Somebody or multiple people at the Massachusetts Department of Public Health have way too much time on their hands. The Commonwealth changed the reporting structure of the daily dashboard once again. Quite frankly, I think the state should focus more on developing a strategy for getting people vaccinated and focus less on changing the dashboard report structure. The files that I used for my analysis have changed as well – I will now work on seeing if all the data I had used for my data updates is still available, and adapt accordingly. Stay tuned.
This is the second post comparing Massachusetts’ Covid statistics to other states, but with a focus on cases. I’m using the daily historical data from the Covid Tracking Project (CTP) to perform calculations, but used the New York Times and Worldometers databases to check figures from the CTP.
|Table 1: Largest Case Discrepancies Among Data Aggregators|
|Data through October 28, 2020|
|2nd Largest||New York||4.6%||Wyoming||10.0%||Wyoming||10.0%|
Table 1 summarizes the discrepancies between the three data aggregators for case data through October 28th. The entries in the tables are the percentage difference in the case totals to date for each pair of aggregators. As with deaths, the best agreement was between the New York Times and Worldometers, and the CTP differed more from the other two aggregators.
In the first post in this series, I defined three phases to the pandemic (Spring – ending May 28th, Summer – ending September 12th, and Fall – ongoing) based on peaks and troughs in national case data from the New York Times. Table 2 shows the case rates and rank for Massachusetts as well as the top-ranked and bottom-ranked states (including Washington D.C.) for each phase of the pandemic. (The W in the Massachusetts rankings stands for Worst).
|Table 2: Case Rates per 100000 and Rank By Pandemic Phase|
|March 1 through October 28, 2020|
|MA||34th W||2,184||4th W||1,377||46th W||430||40th W||378|
Table 2 clearly shows how the geographic concentration of the pandemic has shifted over time. The Northeast had the most per capita cases in the spring; the Southeast and Southwest in the summer; and parts of the Upper Midwest, Rocky Mountain, and Great Plains states in the fall. In fact, the Northeastern states generally had the lowest case rates in the summer and fall – only Hawaii in the fall broke the Northeast’s monopoly on the lowest five case rates. Massachusetts fits this pattern as well, even if its case rate is not among the five best.
The table also indicates how per capita case rates have increased after the spring, even though death rates were much higher in that period. The spring 7 day average national peak was almost 32,000 cases per day, compared to almost 67,000 cases in the summer and 78,000 cases now. Some of this increase is because more testing is being done now than in the spring, resulting in more cases being diagnosed.
In Massachusetts, for example, the average daily number of tests more than doubled from the spring (starting on March 15th) to the summer, and then almost tripled from the summer into fall. However, much of that summer to fall increase has been driven by higher education testing, which now accounts for about half of all testing in the state. Nonetheless, the increase is significant.
These statistics, when compared to the death rate findings in the prior post, point out one of the central puzzles with covid in Massachusetts. Massachusetts is among the states with the lowest number of per capita cases, especially after the spring, but among the states with the highest number of per capita deaths overall, and higher than one would expect in either the summer or fall.
Why? The obvious answer is the case fatality rate (CFR) – the percentage of cases that lead to death. To look at CFRs across states, I compared case rates through the end of the spring and summer phases to deaths rates through the end of the corresponding trough in deaths. For example, I looked at case totals through May 28th in each state, and compared them with deaths through July 5th (from Table 1 in the prior post that defines the phases). Implicitly, I’m assuming about a five week time lag between diagnosis and death. Since the fall phase is still underway, I only performed this analysis for the spring and summer.
Because I’m using about a five week time lag, I may be overstating the CFR for the two phases. Nonetheless, this exercise does provide insight into the Massachusetts conundrum. Table 3 summarizes the results. Death data is only through October 16, as that is the trough in deaths from the summer phase.
|Table 3: Case Fatality Rates and Rank By Pandemic Phase|
|March 1 through October 16, 2020|
|MA||3rd W|| 7.8%
||7th W||8.6%||1st W||5.1%|
The use of a five week lag in deaths does seem to jumble some of the reporting. For example, Arizona has the second highest CFR during the spring. This is almost certainly because some deaths from the summer phase are paired with cases from the spring phase when Arizona had a relatively low case rate.
The table shows the dramatic reduction in CFRs from the spring to the summer. Outside of the Massachusetts outlier, the worst CFRs in the summer aren’t much higher than the best CFRs from the spring. There are likely several reasons for this. As noted before, increased testing has presumably led to the identification of less ill patients. Second, the average age of people diagnosed with Covid has decreased. Finally treatment protocols have gotten better (https://www.nytimes.com/2020/10/29/health/Covid-survival-rates.html)
However, a very surprising result is the out-sized Massachusetts’ CFR in the summer, significantly higher than any other state. Not all of this can be explained by the death lag improperly assigning cases and deaths to the wrong phase. Even as Massachusetts’ relative case rate has come down, its death rate has stayed relatively high. Is this the long-term care issue again?
