How One Epidemiologist Decided Whether to Send Her Children to Group Childcare
How to reason when information is incomplete, uncertain and emotionally-fraught
This issue of Insight is a phenomenal guest essay by Dr. Whitney Robinson on how she navigated the difficult and complicated information landscape early in the pandemic to puzzle through whether to send her two children to childcare. She explains not just her decision, but the underlying principles she used to evaluate uncertain and incomplete information (the meta-epistemology!) in order to arrive at a conclusion, including how she used existing knowledge to try to find the right questions to ask, how she evaluated absence of evidence of certain kinds of events despite high demand (an approach related to this earlier post) and the steps she took to try to guard against her sticky priors (like all the research on influenza and kids) to better look at what was in front of her.
I really love this piece because it is an excellent example of how to provide really useful tools and information to help people make their own decisions, without implying there is a single right answer that fits everyone, or that every other choice was wrong. And the real treat here is that the methods and principles she’s explaining are applicable in a great variety of contexts.
Without further ado, here’s the essay (and see you in the comments!).
More on meta-epistemology: an epidemiologist’s perspective
by Whitney R. Robinson, PhD, MSPH
Associate Professor of Epidemiology, University of North Carolina at Chapel Hill Gillings School of Global Public Health
In the January 31 issue of Insight, Zeynep wrote: “Reflecting on these anniversaries also offers a chance to talk a little bit about metaepistemology: the theory of theory of knowledge. Why did we know what we knew? Which in turn explains: why did we miss it?”
I ate this up and apparently many other Insight readers have, too. Your interest in this topic is likely a reason that Zeynep gave me this opportunity to write about meta-epistemiology from my perspective as an epidemiologist.
My love of meta-epistemology (how we know what we know) likely attracted me to epidemiology. Epidemiology is one of the fields within the health sciences that is uniquely comfortable with uncertainty. Other health fields seek out universal principles. For instance, one of molecular biology’s stunning insights is the double-helix shape of DNA. That shape and structure is a near-universally applicable insight on the nature of life on Earth.
In contrast, one of epidemiology’s hallmark insights is that chronic tobacco smoking greatly increases risk of death from lung cancer and cardiovascular diseases. It doesn’t sound quite as cool. However, with that knowledge, I can reliably predict that, when a lot of people in a community start smoking, in a few decades, that same place will reckon with a greater number of painful cardiovascular illnesses and deaths than they otherwise would have. Unfortunately, I can’t tell you exactly for whom the smoking will make a difference. I can’t tell you which person will smoke for 75 years then die peacefully in her sleep, or which person’s fate was shifted towards an earlier and harder end the moment she picked up her first cigarette.
Because epidemiology can have a fair degree of certainty on the population level but not for the individual, we traditionally make big sweeping recommendations for everyone with the goal of preventing tragedy for an unknown subgroup. We propose measures that we believe will save lives -- but we usually don’t know which lives. That’s a central tension in the field: asking for collective behavior change that may or may not directly benefit a given person or family. The idea is that certain collective actions will benefit many more people than they harm, but we can’t promise that any one person will directly receive more benefits than harms.
Temperamentally and by training, I’m pretty comfortable with this level of knowledge -- the big-picture population kind of knowledge that sits alongside a lot of uncertainty at the individual-person level. And that’s all well and good until you’re a working parent in the middle of a pandemic. It’s all well and good to say, “On the one hand this, and, on the other hand, that,” until you need to make a decision for your own small children, as well as their constellation of caregivers and family members. It’s all well and good to say, “Well, it depends,” until you are in a situation with little trustworthy national guidance and you actually do need to make a specific decision for your loved ones.
