Is COVID-19 Becoming More Dangerous?
Last week, I shared some good news: Although COVID-19 infections are increasing, owing to the spread of BA.5, your behavior still impacts your risk of infection more strongly than other variables, including the variant you're exposed to.
Given that BA.5 is more infectious, we can also ask: Is it more dangerous? For example, does BA.5 lead to more hospitalizations than earlier strains of coronavirus?
In recent weeks, scientists, public health officials, and journalists have offered two, apparently contradictory answers to this question:
Answer 1: BA.5 is milder than earlier variants and less likely to result in hospitalization.
Answer 2: As BA.5 becomes more dominant, COVID-19 hospitalizations are increasing.
If you Google "BA.5" or "BA.5: what you need to know", you can quickly find dozens of examples of each answer. But these answers can't both be right, can they?
In this newsletter I'll walk us through some statistics that resolve this apparent contradiction. At the end, I'll offer some practical advice.
Resolving the contradiction: What doesn't work
We can't say that BA.5 is indeed milder, but some other variant is causing the spike in hospitalizations, because those other variants have been rapidly disappearing. As of July 16, BA.5 was the source of 78% of new COVID-19 cases, up 13% from last week.
We also can't attribute the contradiction to competing ideologies. It's not only laissez-faire conservatives who say that BA.5 cases tend to be mild, just as it's not only worried liberals who point to rising hospitalization rates. Experts and journalists of all political persuasions have been making these observations.
Resolving the contradiction: A first pass
In theory, both answers could be correct. Hospitalization rates could be increasing in recent weeks simply because the total number of cases has increased. This would be consistent with Answer 2. At the same time, if the proportion of cases leading to hospitalization is declining, we'd have a pattern consistent with Answer 1. In short, the contradiction is resolved if there are more COVID-19 patients but proportionally fewer hospitalized ones.
You might be tempted to think that we have all the data we need to evaluate this possibility. For example, during the first week of July, there were 107,174 cases and 5,115 hospitalizations per day, on average. During the second week of July, those numbers were 124,048 and 5,851, respectively.
The temptation here is to divide hospitalizations by cases. If we did so, we'd find that for the first week of July, 4.77% of cases led to hospitalization (that's 5,115 divided by 107,174, multiplied by 100). During the second week of July, 4.71% of cases led to hospitalization (124,048/5851 x 100). In other words, although the total number of hospitalizations increased during July, the percentage of cases leading to hospitalizations actually diminished fractionally. That would be a happy resolution to our contradiction.
Unfortunately, this approach to calculation is too prone to error to be more than merely suggestive:
1. Some people are hospitalized the same day they test positive; for other people, days or even weeks pass between the positive test and the hospitalization. The more of these lags in the data (and the greater their extent), the more error is introduced by dividing daily hospitalizations by daily cases.
2. As I mentioned last week, reported case rates may be anywhere from three to 10 times lower than true case rates. And, there's no guarantee that the extent of error is the same from week to week. Mathematically speaking, if you divide hospitalizations by cases, the result could be wrong, by a lot, given that the true value of the denominator (i.e., cases) could vary so much.
3. Hospitalization data are more accurate than case data, but they're not perfect either. Data may be missing or inaccurate, given how overburdened hospitals and public health staff have been during the pandemic. And, "hospitalization" still isn't consistently defined. A person hospitalized for a cardiovascular issue who's found to have a mild COVID-19 infection may be added to the COVID-19 hospitalization database, even though the infection had little or no impact on the need for a hospital.
4. Not all infections and hospitalizations stem from BA.5. (This is a minor and diminishing source of error, since BA.5 now accounts for 77.9% of cases. BA.4 accounts for an additional 12.8% of cases, and its effects on infection rates and hospitalizations is not markedly different from BA.5.)
In sum, the spread of BA.5 led to higher case rates and more hospitalizations, but it's unclear whether the percentage of hospitalized cases is increasing, decreasing, or staying the same. All we can say at the moment is that throughout the month of July, the percentages of COVID-19 infections that led to hospitalizations seemed to decline fractionally, even though the total numbers of hospitalizations per day increased.
Now for some good news, then some more good news.
The good news
The good news consists of other evidence that BA.5 isn't more dangerous than earlier variants.
