World Health Organizations warns that re-infection cannot be ruled out (Bloomberg)
New study shows promising anti-viral drug remdesivir is not effective against COVID-19 (Stat)
New study shows that as the case definition for COVID-19 changed in China, more cases were detected, highlighting importance of broadening the clinical indicators that should be considered (Lancet)
Experts emphasize that COVID-19 will be with us for months (Axios)
Low agreement on U.S. Death totals Where do you look for information on the number of COVID-19 deaths in the U.S. and elsewhere? It should be straight forward to count deaths, even as we find increasingly that determining the fraction of Americans who are infected is a genuine hornet’s nest. According to WORLDOMETER, the U.S. has now experienced more than 52,843 deaths from COVID-19 as of noon today. Are we done? Not so fast. As the numbers grow, we increasingly see variation in the total death toll depending on where you look. The table below summarizes that variability as of noon on Saturday. What does this mean? This far into the epidemic, one might think that these discrepancies would be shrinking as we gain more experience, but that is not what we see. The average for these estimates is 47,500, but the standard error is a whopping 8,200 deaths! The high and low estimates disagree by 24,000. This means a couple of things: a) different reporting systems are not all using the same sources, b) CDC sites that use death certificates are so severely lagged that we shouldn’t be paying attention to them; c) whether or not a given reporting platform includes ‘probable’ deaths explains a big part of this discrepancy. It’s often difficult to tell what each site is doing with probable deaths. This is not a small issue. The New York Times data excludes over 5,000 probable deaths, and would be much more in alignment with the Johns Hopkins and Worldometer estimates if they were added. I believe that probable deaths should be included, because most probable deaths are people who died in hospitals (or outside hospitals) that had COVID-19 but weren’t tested or the test results were not available at time of death. Given the inadequacy of testing, particularly in hard-hit areas, it’s likely that the vast majority of “probable” deaths were COVID-19 infections. Add to this a false-negative testing rate of up to 30%, and the case for inclusion gets stronger. From a surveillance point of view, I am much more worried about undercounting deaths than over counting them.
Comparison of death totals across data sources as of 4/25/20
Is COVID-19 worse than the seasonal flu? And other mass casualty events? Many people are still wrestling with the question of whether this disease is worth the hit our economy is taking. Isn’t this just like the flu? As states move toward re-opening, big questions emerge about whether we are over-reacting. All of this is understandable. But as Michael Osterholm recently said, we are in the 2nd inning of a baseball game. Take a good look at this new graph I spent the day making. I think it’s pretty impressive. This puts the COVID-19 death toll into a broader perspective. It says that by March 2, we had passed total U.S. deaths for Ebola. We passed SARS deaths around March 7. By March 28, we had exceeded the average deaths from influenza in March and April over the last 5 years. By April 6, we had more deaths than for all flu deaths in March and April for 5 years. By April 7, more people had died of COVID-19 in 5 weeks than all who died of 2009 Swine flu. And by April 21, COVID-19 had killed more Americans than died of influenza over the last 5 years combined. For further context, more us us died by April 10 than died in Ebola, SARS, Los Vegas shootings, the Gulf War, Hurricanes Andrew and Katrina, Swine flu, and the September 11 attacks. Combined. Bottom line: This is not like the flu
U.S. Deaths trending flat on the log scale. Peak approaches, but has not arrived. As I said on Wednesday, it is important that we look at our data the right way at the right time. We are used to seeing daily the solemn record of news deaths and cases. The signal we are waiting for is a switch (or inflection point) between exponential growth and linear growth. That’s the peak that matters. It can be difficult to see that transition point in the usual epidemic curve. Below are two ways to look at the same data (cumulative total deaths by day). The top (purple) graph shows the counts on the linear scale. The bottom (orange) shows the same data on a logarithmic scale. Check the actual numbers to verify it’s the same data. Why? A vital clue about where we are in the epidemic is is easier to see on the log scale, because when the switch occurs and exponential growth yields to linear growth, the curve will look flat or plateaued. It never perfectly flattens, but linear growth looks flat-ish. The orange graph shows that we are not there yet, but we are getting closer.