I’ve written on Twitter about my decision in March 2020 to keep my two young children (1 year old and 5 years old) in group childcare. I feel lucky that I had the option. But keeping them in daycare was a pretty out-of-step decision in my social circle at the time. One day my toddler was the only child in his classroom. It was just him and his two teachers. I’ve talked about my decision publicly over the past year, on Twitter and on my podcast, but I’m not trying to convince anyone else to make the same choices. If I’d been in different circumstances, such as having older relatives living in my household or in a job where I couldn’t risk a two-week daycare quarantine, I might have made a different decision. I’ve talked about my decisions because I was talking about my research life publicly. I couldn’t do that in good conscience without acknowledging the support I was getting through paid childcare. So many families, especially mothers of young children, have dealt with huge levels of gaslighting and burnout over the past year. I didn’t want to ignore the impossible trade-offs many families with young children faced because of the lack of a social safety net. I also wanted it to be clear what trade-offs I was making to reduce the risk for myself and the other members of the daycare community (e.g., no podding, no indoor activities with anyone outside our immediate family, purely outdoor meetups).
Because this was a tangible decision that I made early in the pandemic, when research was limited, I thought it was a good example for Zeynep’s meta-epistemiology series. When almost everyone else was keeping their kids home, how did I decide to send mine to daycare? In the spirit of Zeynep’s previous posts, I will answer that question using 3 principles.
Principle 1. “Look to previous phenomena to know what questions to ask”
I’m not an infectious disease epidemiologist, but I had enough basic knowledge from my PhD training to know what questions to ask. As a student, I learned two important things about the dynamics of another respiratory infection, influenza. First, I learned that, with respect to age, sickness (morbidity) and death (mortality) typically show a J-shaped curve. The very youngest (especially infants younger than 12 months) and oldest are mostly likely to die of influenza during an outbreak. The very oldest (75 years+) suffer the highest rates of flu death. In a typical flu season, school-aged children and young adults are at the least risk. So my first question was, are young children likely to get very sick or die during a COVID-19 outbreak?
But when I was an epidemiology student in the early 2000s, I’d also absorbed the lesson that groups at low risk of morbidity and mortality could, counterintuitively, be high-risk transmitters. This insight was exemplified by studies like this 2001 New England Journal of Medicine article which concluded that vaccinating school-aged children in Japan prevented flu deaths in the elderly. The article determined that stopping that school vaccination program resulted in a big increase in influenza deaths among the elderly. Even though school-aged children faced low risk of death themselves, they were a key driver of flu deaths for older people. So my second question was, are young children especially likely to spread COVID-19 to other people?
Principle 2. “Observed versus expected." In other words, “Pay attention to unexpected data that has no natural constituency and to lack of data that are in high demand”
The best sources of data about morbidity and mortality were hospitalization and mortality data from China and hard-hit countries like Italy. And the data were consistently different than the classic J-shaped mortality or hospitalization charts of the flu. Instead, hospitalization and mortality rates increased exponentially with age. In fact, mortality for the youngest children was similar to that from flu. That was a helpful anchor for me in thinking about absolute risk to young children.
Understanding children’s contribution to transmission in childcare and school settings was more complex. There was not going to be a randomized controlled trial or enough time for a natural experiment like Japan’s experience with changing school vaccination policy. Instead, I triangulated among multiple data sources. First, I wanted to understand infection: there’s no transmission without infection. US surveillance data were useless because of limited and haphazard testing. To understand age-specific infection risk, I relied on places like Iceland and the Faroe Islands that were conducting representative or extremely widespread testing of current or past infection, regardless of symptoms. When places conducted this kind of testing early on, before the most at-risk groups knew how to protect themselves, young children consistently had low prevalence of infection. (In China, schools had been closed when the outbreak came to light, but the same wasn’t true in all other countries.) Then there were the household studies. The strength of these studies is that the researchers study a family in which at least one person is infected. Everyone in the family is exposed and presumed to be at high risk of infection. Age differences in which family members get infected tell us about susceptibility to infection. There are lots of these studies now. The early ones weren’t all perfectly conducted, but they consistently indicated that the young kids in these families were least likely to become infected when there was a positive family member. And there were also early indications that they were less likely to be the index (or first) case in a household as well.
