Western Massachusetts counties now have their highest numbers of new Covid-19 cases since the spring (And why are you learning about that here?)

The article below was written with data current as of November 13th 2020. The New York Times provides the latest county new case rates, and an interactive county-level new case graph is now also available on their website. See this page for up-to-date information on Northampton (note esp. that unlike the spring most cases are in the community rather than in long-term health care – I missed that when I wrote the article).

A Nov. 21 Hampshire Gazette article covers this work and provides guidance and further information from Joanne Levin of Cooley Dickinson Hospital – see also Karen Foster’s Ward 2 Newsletter for more from Levin. I have also written this related article with a warning about changes to the MassDPH color coding scheme, and an illustration of what Hampshire looks like using Colorado’s, and co-authored a paper on the influence of higher ed testing on local numbers.

=============

On November 10th Governor Baker gave a Covid-19 update, and announced a statewide 300% increase in new cases since Labor Day. This announcement came in a discussion of the state’s new public health measures, which include a face covering requirement in all public areas regardless of distance between people, a stay at home advisory from 10 pm to 5 am, and tighter restrictions on size of gatherings. Here in western Mass, we have had our own surge in this same timeframe, and now have higher numbers of new cases in all four counties than at any time since the spring. In all but Hampden County, the levels remain relatively low compared to the rest of the state, and to the rest of the country (relatively low compared to another place does not imply “safe”, of course). Hampden now has one of the highest per capita new case rates in the state, a rate that is even higher than the United States as a whole.

After presenting current and historical county level data from western Mass, with some comparisons with statewide and nationwide numbers, I turn to more local data from towns and cities, and higher education institutions. In that context I discuss the deficiencies in local COVID-19 data reporting, which seem characteristic of the state as a whole, and probably the entire country. This is one of many governmental failures in managing this crisis, and one that should, and probably can be fixed quickly as we enter a particularly dangerous phase, and hopefully get better leadership at the federal level.

I have been making graphs like the one below for Hampshire County since the summer, when I was trying to monitor the local COVID-19 situation in preparation for whatever decisions we would have to make for the school year for our child. I couldn’t find longitudinal new case rates for the county elsewhere, so I made them myself with raw data downloaded from the Massachusetts Department of Health website; see the section at the end for the details, and for other data details that aren’t in the body of this article. The graph shows the seven day total of new cases for the week ending at the date shown on the horizontal axis. For the most recent week ending Nov. 12th, there were 181 cases, the highest since the week ending April 29th, in which there were 187. In the week ending September 14th, only 7 cases were reported, the lowest number since March 24th, in the first week of reported cases.

It might be a mistake to conclude from these figures that the prevalence of COVID-19 in Hampshire County is now as high as it was in April, since testing has ramped up considerably since then. “Cases” refers to cases that are diagnosed by a positive molecular test (although the data before August 12th also contain probable cases, which make up about 10% of the total). When all else is equal, more testing will yield more cases given a fixed prevalence in a population, so given the same number of cases with more testing, the underlying prevalence is likely lower. There do not appear to be any publicly available county level data on the number of tests performed in the spring, but statewide there was a daily average of 12,769 tests performed in the week ending April 29th, and 77,630 for the week ending Nov. 12th. There is another difference between these time periods though: tests in the earlier period were generally performed only on symptomatic individuals, and there was much more asymptomatic testing in the later period. So it is possible that there is a similar prevalence in the two periods. (See this article for a more detailed discussion of comparison with the spring, with a local focus on Lewisburg Pennsylvania).

It seems unlikely that the change in the last two months is due to an increase in testing. In the two weeks ending September 16th, 32742 tests were performed in Hampshire county, with 38 positive, for a positivity rate of 0.12%. In the two weeks ending November 11th, 41751 tests were performed, with 262 positive, for a positivity rate of 0.63%. Though there was an increase in testing, it doesn’t match the steepness of the increase in cases, and the positivity rate has increased more than 5 times. (Update: it’s up to 0.92% for the two weeks ending Nov. 18; note that this includes UMass numbers, which are based on a relatively high proportion of asymptomatic testing – the statewide positivity numbers for higher ed are much lower than those from the rest of the population, as can be seen in the demographics section of the state dashboard).

One indication that the situation statewide is less dire than in the spring is that although COVID-19 hospitalizations have about doubled since September, they remain far lower than in the spring. There was a daily average of 3,868 in the week ending April 29th, which was down to 328 in the week ending September 14th, and is now up to 592 in the week ending November 12th. The current situation in Massachusetts is also better than the country as a whole, which this week hit an all-time record of hospitalizations. There seems to be no publicly available county level hospitalization data, but the numbers of deaths in Western Mass seem to have a similar pattern, though the increase since September is smaller. Hampshire had 20 in the two weeks ending April 29th, none in the two weeks ending September 14th, and 3 in the two weeks ending November 12th (I’m using a two week window since the numbers are small). Hampden had 205, 21, and 28 in the same time periods, Franklin 10, 2, and 2, and Berkshire 11,1, and 0. The Massachusetts daily averages of deaths in the weeks ending at those dates were 172, 12, and 23 respectively.

