What Am I Looking At? Reading Data in times of COVID-19
Authors Note: This story is co-authored with Dr. Mauricio Garnier-Villarreal, Research Assistant Professor for the College of Nursing at Marquette University. Medium does not have a platform for co-authorship.
As COVID-19 sweeps the globe we are being bombarded with data and statistics that communicate to us infection rates, risk assessments, and death. Yet, as we glare at our screens many of us might feel dumbfounded at the sheer amount of data that is circulating through our webs of social media and news sources. In efforts to contribute to better literacy of data in times of COVID-19 we have authored this piece enabling people to read data with a keener eye.
We write this article from Milwaukee, WI. The spread within the county of Milwaukee has unsurprisingly tracked the hyper-segregation of the city with a hyper concentrated spread in the North side; the historically black part of Milwaukee reminding us that death and debility are socially constructed. Further, we are still glaringly missing data on the infection rate in the Hispanic population of the city as our testing does not seem to be accounting for ethnicity. In this context, it is crucial to remember that the data on infection spread is tracked by testing, which is itself a scarce resource. Testing mechanisms are impacted by the social ordering in which they are placed. Testing is reflective of social and political conditions that construct our relationships with a healthcare system that runs for profit and is not equitably accessible. Moreover, we need to keep in mind that testing is not a value neutral experience and for certain communities testing means waging between the possibilities of positive confirmation and the possibilities of the loss of livelihood; both of which could be lethal. Hence, we should, as the Milwaukee County COVID-19 dashboard indicates presume that the cases are likely higher. Now, more than ever, it is crucial that we broaden our data literacy skill sets as we attempt to decipher what is really happening to our communities as COVID-19 continues to spread.
The silver lining: there is a lot we can still learn from the infection trajectories of other states and countries, but in order to do that we need minimally comparable metrics along time and population size.
How are we tracking COVID-19 rates of infection? How should we read the data of confirmed positive cases? How should we understand the impact of social factors on testing? Do we still have time to impact the outcome of the infection trajectory? Our goal is to shed light on the fact that the COVID-19 infection processes are multidimensional. We will witness both exponential growth in the rate of infection as well as its slowing down. It is just a matter of how we choose to behave on large scales that will change the outcomes of how that trajectory presents itself in data plots. If we intervene on spread through social measures that minimize community spread, social contact, and increase accurate testing we can alter the path of the infection trajectory. The silver lining: there is a lot we can still learn from the infection trajectories of other states and countries, but in order to do that we need minimally comparable metrics along time and population size. Hence, we must be vigilant of the categories with which we compare data and never forget that there is no such thing as value neutral terms.
The Infection Trajectory: Reading COVID-19 Data
COVID-19 data is largely presented in graphs that track infection counts, premised on positive testing in a population. The graphs are presented on an X, Y axis, where the X axis tracks time. The Y axis tracks the number of confirmed cases. In other words, what COVID-19 charts are tracking are infection rates, determined by positive testing, over the course of time. The timeline begins with the first confirmed positive case, which means that different places are operating on different clocks. For instance, New York saw its first confirmed positive case on March 2, 2020 whereas Wisconsin saw it on March 9, 2020. So, we are working on different timelines.
A COVID-19 infection process follows an S shaped trajectory (See Figure 1) where most states in the U.S. are currently in the first half of the development of the S shape, which follows exponential growth. If we look at the trajectory of a place like China, we can see the full S shaped trajectory whereas in the United States we are currently in a moment of exponential growth (See Figure 2). It is important to note that what we are looking at is not an exponential growth model, but rather a plotting of positive confirmed cases that is following exponential growth, and will at a certain point start to slow down. The point where the rate of infection starts to slow down is called an inflection point, featured in Figure 1 as the red vertical line and can be located at the middle point of the S shaped trajectory. The red lines in Figure 1 present the inflection point for two hypothetical infection trajectories. The part of the S shaped curve that comes before the red line (inflection point) is tracking exponential growth, and the part of the curve after the red line tracks the slowing down of the total number of infections. At the inflection point, the trajectory of the S curve changes, it starts to slow down, until it reaches the maximum number of infections (the asymptote) and stops spreading. In Figure 1 the maximum number of infections is identified by the blue horizontal lines. The maximum number of confirmed cases will be affected by multiple factors, these factors include but are not limited to, infection spreading processes, social contact, traveling, testing as a resource, and state and federal interventions on.