I gave these studies a great deal of weight because the results were unexpected, consistent, and seemed to please no one. There was no natural constituency for the narrative that kids under 10 years old in particular definitely could get infected but were much less likely to get infected and transmit than older people. It was an oddly specific finding that no one had predicted or set out to prove. Because the particulars were so unexpected and seemed to serve no one, I believed they were likely to be valid.
On the other hand, by late March, there was a huge demand for another type of story: news reports of COVID-19 spreading in schools and childcare centers. I am a member of several Facebook groups for academic researchers who are mothers. The members of these groups were hungry for information on children and SARS-CoV-2 transmission dynamics. As I watched the same stories circulate again and again in March, I became convinced that the many childcare centers and primary schools that remained open across the country were not producing an outsized number of outbreaks. Many of the stories I saw, besides being rehashes of the same incidents, stretched the truth to make it seem as if children were fueling outbreaks. There would be a headline about 20 cases among teachers in a school district in Georgia. But when I read the article, the spread would be among teachers gathering in a building for a meeting among themselves or to teach remotely. I saw multiple instances of COVID-19 outbreaks among adults conducting remote teaching purposely given headlines that implied that schoolchildren were involved. This told me that the media knew there was strong demand for stories of children at risk for COVID-19 in childcare settings and that there weren’t enough verified outbreaks to meet that demand.
There were alternative explanations for the relative lack of stories of daycare-based outbreaks. As I mentioned earlier, testing was extremely limited in the U.S. in late winter and early spring 2020. That meant cases among children would be easy to miss. However, if cases were spreading quickly from childcare centers to families, I would expect some of those to result in severe illness among family members and lead to contact tracing. Because childcare and school settings were seen as high-risk early in the pandemic and as critical infrastructure for first responders, most local public health agencies would have prioritized a possible outbreak at a school setting. Further, schools and childcare centers are among the few settings that had existing infrastructure for infectious disease disease reporting before the COVID-19 outbreak. For instance, in North Carolina, the only entities obliged by law to report known cases of communicable diseases are school principals and childcare operators, physicians, restaurants, and operators of scientific laboratories.
Both childcare and restaurants are highly regulated businesses, but childcare facilities are even more likely than restaurants to detect and report a case because the population it serves is fixed and has ongoing relationships with the center. If a big group comes to a restaurant and a presymptomatic diner infects others in his party, the restaurant owner would have no way to know. However, if a child stopped showing up at daycare because family members were sick, the daycare operator is likely to find out. Even with the challenges of limited testing and anecdotal evidence of some families hiding children’s infections, the demand for news stories about kids and COVID-19 in schools, ongoing relationships that schools and childcare centers have with families, and the data infrastructure of reporting that preceded COVID-19 convinced me that many school-based outbreaks would receive the wider attention if they were happening on a frequent basis.
Footnote 1: And then there was this case study about families on ski vacation in the French Alps in February. A thorough contact tracing effort found that an infected child did not transmit to anyone despite many school-based contacts at 3 different schools. It’s totally anecdotal, but travel envy really got this scenario stuck in my head.
Footnote 2: Despite the evidence, many people remain convinced that school settings can’t be safe because they’re environments where so many people are indoors together. I think this is where exponential growth frustrates intuition. Why would it be safe to have a bunch of bodies indoors in one context but not another? The way I think of this is the way I think of more infectious Variants of Concern, like B.1.1.7. Even with only 15% more transmissibility, a Variant of Concern could catalyze a huge increase in virus spread because of the counterintuitive math of exponential growth. But it goes the other way too. A population, like young children, that is ⅓ or ½ as likely to get infected or spread disease, can experience an exponentially lower level transmission together than adults doing the same tasks.