To check this, I performed the same hypothetical as in the prior post. What if the percentage of deaths from long-term care facilities in Massachusetts were 40% instead of 70%? In that case, Massachusetts’ overall CFR for the spring and summer would have dropped to 4.6%, ranking it 8th highest overall instead of 3rd. Similarly, the state would have improved only to the 5th worst CFR in the summer period. Some improvement, but the state is still a laggard. There is something else going on.
I’ve focused most of my attention on tracking the internal dynamics of the coronavirus in Massachusetts, with only one previous post with a comparison to other states, (https://www.masscoronavirus.net/massachusetts-isnt-as-great-as-it-thinks-it-is/) back at the end of August. But one of the things that has stood out is Massachusetts’ very high death rate relative to other states. It has either the second or third highest (more likely 3rd, more on this later) per capita death rate among states over the course of the pandemic, but a high percentage of deaths in Massachusetts were during the spring. As a consequence, I wanted to analyze how well Massachusetts has been handling the pandemic relative to other states since the spring – if possible examining cases, deaths, testing and test positivity.
I used data from the Covid Tracking Project (CTP) (https://covidtracking.com/data) as my primary source, because it is easy to download daily historical state-by-state information from their website. However, I checked the CTP’s overall case and death statistics against the New York Times (https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html?action=click&module=Top%20Stories&pgtype=Homepage) and Worldometers (https://www.worldometers.info/coronavirus/country/us/) databases. All data is as of October 28.
I defined three phases of the pandemic to date using the New York Times calculation of the 7-day average of the national total of cases, looking for troughs in the data to define the end of each phase of the pandemic. Table 1 summarizes peaks and troughs for both cases and deaths.
|Table 1: Pandemic Phases Based on Data Peaks and Troughs|
|March 1 through October 28, 2020|
|Peak and Trough Dates|
For example, based on national case data, the spring phase of the pandemic lasted from March 1st through May 28th, because May 28th is the day when total cases fell to a minimum after peaking on April 10th (they began increasing again on May 29th). Based on national death data, the spring phase would have ended on July 5th.
There are some interesting tidbits in Table 1. First, the troughs in cases coincide with a few days lag to the holidays that bookend the summer – Memorial Day and Labor Day. This does not necessarily imply that increased social activities on those holidays led to upswings in cases, but they certainly may have played a part.
Second, the death peaks for the spring and summer phases occur roughly one week after the case peaks, but the death troughs occur roughly five weeks after the case troughs. I would not read too much into this, as the national statistics are just the sum of state statistics, each of which has its own ebb and flow of cases and deaths. In addition, this is based on a sample of only two phases for the pandemic (and let’s hope there are not many more). However, the lag between the case and death troughs correspond more naturally to our understanding of how covid progresses from diagnosis to death.
Based on the peaks and trough data in Table 1, I defined the pandemic phases using troughs based on cases, even when I analyze death data. This is primarily because the current (Fall) phase based on death data has had too few days to draw any meaningful conclusions (twelve days so far)
A few caveats about the data are in order. First, the CTP generally obtains its data from state agencies charged with compiling and publishing Covid information for that state. To the extent that different states have different procedures for counting cases and deaths, the data published by Covid Tracking is not uniform across jurisdictions.
Second, the CTP (and Worldometers, for that matter) are not particularly adept at handling data restatements from the states. For example, when Massachusetts dropped roughly 8,000 probable cases and 26 deaths from its tally in early September, both sources just showed one day drops in case and death totals reflecting those changes. In other words, neither data aggregator restated the data historically, so that it appeared that there were a large number of negative cases in Massachusetts on September 2nd. This means that although the total number of cases and deaths are accurate, they are improperly distributed into the three pandemic phases.
Finally, although agreement among the three data aggregators is generally good, there are some large discrepancies in aggregate death and case statistics that call into question some of the results from the Covid Tracking Project.
Table 2, a comparison of the largest differences between the cumulative death rates among the three data sources, illustrates these issues. The entries in the tables are the percentage difference in the death rate to date for each pair of aggregators. As is evident from the table, the agreement was best between the New York Times and Worldometers, and second best between the CTP and Worldometers.
|Table 2: Largest Death Discrepancies Among Data Sources|
|March 1 through October 28, 2020|
|Largest||Alaska||5.6%||North Dakota||30.4%||New York||30.5%|
|2nd Largest||Kentucky||4.6%||New York||28.5%||North Dakota||29.1%|
Table 2 points out particular issues for the Covid Tracking Project data for New York and North Dakota, where the agreement between the New York Times and Worldometers is close. The CTP shows 25,773 deaths for New York, compared to over 33,000 for the other two aggregators. The CTP data aligns with the NY State website, which excludes probable cases from its totals. This implies that the CTP is under counting deaths in New York. (Is Cuomo fudging the stats?)