The effects of our interventions are delayed, and will be affected by what measures people take and how well they do them.
We should not treat the maximum number of cases as a fixed number, but rather it is a number that we as a society are trying to control and keep as low possible through mechanisms like social isolation and quarantine. The theoretical maximum number of confirmed cases is infection of 100% of the population. Similarly, the inflection point or the point where the rates of infection start to slow down, is also an unknown, as we are socially trying to intervene on spread in the hopes of making the inflection point happen sooner. The moment of inflection is further hard to predict because individual infection processes can take up two weeks, so any action that we take will be delayed for the same amount of time. In short, the effects of our interventions are delayed, and will be affected by what measures people take and how well they do them.
So What Does Flattening the Curve Really Mean?
The objective of flattening the curve is to alter the trajectory of infection, which can be seen by changing multiple characteristics: speed, inflection point, and/or maximum number of cases. Figure 1 presents two possible hypothetical trajectories of infection using the Wisconsin population as a metric. The first black S curve has an inflection point occurring 100 days from the first confirmed case and a maximum number of cases of 20% of the population of Wisconsin, which is 1,160,000 people. The second S curve identified in green presents a possible scenario of a flattened curve in which the speed of infection is half of the black S curve meaning that the infection rate is slower. By slowing down the infection trajectory the inflection point happens later on day 150. Assuming that population behavior is effective the maximum number of cases is 10% of the population of Wisconsin, which is 580,000 people. Yet, it is important to note that change can present in many possible ways contingent largely on social behavior because infection is a population process.
Any social and political interventions need to be made by a large majority of the population in order to have impact on infection rates…Just because you are not witnessing or directly in contact with the effects of COVID-19 does not imply that it is not spreading.
Infection is a Population Process
The infection trajectory of COVID-19 is a process that happens across a large number of people. For instance, in the case of Wisconsin, we are describing a process that affects approximately 5.8 million people. Small groups of people may not see or experience the infection directly. Similarly, the behavior changes of a small group will not have a large impact. Any social and political interventions need to be made by a large majority of the population in order to have impact on infection rates. Part of the reason we are highlighting the fact that COVID-19 infection is a population process is to prevent a false sense of security. Just because you are not witnessing or directly in contact with the effects of COVID-19 does not imply that it is not spreading. In this context, what matters is the behavior of large portions of people to commit to social isolation even if not state mandated. Impact will only be made if behavior is altered on large scales. Large scale social action can save lives.
Why Infection Trajectory Comparison Matters
We are witnessing a disorganized presentation of incomparable numbers between countries and states with no context on population size and the commencement of testing.
A notable trend circulating social media and news dissemination is the use of comparison that tracks numbers of deaths and confirmed cases across countries or states. We are witnessing a disorganized presentation of incomparable numbers between countries and states with no context on population size and the commencement of testing. We want to heavily emphasize that in order for COVID-19 infection trajectories to be truly comparable we need to adjust for population size (typically per 100,000 people) and the moment/time of the first confirmed case. We should presume that the moment of the first confirmed case is itself inaccurate as it is reliant on access to testing.
Comparisons are further complicated by social and political interventions. Processes of infection are going to be affected by when we start testing and when social distancing and quarantine are put into practice, not just when they were mandated. It is difficult to do fair comparisons if we are all working on different clocks that commence with the first confirmed positive case and with different population sizes that behave differently. The basic metrics of comparison are population size and time. However, there are many characteristics that affect infection trajectories and could be taken into account for more detailed comparisons. However, some of these relevant characteristics are difficult to track. Consider the challenges in tracking: adherence to social distancing, geographical location, migrations of infected peoples, access to testing resources, and culturally competent testing. These will vary by location and even vary within a country.