The current lower number of hospitalizations and deaths than in the spring might seem to point to a lower prevalence of the disease even when the new case count is similar across times, but there are other (non-exclusive) explanations: the cases are now more often younger people who are less likely to develop symptoms, to be hospitalized, and to die, and treatment is improving.

To compare across counties, and to compare our local figures with those of other places, we need to relativize the numbers to population. The convention followed by the New York Times and by the MassDPH is to present a daily average over 7 or 14 days per 100,000 people. This means dividing the weekly count by 7, multiplying it by 100,000 and dividing by the population. With Hampshire County’s weekly count of 181 and population of 160,830 this yields 16.1, which matches the currently displayed rate on the New York Times website. Brendan O’Connor of the College of College of Information and Computer Sciences at UMass Amherst provides a daily updated graph of longitudinal new case rates for the four Western Mass counties, as well as the state as whole, and the country. Today’s graph is shown below. Here the rates are presented per million, so Hampshire is at 161.

As you can see by comparing the endpoint of each of the solid lines leftwards – backwards through time – the current new case numbers in all 4 counties are the highest they have been in months. You can also see that there are some big differences amongst the counties. Hampden’s current new case rate is the only one that is higher than the state as a whole – it even exceeds that of the United States, which itself is at a record high. Further comparisons can be made with the current rates beneath the graph. For example, Canada’s new case rate is 111 per million, 28% of that of the United States but still alarmingly high, while Europe (i.e. the EU and the UK) has an even higher rate than the US.

Why are new cases going up? Governor Baker’s explanation is that we are “dropping our guard” because we are experiencing “Covid fatigue”. One might also look for explanations in changes in the extent of contact between people due to being indoors more as the weather gets colder, or in changes brought about by various aspects of reopening, including the beginning of the school year and the wider opening of various kinds of businesses. Maybe there is even a new more transmissible strain of coronavirus now in circulation. Here I will focus on what we can find out by looking at more fine-grained local data: data from the towns and cities, and institutions, that make up the counties. Massachusetts appears to have relatively good local data reporting compared with the rest of the country. However, it is nowhere near as good as it could, and should be, to help guide and protect its citizens. I am focusing on Hampshire County, whose data I am most familiar with since that is where I live and work, and where my child is enrolled in the public school system. I suspect that its local data reporting is the best of the four counties, and that there is probably an even more urgent need for improvement elsewhere in western Mass.

The first problem is that it’s not at all obvious how you find the more fine-grained local data. It’s not even that obvious how you find the most up-to-date county data. Perhaps surprisingly, the best place for the county data is usually not the Massachusetts Department of Public Health’s website, which is where you can get the most detailed current statewide data, but which updates local data only weekly (the latest daily county data can be downloaded in raw spreadsheet form, the source for the graphs above and the New York Times website). We have no county board of health, and the town and city boards of health display very little data, if any, on their websites. Rather, you need to go to the New York Times’ website, which has an excellent presentation of state and county data, based on numbers it gets from the Mass DPH. It does not, however, present the county data over time, and I have never seen graphs like those above for Western Mass outside of those produced by me and my UMass colleagues from raw data downloaded from the MassDPH or the New York Times. This is the likely answer to my question in the title of this article: no one told you about the historically high local new current rates because the right kind of data visualization in which to see them isn’t being widely disseminated. And the county testing rates, which are as discussed above are important for interpreting the new case data, are only released once a week on the MassDPH website, and do not seem to be available at all for the spring.

I’ll stop complaining for a minute, and cut to the chase about what we know about the local rise in cases, which is due to an instance of exceptionally good local data reporting. The UMass Amherst COVID-19 Dashboard presents daily counts of new cases and a rolling 7 day daily average, from Aug. 6th to the present. For the week up to the most recent date with tests returned, Nov. 10, there were 90 cases. Thus, about half of the new positive cases in the last week in Hampshire County are UMass Amherst staff and students. The number of tests across time are also presented on the same page, and we can see that the recent increase in new cases is not due to increased testing. We can also see that the spike in cases in early October in Hampshire County is largely, if not fully, due to a spike in UMass Amherst cases. (Update 11/19: there were 76 UMass cases in the week ending 11/16, so if this lines up with the week ending 11/18 in the MassDPH data which had 249 for Hampshire County, UMass is now accounting for 31% of the cases. Also a caveat: some of the UMass cases are residents in other counties, and wouldn’t be part of the Hampshire numbers).

Town and city data are released once a week in the MassDPH in a public health report, the latest of which latest of which came out on Nov. 12th. This one reports data over a two week period ending Nov. 11th (the fine print tells us “current at 8 am that day”). The first page of the data, which includes Amherst, is shown here. Note that the color coding changed on Nov. 5th (see the current report for the criteria). Many of the current yellow designations, including Amherst’s, would have been red under the old system – the red designation used to indicate a rate greater than 8 per 100K.