For less clear reasons, The CTP is also under counting deaths for North Dakota relative to Worldometers and the New York Times. Death totals from these two sources align much more closely with the North Dakota state website.
These data discrepancies have implications for the results shown in Table 3, which show the death rates and rank for Massachusetts as well as the top-ranked and bottom-ranked states (including Washington D.C.) for each phase of the pandemic.
|Table 3: Death Rates per 100000 and Rank By Pandemic Phase|
|March 1 through October 28, 2020|
Based on the CTP data, Massachusetts has the second highest per capita death rate from the pandemic in total. However, Massachusetts would drop to 3rd place if New York State included probable cases. Even so, Massachusetts has had a very high relative rate of deaths in each phase of the pandemic, only dropping to 9th worst during the summer, and to 21st worst for the fall pandemic to date. Not an enviable record.
This table makes concrete how the pandemic has affected different areas of the country in the three phases to date. As it well known, the first and deadliest phase in the spring affected the Northeast most heavily – of the eleven states (and D.C.) with the highest death rates, all but Michigan and Louisiana were in New England or the Mid-Atlantic. Only the more rural states in New England were spared.
The summer phase impacted the sunbelt most heavily, with eight of the ten states with the highest death rates in the Southeast or Southwest. But Massachusetts and Rhode Island were in the top ten as well. Finally, the current phase is more of a mixed bag, although the states with the highest death rates are generally more rural. There is less clarity about this phase to date.
Overall, the states with the lowest death rates overall are sparsely populated, or geographically isolated (Hawaii), or both. This is true for the ten states with the lowest death rates overall, except for Oregon and Washington. While there have been some changes in the ranking of these low death rate states during different phases or the pandemic, for the most part this has been the case.
Why has Massachusetts fared so poorly? An obvious thought is that this is from the high percentage of deaths in long-term care facilities in the state. Almost 70% of Massachusetts deaths have been in long-term care facilities, compared to about 40% nationwide.
However, this is not the entire answer. As a hypothetical, for each phase in the pandemic and in total, I adjusted Massachusetts’ deaths so that 40% instead of 70% were in long-term care facilities. What would have changed? In overall death rates, the state would have dropped from 2nd (or 3rd) overall to 9th, from 4th to 7th during the spring phase, from 9th to 28th during the summer phase, and from 21st to 30rd during the fall phase. Improved, but not stellar. Because death rates during the spring were so high compared to later death rates, those early deaths dominate the overall results, even adjusting for long-term care deaths.
For everyone dealing with the coronavirus, there is both a societal and personal calculation. The societal calculation revolves around the enormous global and national costs of the pandemic – the staggering number of illnesses, hospitalization, and deaths, as well as the economic toll – massive unemployment, shuttered businesses, and food insecurity.
But there is a personal calculation with which most people wrestle. How likely am I or the people close to me to get sick; and if they get sick, how sick will they be? What are the odds that they will be hospitalized, or obviously even worse, how likely are they to die? This calculation clearly is highly dependent on personal circumstance – age, type of work, underlying health conditions, etc. But a starting point for understanding this is the number of people hospitalized from the coronavirus. (Obviously, this is an incomplete measure of severe illness, as many ill people in long-term care are never hospitalized regardless of how severely ill they become). And here, the available information in Massachusetts is confusing.
The Commonwealth has published a running total of the number of confirmed and suspected Covid-19 cases, total hospitalizations, and total deaths through time. (According to the Dashboard, this information comes from the Bureau of Infectious Disease and Laboratory Sciences). However, when the state changed the definition of probable cases earlier this month, they restated the cumulative number of hospitalizations without providing the details of the historical revisions – unlike what they did for cases and deaths. The number of confirmed and suspected hospitalizations dropped from 13,386 on September 1 to 13,295 the next day.
Fortunately, there is another source of hospitalization data – that provided by hospitals themselves and submitted to the state Department of Public Heath and federal government. Hospitals report both the number of patients currently hospitalized for Covid, and the number of new hospitalizations. Unfortunately, these data on new hospitalizations do not track the data collected by the state – in fact, the number of new probable case hospitalizations reported don’t make much sense if taken at face value.
For the week ending September 3, hospitals in Massachusetts reported an average of 312 patients hospitalized with Covid, an average of 19 new confirmed case admissions, and an average of 126 suspected case admissions, for a total of 145 new admissions. These statistics do not square with what we know about the hospital stays of Covid patients. Since the only way patients leave the hospital is if they are discharged or die, this would imply an average hospital stay of roughly two days, much shorter than what one would expect.
