Is Delta More Dangerous?
Being old and a little cranky, I sometimes wonder why people don't look at studies that make the news. After all, news reports often include hyperlinks, and when the linked studies are open access, one click takes you straight into the weeds. Why not have a look?
Here are four reasons for not looking that people have shared with me over the years (and which, in some cases, explain my own reluctance to look):
1. There's no point in reading the study, because it's technical. I wouldn't understand it.
2. There's no point in reading the study, because the details aren't meaningful.
3. There's no point in reading the study, because I trust the source. I'm happy with a brief summary.
4. There's no point in reading the study, because I can't trust the source. I don't want to waste time on something that’s "rigged" to promote someone’s scientific or ideological agenda.
Most of this newsletter focuses on a Delta variant study, published last week, which illustrates that #1 and #2 can be wrong. Sometimes a study of great personal relevance will be deeply, hopelessly flawed in a way you can easily spot, even if you're not an expert in the field. When that happens, you don't have to choose between #3 and #4 (or find some nuanced middle ground between them). You can judge the study for yourself.
Since not all studies are open access, I will close with a brief look at this week's news reports on "Delta plus", in order to show that #2 can be wrong even with respect to second-hand discussions of research. Often the smallest details matter.
Delta Dangerousness
We've heard for many months now that the Delta variant is more dangerous than earlier strains of SARS-CoV-2, in the sense that Delta is more infectious, and more likely to cause severe illness and death. However, on October 22, the CDC published a study showing that among adults hospitalized for COVID-19, the proportions of severe health outcomes were not greater during the Delta period as opposed to pre-Delta.
The CDC study did confirm that Delta is more infectious, and that hospitalization rates increased after Delta became the predominant strain. However, the researchers also claimed that Delta is not otherwise more dangerous. Specifically, they examined the medical records of 7,615 adults hospitalized for COVID-19 and found no significant differences in percentages of severe outcomes between the pre-Delta period (January-June 2021) and the Delta period (July-August 2021). ("Severe outcomes" consisted of admission to an intensive care unit, receipt of invasive mechanical ventilation, or death.)
You may be wondering about the phrases "pre-Delta period" and "Delta period". Why didn't the researchers refer to "patients with a pre-Delta variant" versus "patients with Delta"? Well, because they couldn't. They didn't do the genomic sequencing necessary to identify which strain of SARS-CoV-2 each patient had. Instead, they compared patients hospitalized at one time period (January-June 2021) to patients hospitalized at a later time period (July-August 2021), under the assumption that the former represents a pre-Delta group and the latter represents a Delta group.
Perhaps you've already noticed an enormous, honking methodological problem here. It's almost certain that some of the pre-Delta group had Delta, while a few members of the Delta group had a pre-Delta strain. The fact that the researchers found no significant differences between the groups may be attributable to this overlap. In other words, Delta patients in the pre-Delta group may have increased the percentage of severe outcomes for that group so that it became closer to Delta group levels, while pre-Delta patients in the Delta group may have lowered the percentage of severe outcomes for that group so that it became more like the pre-Delta group. This is like reporting no differences in intoxication between a group who drank 4 ounces of beer and a group who drank 4 ounces of vodka, but discovering later that the beer group included vodka drinkers, and the vodka group included beer drinkers. Not very convincing, right?
The problem here isn't merely speculative. Here’s a stat (from the CDC itself) that hints at the possible extent of error: During June 2021, the percentage of new COVID-19 cases in the U.S. caused by the Delta strain (according to individual-case genomic analyses) increased from 9.5% to 82.2%. Thus, in a "pre-Delta" group sampled between January and the end of June, some of these folks almost certainly had the Delta strain. A lesser concern, but still concerning, is that at least a few members of the Delta group (July-August 2021) might've had a pre-Delta strain, because over the course of July, the percentage of new cases that were still pre-Delta ranged from 17.8% down to 3.5%. In short, I'd say that a substantial number of folks in the beer group were drinking vodka, and a few people in the vodka group were drinking beer.
There's no way of correcting for these kinds of misclassifications, because the CDC was unable to run genomic analyses, and they didn't even note how many patients were sampled each month. All the same, we can do a little guesswork as to the possible extent of error. If approximately equal numbers of patients were sampled during each month of the study, 952 "pre-Delta" patients would've been selected in June, and 952 "Delta" patients would've been selected in July. Let's assume then that the percentages of misclassified patients in each group represent the middle values documented for that month. In other words, since the percentage of new cases caused by Delta rose from 9.5% to 82.2% in June, let's round off those numbers and take the middle value of 46% as an estimate of how many June patients in the pre-Delta group actually had Delta. And, since the percentage of new cases caused by pre-Delta variants ranged from 17.8% to 3.5% in July, let's take the rounded-off middle value of 11% as an estimate of how many July patients in the Delta group actually had a pre-Delta strain.
For the June sample, 46% of 952 would be 438 patients who were misclassified as pre-Delta. For the July sample, 11% of 952 would be 105 patients who were misclassified as Delta. That’s a total of 543 patients, or 7.1% of the entire study sample. (The breakdown would be 7.4% of the entire pre-Delta group misclassified as Delta, and 6.3% of the smaller Delta group misclassified as pre-Delta.)