1. Although hospitalizations have been increasing since mid-April, the extent of increase has been small. There are different ways to operationally define "extent of increase." For instance, the average numbers of hospitalizations per day (5,851 through last week) are still lower than they've been at any time during the pandemic except for May through June of 2021 (and, perhaps, during spring 2020 when hospitalization data were particularly inaccurate).
It's also helpful look at how rapidly hospitalizations are increasing. Since the lowest point this year (the week of April 11), we've seen a roughly linear rise from an average of 2,114 hospitalizations per day to 5,851. In short, hospitalizations have nearly tripled since mid-April. Although that's concerning, comparable spikes in hospitalizations have occurred four times during the pandemic. It's unclear how long the current trend will continue, but I wouldn't call it bad news just yet.
Finally, we can look at predictions of future changes in hospitalization rates. The CDC does so based on 16 different models provided by organizations ranging from the Harvard School of Public Health to Facebook AI. The CDC combines these models into a single "ensemble" forecast which, to be blunt, is shockingly unhelpful. For example, in their July 20 forecast, the CDC estimates a 95% probability that by August 12, the number of hospitalizations per day will be somewhere between 3,100 (which is almost half the current rate) and 13,800 (which is almost three times the current rate.) In other words, hospitalizations may go down a lot, go up a lot, or fall somewhere in between. Who could've guessed?
3. Hospitalization data from other countries are mostly consistent with trends in the U.S. In the U.K., France, and Italy, for example, hospitalizations this summer have been increasing from historic lows but still remain lower than at any prior point during the pandemic.
Two interesting exceptions to this trend are South Africa and Portugal. In South Africa, hospitalization rates have barely changed since BA.5 became dominant, even though the number of cases increased and vaccination rates remained low (just under 50% of South African adults are fully vaccinated). The main explanation proposed for this is "hybrid immunity" resulting from vaccines plus an unusually high rate of prior infections across several variants. As for Portugal, hospitalizations have been rising to pre-omicron levels, even though the national vaccination rate is roughly 87%. This may be attributable to the fact that Portugal has one of the oldest populations in the world (median age 46.2 years).
More good news
With respect to the dangerousness of BA.5, there's more good news when we look beyond hospitalization rates to more serious outcomes – intensive care unit (ICU) admissions, and mortality.
The percentages of hospitalized patients in the U.S. who are admitted to ICUs have remained constant at 11% per week since early May. This is remarkably consistent, and it's the lowest percentage for any week throughout the pandemic (at least since summer 2020, when reliable data were first available).
As for mortality rates, everything I've written about hospitalization rates applies roughly equally well to this outcome. COVID-19 mortalities have been increasing recently, but the extent of increase has been small (even smaller than for hospitalizations). As with hospitalizations and ICU admission rates, mortalities are disproportionately observed among patients who are unvaccinated and/or elderly and/or immunocompromised.
Summary and advice
At the outset of this newsletter I mentioned a pair of contradictory statements you often see the news. Here's how I would reconcile them: BA.5 is more infectious than most earlier variants, but it's also relatively mild. COVID-19 hospitalizations are increasing because there are more cases overall. The percentage of cases leading to hospitalizations isn't changing much, nor is the percentage of hospitalizations that are lethal or require ICU care. As with prior variants, the majority of hospitalized patients are unvaccinated or otherwise vulnerable. In a word, BA.5 isn't particularly dangerous.
At the outset I also promised some practical advice related to personal risk. In sifting through the hospitalization stats, I was reminded of how many different variables may – or may not – influence the chances of getting a BA.5 infection as well as its severity. I was also reminded of how much statistical uncertainty surrounds each of those variables. Whether or not you've been vaccinated is critical, as is your immune system functioning. Beyond that, risk is influenced by boosters, length of time since last shot, numbers and types of prior infections, case rates in your city, etc., but the impact of these and other variables is small and uncertain. So, what practical advice might be drawn from these data?
My advice – besides getting fully vaccinated, if not boosted, and minimizing time in densely crowded spaces (unless properly masked) – is to relax. A close look at the data tells us that strong statements about managing personal risk are mostly unwarranted. If you're vaccinated, then the main determinant of risk is your social behavior, and each behavioral decision you make influences your risk by some unknown amount – probably a small amount, but certainly unknown. Another way of saying this is that most of the statistics aren't very helpful in assessing personal risk. Imagine reading that in a statistics newsletter!