It is unclear exactly what these suspected hospital admissions are tracking, but the definition seems overly broad. According to the dashboard, these suspected cases “are those with symptoms who have not had a test result yet”. Perhaps many of these originally suspected cases turn out to not be Covid patients at all, or there is something else not transparent about this reporting.
However, the number of confirmed hospital admissions does closely track the number of newly hospitalized cases reported by the state up to the point at which the probable case definition was changed, as shown below. Both figures also reinforce the idea that the state has been in rough equilibrium for about the past five or six weeks (this is true for cases, hospitalizations, and deaths), with relatively low case and hospitalization rates.
Massachusetts has done a great job in bringing the coronavirus under control since the early, dark days of Covid-19, when it looked like hospitals in the state would be overrun like those in New York City. And relative to many states in the country, our statistics look pretty good. However, if we compare ourselves to peer states in New England and the Northeast, we’re at the back of the pack, as the following table shows. I’ve defined the peer states as the states with which Massachusetts shares a border, as well as Maine and New Jersey. The data is from Johns Hopkins’ excellent coronavirus tracking site, with population estimates taken from worldometers.info, which also tracks covid statistics.
Johns Hopkins Testing and Death Statistics
Massachusetts and Peer States
As of August 29, 2020
|Deaths Per 100,000 Residents||7 Day Positivity Rate||Testing Rate (per 100)|
|State||Last Week||Last Month||Last Week||Last Month||Last Week||Last Month|
I’ve pulled together three statistics: death rates, testing positivity rates, and overall testing rates; and show each for the past week and past month, as per Hopkins. The states are sorted from worst to best death rates over the past week. Massachusetts continues to have a stubbornly high death rate, with only Rhode Island coming close. The other peer states have significantly lower rates – including New York and New Jersey, which were much more heavily ravaged at the beginning of the outbreak.
Why is this? It is difficult to know precisely, but about 70% of our deaths are coming from people in long-term care settings. I don’t know the percentages in other states, but it is surprising that the high death rate in those facilities continues months after the peak.
We’re also at the back of the pack when it comes to test positivity rate, although the gap between Massachusetts and our peer states is much lower. It may be harder to control the virus in a more urban state, which would explain why the positivity rates in Rhode Island and New Jersey are also relatively high, and Vermont and Maine have the best recent record here. But New York and Connecticut are also urban, and have much better statistics than we do. (New Jersey, Rhode Island, Massachusetts and Connecticut are the mostly densely populated states in the country in that order, and New York is the 7th most densely populated).
In terms of testing, we’re squarely in the middle of the group. Rhode Island leads the testing Olympics, perhaps because CVS is headquartered there (over the entire pandemic it is the top state in the country for testing per capita), followed by New York, and Connecticut. We’re doing relatively well, but not terrifically so. We started out as one of the better testing states, then fell back for some time. We’re now ranked 13th overall in the country per capita, but for a period of time we were testing less than the national average. Testing has picked up again, a positive sign.
This is the inaugural post for this site dedicated to reviewing the state of coronavirus in the Commonwealth of Massachusetts. I had been thinking about a blog for quite some time, but had been satisfied with occasional comments on Boston Globe articles about the coronavirus – in particular, the Globe’s daily article which focuses on the latest statistics published at roughly 4 pm each day by the Commonwealth in its Covid-19 Dashboard. That daily article seemed generally to just parrot the headline numbers released by the Commonwealth, without any deeper analysis.
I have felt increasingly constrained by the Globe’s comment section. First, although I’m interested in a daily recap of the statistics, I thought that would become quite repetitive and boring for most Globe readers. However, I realize that there is a subset of readers who are interested in that, as well as other analysis of what is happening with testing, cases, hospitalizations, and (unfortunately) deaths. So this format gives me freedom to write about what is interesting to me (and hopefully others), and to go somewhat beyond the headline numbers. Second, the Globe comment section has been increasingly taken over by individuals with a particular axe to grind (I hesitate to call them trolls), with whom I’m tired of dealing. No need to name names, but anyone who finds their way here from there will know to whom I’m referring.
The purpose of this blog is to provide relatively untainted commentary about the coronavirus in Massachusetts (I use the word relatively because I understand that I, like everyone else, have particular biases that influence what I write about and how I write about it). I have been relatively impressed with the response to the coronavirus in Massachusetts after some initial missteps (and those might not have been avoidable), but a bit dismayed with recent changes in the Dashboard, which have reduced the amount and variety of information published. The Commonwealth had done an excellent job of providing information sliced and diced in various ways. It has backed off of that, for reasons I’m unaware of.
In any case, off we go.