Although the CDCs statistics were crude (nothing but Chi square tests) and under-reported, as far as I can tell my estimated misclassification rates for each group (7.4% and 6.3%) are sufficient to account for the absence of group differences. In fact, I reran two of the CDC's analyses, based on additional assumptions about their data, and found that in each instance, the proportion of severe outcomes was significantly greater for the Delta group than for the pre-Delta group. So, yes, Delta is more dangerous, in many senses of the term "dangerous". However, the point of this newsletter is not that we should be re-analyzing study data (it's usually impossible without the raw numbers). The point is that if you have the time, it's worth looking at actual studies rather than just summaries of studies, if only because huge flaws may be obvious.
The CDC's reputation has been tarnished in recent years owing to evidence of susceptibility to political influence (from both the Trump and Biden administrations) and to mixed messaging related to the pandemic. So, before making a final, particularly negative remark about their study, I want to acknowledge the agency’s enormous impact on public health (see here for some examples). As for the pandemic in particular, one of the CDC's many contributions is its Data Tracker site, which provides a rich trove of regularly updated information on case rates, outcomes, vaccinations, etc. I’m grateful for what the agency has done and continues to do for us.
Having said that, I would call this particular study one of the most disturbing ones I've encountered this year. Why such strong language? Because at least four larger studies using genomic sequencing have shown that the Delta variant results in more severe outcomes than pre-Delta variants do. For example, a study of 212,326 Canadians showed that compared to earlier strains of SARS-CoV-2, Delta infections significantly increase rates of hospitalization, ICU admission, and death. In light of studies like that, who benefits from the CDC publishing a study that's much smaller and, lacking genomic sequencing, can't clearly distinguish Delta from non-Delta strains? That study was a terrible waste of our tax dollars. (I did say I'm old and a little cranky, didn't I?)
Delta Risk
Now that I've suggested we ignore the CDC study in favor of larger, better ones attesting to Delta's relative dangerousness, I want to offer some reassurances, first concerning Delta and later for "Delta plus".
Being fully vaccinated still offers substantial protection against all known variants of SARS-CoV-2. Estimates of effectiveness against Delta vary widely, so it's helpful to remember that effectiveness stats are relative rather than absolute values (as I explain here). 60% effectiveness doesn't mean that 60% of vaccinated people won't develop COVID-19. Nor is it a prediction that 60% of the time a vaccinated person encounters the virus they won't develop the disease. Rather, it simply means that compared to someone who isn't vaccinated, vaccination reduces a person's chances of developing COVID-19 by 60%. Although we’re happier to see vaccine effectiveness stats closer to 90% than to 60%, the latter is still better than nothing. 60% better, actually.
Nobody is ever quite satisfied with what I just wrote, because we all want to know: What are my chances of getting COVID-19? Your actual, personal risk of infection depends on many factors, including some you can't control (e.g., who crosses your path) and some that you can (e.g., how vigilantly you mask). In short, effectiveness stats don't translate into precise estimates of personal risk. All you can be sure of is that the specific actions you take, from vaccination to routine behaviors like masking at a grocery store, each reduce your risk of infection by some specific (if not large or readily quantifiable) amount.
I've been asked about the apps you can use to estimate personal risk, These apps do take into account important variables like case rates where you live, the kinds of activities you'll be pursuing, and so on, but they make a lot of assumptions (e.g., how many people at a gathering you plan to attend will have COVID-19) and they're not able to include a lot of information of relevance (e.g., how ventilated that gathering place will be), so their risk estimates are necessarily crude. If nothing else, they serve as useful reminders of the kinds of things we can do to protect ourselves.
"Delta Plus"
In June, a new sublineage of Delta, AY.4.2, was identified in 11 countries. This sublineage, referred to popularly as "Delta plus", now accounts for the majority of new COVID-19 cases in the U.K. and appears to be more transmissible than all previous strains.
It's too early to know what the scope and impact of Delta plus will be, but concerns have been raised by the U.K. data, and by phrases like "more transmissible than all previous strains". (Even the name "Delta plus" sounds a little scary.) However, a close look at the news this week offers some reassurance.
First, although a few cases of Delta plus have been identified in the U.S., they don't seem to have clustered or occurred with increasing frequency over time, according to the CDC director (also see here). This is not to say that Delta plus won't spread rapidly someday, but the low frequencies thus far seem tentatively reassuring. (In contrast, Delta showed weekly increases in frequency.)
Second, Delta plus has been estimated to be 10 to 15% more transmissible than Delta. Although this isn't good news, I take some reassurance in recalling that Delta was roughly 60% more transmissible than Alpha, and Alpha was roughly 50% more transmissible than earlier, wild-type strains. In other words, Delta plus may introduce only a relatively small increase in current levels of transmissibility.
Finally, it's also reassuring (if you're fully vaccinated) that "transmissibility" is defined in terms of rate of spread. Although Delta plus may spread more rapidly than other strains, new infections primarily occur among unvaccinated people (and roughly 99% of those who become hospitalized for COVID-19 and die are unvaccinated). So, if you're fully vaccinated, and you tend to comply with the familiar litany of advice (distancing, masking, avoiding unventilated spaces, etc.), the increased "transmissibility" of Delta plus may not imply an appreciable increase in personal risk of infection.
Thanks for reading!