Covid-19 Hospitalizations Are on the Rise — Here’s What You Should Know (Part 1)
How to break down the raw numbers like a pro
This article belongs to a three-part series written for anyone that is looking to understand why experts are expressing concern about rising COVID-19 hospitalizations. Part one explains what the number of hospitalized patients actually means, part two explains why hospital bed distribution is important, and part three covers the nuances of hospital and ICU occupancy.
On September 20th, U.S. Covid-19 hospitalizations reached a nadir not seen since mid-June.
Just one month later, the trend is moving in the opposite direction at an alarming rate, threatening to overwhelm resources in communities with limited hospital capacity.
Visualizing data at the national level is helpful to identify national trends, but mobilizing resources requires far more granular data (and a lot more context). A number of people involved in the Covid-19 response effort at the federal level recognized these data gaps as early as March, and had taken concrete steps to address the issue by the time I was looped into a related meeting in late April. The meeting was an opportunity to brief the data architects responsible for building the Covid-19 Data Hub on the technical details of the Covid-19 testing site dataset produced by our volunteer effort. During that meeting, the data hub project leads shared with us their broader vision, which included near-real time hospital capacity and even weekly snapshots of PPE inventory.
We know from public reporting that the initial phase of the HHS Covid-19 Data Hub project reached a milestone in late July, but key details remain opaque and the public still lacks access to key details that are necessary to fully grasp the extent to which the risk of Covid-19 transmission did/does/will shape our day-to-day lives.
The next few sections offer a (mostly) jargon-free break down of hospitalization data that highlights what we know, and what we don’t know — and explains why it’s important to avoid drawing conclusions or drafting policy based solely on the data that is currently made available to the public. The goal is to teach non-experts how to read and interpret hospitalization data, and to illustrate the value of increased data transparency.
Looking at raw data: how health data experts think about the number of Covid-19 hospitalizations
The first thing to know about hospitalization data is that the raw number — that is, the number of patients hospitalized with Covid-19 — is not a meaningful measure on its own — it needs proper context. That’s because the risk of being hospitalized for Covid-19 can vary due to any number of factors, including some that we may not yet recognize. Age is a well-known factor:
This chart tells us that if you were to take two regions of the country with comparable-sized populations, similar rates of new cases, and similar % test positivity — but one region was comprised entirely of 18–29 year olds and the other was entirely 75–84 year olds, we could expect the number of hospitalizations in the older group to be 8X higher. The real world doesn’t work that way, but it does help to explain why Covid-19 hotspots in Tampa, Florida are more likely to increase hospitalizations than a hotspot in Miami. It also underscores why we can’t just look at the raw number of hospitalizations.
The number of patients hospitalized with Covid-19 is not a meaningful measure on its own — it needs context.
Based on the information we’ve covered thus far, we can expect the number of hospitalizations to increase (or decrease) as confirmed Covid-19 cases rise and fall in the general community — as illustrated here using data for the state of Florida. This chart also illustrates that the number of patients hospitalized tends to lag the onset of symptoms by 8 to 12 days. This explains why the curve for hospitalizations (bottom) appears to be slightly shifted to the right as compared to the curve for new cases (top).
But each point on the curve is a just a snapshot in time — it doesn’t tell us anything about the number of days a specific person has been in the hospital. The best available data we have to measure the typical length of stay (LOS) for persons hospitalized with Covid-19 was collected very early in the pandemic; it tells us to expect a Covid-19 patient to spend 10 to 13 days in the hospital. But the treatment of Covid-19 has evolved over time.
Assuming that this evolution translates to clinically meaningful improvements, then we can expect to see measurable improvements in hospital LOS (i.e., a statistically significant decrease) — and we do. There are two treatments that are proven to be helpful in managing Covid-19 in hospitalized patients: remdesivir and dexamethasone.
- Remdesivir has been shown to reduce the typical (median) hospital LOS from 15 to 10 days.
- Dexamethasone was shown to decreased the amount of time Covid-19 patients on mechanical ventilators remain intubated. There was also a slight reduction in hospital LOS from 12 to 13 days.
Using the state of Colorado as example, we can see that hospital LOS decreased between March and July; as did ventilator use.
If the way doctors manage hospitalized Covid-19 patients has improved over time, then the data should reflect that — and it does.
Key takeaways
When looking at the number of Covid-19 hospitalizations over time, bear in mind that:
- Daily hospitalization numbers are a summary snapshot for the day — the value doesn’t say much about the patients themselves
- The way doctors treat hospitalized Covid-19 patients has improved, and that’s translated into shorter hospital stays.
- Assuming nothing about the disease or the patient population has changed, then the number of patients hospitalized with Covid-19 on any given day should be lower today, than it was over the summer.
Another way to say it
Saying that you have 100 Covid-19 patients in a hospital today is (roughly) the same as saying that you had 70–90 patients back in March.