Nevertheless, comparison matters as it is a window into possible scenarios of infection trajectories. Comparison of infection trajectories creates intuitive predictability of where states or countries might be at on the infection clock. So, we might ask ourselves: Who do we (Wisconsin) look like given that our infection clock started on March 9? Comparison allows us to see the effects of policy decision making in light of who has had better outcomes depending on the implementation of testing, social distancing, quarantine, or stay at home policies. With an eye toward Wisconsin we can appreciate that based on geographical location, population size, and time line we have an advantage that others simply do not. This is crucial and we still have time to make proactive decisions that can save many lives. We are behind the coastal clocks (California and New York) and have the opportunity to look at the experiences of other countries and states to try and make proactive interventionist decisions that slow the rate of infection rather than merely reacting to COVID-19 spread.
Injection Trajectory Comparison by Country: Adjusted for Population and Time
Figure 2, on the left, demonstrates the importance of adjusting for population size per 100,000 people and time. The unadjusted plot records number of confirmed positive cases in relation to the date that the case was confirmed. It is a plot of raw data across China (red), Italy (green), South Korea (blue), and U.S. (purple). On the right we have a plot adjusted for population size and time of first confirmed case. On both plots South Korea (blue) has a low total number of confirmed cases over time even if adjusted for population size. China (red) on the other hand in the first plot presents a high number of total cases with a steep rise in the red curve, but when adjusted by population size in the second plot we notice that they have a low number of positive cases relative to their size. Italy (green) by raw numbers looks like it has the second highest number of cases, but when accounting for population size we see that their infection rate is higher by population size. Finally, the U.S. (purple) presents the highest total number of cases (currently the highest in the world) and the second highest relative to their size (Italy being the first).
The timelines in each country start relative to testing. We can appreciate that China, South Korea, and the U.S. have 70 days of testing positive cases whereas Italy has 61. Italy presents the highest rate of infection per population size over the course of less time. Further, it is important to notice that South Korea (blue) and China (red) are in the second stage of the S shape trajectory, as described above, meaning that they have reached their maximum numbers of positive cases while Italy and the U.S. are still in the first part (exponential growth).
Infection Trajectory Comparison by State: Adjusted for Population and Time
Wisconsin, on the other hand, has a slower rate of infection, at this moment in time, than Michigan or New York, which means that we are behind New York and Florida’s clock allowing us to make social interventions and better prepare for the pressures that about to come down on our health care system in the state. Wisconsin: It is not too late to act!
Figure 3 (above) is an infection trajectory comparison by state: Florida (red), Michigan (green), New York (blue), and Wisconsin (purple). The graph on the left is unadjusted for population size or time. The graph on the right is adjusted to account for population per 100,000 with the clock commencing with first confirmed positive case. The unadjusted plot reveals that Wisconsin has less cases than Michigan and Florida and further that Michigan and Florida are on a similar same path. However, when looking at the adjusted plot we see that Florida has about the same relative number of cases after more days of testing than Wisconsin. We can further appreciate that Michigan has a faster rate of infection relative to its population size with less days of testing, which is worrisome. Michigan appears to be in a closer path to the longer trend set by New York. In New York we witness the most days of testing with the highest relative number of infection. Wisconsin, on the other hand, has a slower rate of infection, at this moment in time, than Michigan or New York, which means that we are behind New York and Florida’s clock allowing us to make social interventions and better prepare for the pressures that about to come down on our health care system in the state. Wisconsin: It is not too late to act!
The Future is Now
Reading the data on the COVID-19 infection trajectories more closely enables us to see that time is of the essence. The actions we take now will not be felt immediately, but that does not mean that they will not work. Adjusting for population size and time our comparative infection trajectories reveal the importance of social/political/medical interventions in a future that is written by what we do now. Let us not forget that there is much to learn from this monumental moment, and the better testing and tracking data we aggregate now the more effectively prepared we will be in the future; come what may.
