalizable, even if the magnitude of the effects might differ depending on the specific population. The estimates that Dr. Song spoke about may be, in fact, conservative ones. As I mentioned, our data represent those who receive coverage through their employer, so our sample doesn’t include uninsured individuals or those with other commercial insurance, those covered with individual plans through the ACA or Medicaid.
There’s a large segment of the population that we’re not seeing. But many of those — those on Medicaid, individual plans, the uninsured — might be more likely to be essential workers than nonessential workers. I think what we’re reporting are potentially conservative estimates.
Knowledge@Wharton: It’s been widely reported that COVID-19 is disproportionately affecting minorities and neighborhoods of color. How does that information factor into the findings of your study?
Song: I think that’s a really important aspect of what everyone should be paying attention to. This is part of what also motivated us to start thinking about intra-household transmission risk. If you think about some of those populations, what was being reported in the news in terms of possible differences between higher-income areas versus lower-income areas, is the density of housing and the inability to perhaps quarantine if one member of your household comes home sick. There’s not enough room to do that, and you tend to have more people living in a smaller space. That’s part of what we hope we are able to speak to through that intra-household transmission aspect.
We do control for these potential characteristics that may differ across these populations when it comes to not only the age and gender of the population, but also the ZIP code in which someone is living, and how that might be related to some of these factors that you’re mentioning. We do our best based on the data we have to be able to account for those factors, but I will acknowledge that we haven’t specifically drilled down into certain segments of the population beyond the initial categorization of essential versus nonessential workers.
Knowledge@Wharton: What can we take away from this study that perhaps will help us reduce the rate of transmission and get us to the finish line faster in this pandemic?
Song: We hope that by quantifying these effects, we can help policymakers develop and think about what’s the appropriate and necessary protection that workers need, both when it comes to workplace safety, but also for things like the right to refuse unsafe work. It’s whatever can be done so that people don’t have to be making a choice between their paycheck and their health.
“We hope that by quantifying these effects, we can help policymakers develop and think about what’s the appropriate and necessary protection that workers need….” –Hummy Song
Smith-McLallen: In a broad sense, what the nonessential business closure policy did was to create a situation that limited interpersonal contact for nonessential workers who were staying at home. But it also limited contact for essential workers who were perhaps commuting with fewer people, for example, and not necessarily exposed to all of the people who were staying at home. That secondary protective effect was very effective at reducing cases.
Another thing about that secondary protective effect is we might think that if there would have been no nonessential business closure — if the nonessential workers had gone out to work — their infection rates would have been the same as we observed among the essential workers. There would be no difference. That’s what the results of our study speak to. However, there is a real possibility that the rates for everyone would have been considerably higher, even higher than what we observed in the essential worker population, just because of the increased contact and exposure across the board.
What I think policymakers should take from this research is that with new strains of the virus being discovered, if we reach a point where we need to aggressively limit contact and transmission, nonessential business closure policies can be effective. And now we can quantify just how effective they can be.
For example, if a state had information that X percent of the population worked in essential industries, and Y percent worked in nonessential industries, they could take the results of our analyses and come up with conservative estimates of the number of cases that could be avoided, and potentially the number of deaths that could be avoided, with business closure policy. Obviously, those benefits would have to be weighed against the cost, the jobs, the economy, and so on, but at least now that benefit can be quantified whereas before it couldn’t.
Knowledge@Wharton: When you’re talking about that secondary protective effect, you mean that nonessential workers create a little bit less exposure when they are staying home, correct? That makes sense if the goal of less contact is to flatten the curve.
Smith-McLallen: Exactly. When everybody’s just out there in society, there’s just that much more interpersonal contact and connection with people. If you take a large segment of the society and keep them home, obviously they’re protected. But the people who are out there doing essential business and essential work are now not exposed to a huge segment of the population, so it reduces their risk as well. If everybody were out there, if we didn’t have the nonessential closure policy, I think the rates for everyone would be a lot higher than what we observed in our study, even among essential workers.