Principle 3. “Beware of ‘sticky’ priors”
A major topic of conversation in Zeynep’s newsletters has been why expert consensus got some important things wrong at first. I have two thoughts on this. First, experts are human and can get too attached to knowledge gained in other settings. I think that some types of knowledge are particularly “sticky.” For instance, counterintuitive facts that we learned during formative stages of training, such as the Japanese school-aged flu transmitters, can be particularly hard to let go. Also, observations that are reinforced by personal experience often have outsized weight when we are evaluating new situations.
With regard to COVID-19, I didn’t have a lot of strong priors because I had never worked in infectious diseases. But I have worked a lot in racial inequities in non-infectious diseases and also in fields like obesity and cancer. Because of that work, I did have strong priors about the causes of racial inequities in COVID-19 and obesity as a risk factor for COVID-19 mortality. My priors about racial inequities were mostly correct. However, my skepticism towards obesity as a risk factor likely underestimated its riskiness. That’s the thing about strong priors -- sometimes they are a shortcut to truth, sometimes they mislead. And it’s difficult to predict when one’s hard-won scientific intuition will be prescient and when it will be foolish.
Another potential explanation for notable failures to integrate new data quickly and shift thinking is exemplified by this article: “Does Science Advance One Funeral At A Time?” I often think of this article. Its central premise is that a few star senior scientists can impede the flow of powerful and novel ideas in whole areas of inquiry. As explained in this summary in Science, the effect can be intangible and even inadvertent. Phenomena include “Goliath’s shadow,” “intellectual closure,” and “social closure.” I’ve worked in fields like this, where a dominant paradigm is so strong that it feels impossible to get a foothold for an alternative direction for research. Advancing a new way of seeing things -- even if that alternative view is pretty obvious to an outsider carefully looking at the data -- seems to hit roadblock after roadblock or just get ignored. In these circumstances, one has to become near fanatical to break through. I’m not sure if this is the case with some areas of infectious disease. But, for better or worse, scientific micro-communities have cultures just like all other communities.
When people ask me my opinions about COVID-19 and kids now, I mostly demur. On the one hand, I’m still sending my younger child to our longtime daycare. And my older child is now attending in-person kindergarten. We are still cautious in our daily lives, but we will see extended family soon (all adults are vaccinated) for the first time in over a year. Here are some reasons why I feel pretty confident about my kids being in in-person group childcare. Vaccination rates are high among the staff at the daycare and the school. Mask compliance is high in both settings. Both settings have very long, evidence-based handbooks of their safety procedures. For instance, although my kindergartner eats lunch inside at his school, mask-off time only lasts 10 minutes and is silent: the teacher puts on an episode of The Magic School Bus. The kids are riveted.
However, the justifications in the previous paragraph are not the whole truth. The rest of that truth is that I’m leaning on my priors now. I haven’t done a comprehensive review of the literature since summer 2020. Instead, my decisions today are heavily influenced by my priors from my deep investigation last spring. The truth is, I made a decision a year ago that worked out well. That experience has reinforced my beliefs.
But things are different now than they were in March 2020. Now we know about Long COVID and MIS-C. There are more infectious variants and some with potential to partially evade vaccines. But there are also amazing vaccines. They are so, so remarkably powerful. When I do quick mental math, I assure myself that the high vaccination levels in our community, my family’s continued precautions, and the thoughtful precautions at our daycare and school mean that my family and school staff remain at extremely low risk of infection. But I also know that I am taking a mental shortcut. I hope you’ll have some grace for me. It’s been a long year.
This was wonderful. I wish every scientist wrote like this! Love the phrase "sticky priors." I've always thought of them as axioms, or maybe that's related, but am switching to "sticky priors." Thank you for sharing this with us.
Thank you for this! Especially the section on unexpected or absent results, factoring in the supply and demand for different types of data and findings. That's a tough skill for lay people like me, but the essay handles it elegantly and practically. I try to put on my Bayes hat to weigh priors and new data, but sometimes it narrows my perspective. The essay gives such a great model for improvement.