I extracted the case counts for the towns and cities in Hampshire County, and found that the total seems to match the 14 days ending Nov. 10th in the MassDPH daily county data I used for the above graph (“For populations <50,000, <5 cases are reported as such or suppressed for confidentiality purposes”, so an exact sum is not possible). The 81 cases in Amherst seem to match the two weeks of data ending Nov. 8th on the UMass Amherst website, which makes sense if there is a reporting lag. In the table above, there appears to be an asterisk missing beside Amherst, which is supposed to be there when a large proportion of cases are from a long term care facility or higher education institution. But more than an asterisk is needed to convey to Amherst residents the extent to which spread has affected their community (e.g. the positivity rates are skewed by all the asymptomatic testing at the university). And a weekly report is often out of date with this quickly spreading disease. (Update 11/24 – Emily Riddle points out that I missed another bit of fine print on the public health report, which states the two week period as ending 11/7. She also confirms that a two-day lag seems to best line up the UMass and MassDPH dates).

Pulling back from the details, it’s clearly not easy to go from an observation of a spike in county data on the New York Times website to any finer grained analysis of what’s happening locally, and again, it’s not obvious that we should be going to a national newspaper’s website for up-to-date county data. As individuals navigating this epidemic, it’s like driving our cars at night with the headlights broken and not being able to pull over.

Risk analyses for individual and local policy decisions involve using local new case data and positivity rates to estimate risk. As my attempt to find out what’s happening in Hampshire County should show, it’s not easy to get good data (the just-linked individual decision article provides further discussion and examples from elsewhere in the country). And there is much more data that would be helpful for individuals trying to avoid catching and spreading this disease, as well as for those deciding local policy. The discovery that about half of the recent new cases in Hampshire County are fairly localized comes as a relief to those of us who are not in contact with UMass Amherst students and staff, because it means that the infections are not as widespread through the broader community (UMass also carefully documents the quarantine status of the individuals). Similarly, knowing that an outbreak occurred in a long-term care facility would make one feel relatively secure about going to the supermarket, whereas knowing that it occurred in the supermarket would have the reverse effect (if it remained open). But that sort of information is not shared systematically by our public health departments. I want to emphasize that the blame here should not be placed on the people doing the incredibly difficult public health jobs. The problem is almost certainly lack of resources, caused by the decades of cuts to our public health system.

Because of its resources, UMass Amherst can provide excellent data reporting, including detailed information about the status of individual cases. This could serve as one model for local governments and their health boards, if they are given the necessary resources by the state and federal governments to implement improved reporting.

Those interested in improving public reporting of data could also look to countries where public health systems are in better shape, like Canada. Kingston Ontario has done exceptionally well against COVID-19, thanks to the work of their public health board, and their Covid-19 dashboard is excellent (their total case count is right now just above Hampshire County’s weekly count, with about the same population). The fact that our boards are town- and city-based, and that Kingston’s serves a population of about that of Hampshire County, presumably also contributes to the differences in the quality of the reporting. Perhaps as well as further investment, a fundamental reorganization of our public health system is needed for it to function more successfully. See this paper for a discussion of the response of the public health system to the current situation, and some historical perspective; it discusses the fragmented nature of the response, which seems acute in Massachusetts with our town-city boards.

In his Nov. 10th briefing, Governor Baker pointed to the incoming Biden-Harris administration as providing hope for a more effective federal response to our current health crisis. As in many other things, we should not be satisfied with a return to the pre-Trump situation in what our government provides for us in public health. We should also not be led to believe that it’s purely a matter of individual responsibility to stop the spread of disease, just like it’s not purely a matter of individual responsibility to slow climate change.

==============

Further details on the sources of data.

Daily counts of new cases and deaths for Hampshire County came from three sets of raw data, since historical data for all dates cannot be found in the latest one, and the earliest data was also removed at some point. They were downloaded from https://www.mass.gov/info-details/archive-of-covid-19-cases-in-massachusetts. The data from March 24th to April 17th came from the May 16th report, those from August 11th came from the August 11th report, and those from August 19th to November 12th from the November 12th report. The compiled new case data and graphs can be downloaded in a Numbers spreadsheet; other formats available on request.

Daily averages of testing, hospitalizations and deaths for Massachusetts were found using the interactive graphs on the New York Times website on November 13th.

==============

Acknowledgments: Thank you to Carla Caruso, Will Meyer and Joe Pater (père) for reading and commenting on a draft of this piece, and to Brendan O’Connor for making the place comparison graph. Comments are welcome below; they may not appear immediately since they may need to be moderated. I also posted this in Facebook and am discussing it there; it’s clear from that discussion that at least some people did hear about the spike for the first time from me, which is really bizarre and problematic.