U.S. Deaths decline Thursday. Hot spots in Minnesota, Alabama, West Virginia and Pennsylvania On Thursday, an additional 1,840 Americans lost their lives from lab-confirmed COVID-19. This represents a 4% rise, but is lower than the 2,119 on Wednesday. California (115), Colorado (44), Nevada (23) and Minnesota (21) set or matched the previous high death toll. Despite these trends, Colorado and Nevada plan to reopen soon. Deaths have declined in absolute numbers for 3 days, however, part of yesterday’s decline comes from the removal of 201 deaths in Pennsylvania that were reclassified as “probable” deaths requiring further investigation. This highlights the need to interpret daily totals with caution. A more robust metric is the 7-day moving average, which suggests that deaths have been flat over the last week after surging on Tuesday. The graph below highlights hot spots in Minnesota (+112%), Nebraska (+94%), Alabama (+139%), North Carolina (+93%), West Virginia (+133%), Massachussetts (+90%), and Pennsylvania (+101%). All 7 of these states have approximately doubled their cumulative death tolls this week.
Deaths and new cases remain flat. Reopening states Georgia, South Carolina, Florida, Tennessee, all saw 30% rise in new cases last week On Wednesday, 29,743 new COVID-19 cases were reported, a 4% rise in the cumulative total. Another 2,146 American’s lost their lives to the virus, a 5% rise in the total death toll. Trend lines based on 7-day moving averages show that both deaths and new cases are flat (Graphs 1 and 2). Seven states matched or set new daily high deaths including Arizona (21), Minnesota (19), Missouri (19), Alabama (52), Washington DC (15), Massachussetts (221), and New Hampshire (6). Regionally (Graphs 3 and 4), cases are surging in the Midwest, with greater than 50% 7-day growth seen in Iowa, Minnesota, North Dakota, Nebraska, Ohio, and South Dakota. New case hot spots also seen in the South (Mississippi, and Virginia) and the Northeast (District of Columbia, Delaware, Maryland and Rhode Island). However, as deaths lag behind cases, the mortality hot spots differ, including Minnesota, Alabama, North Carolina, Virginia, West Virginia, Connecticut, Massachussetts, Pennsylvania and Rhode Island. It is possible that deaths will rise next week in midwestern states after rising cases this week. Several states move to reopen businesses according to a report in the Washington Post including Georgia, South Carolina, Florida and Tennessee. All four of these states have seen 30% jumps in new cases over the last week (Graph 3). Deaths rose over 50% last week in Florida and Georgia and by a third in South Carolina and Tennessee. Experts are unified in the belief that the reopening of these states will lead to sharp increases in cases and deaths.
A sharper picture of prevalence and CFR begins to take shape The big news today, covered in today’s Top Pick, is the release of results of a seroprevalence study of 3,000 people in New York. The results, while preliminary, can now be added to the growing number of studies trying to get to a better estimate of the total fraction of the population infected and what the true case fatality rate is. I have repeatedly emphasized how vital these two pieces of information are. I have also stressed that all studies have limitations and that no single study can answer big questions alone. Instead, as disease detectives, our strategy is to triangulate across studies, considering the strengths and weaknesses of each, paying close attention to the overall patterns. So, let’s look at the studies that have so far come on to our radar screen (see table below). There are more studies than this, but let’s keep it simple. The top row of this table is the new study today in New York. It suggests that 14% of New Yorkers may have been infected, which means as many as 2.7 million cases, a 10-fold higher number than current testing now suggests (257,000+). The New York prevalence is quite similar to the much earlier German study. New York estimates a CFR of 0.5%, five times higher than seasonal flu and higher than the rest of the estimates. The two California studies estimate a prevalence of 3-4%, which is dramatically lower than New York. What this means? We need to squint at this table. All these studies have limitations. We don’t focus on any one study, but instead look for patterns. Overall, these studies are telling us that the prevalence of COVID-19 infection is orders of magnitude greater than current testing indicates, somewhere between 10 and 50 times higher. It also says the CFR is in the neighborhood of 0.2% and 0.5%. This number is thankfully well below the crude death ratios suggest, but indicate that COVID-19 will take substantially more lives than seasonal flu.