GitHub Repository

Executive Summary

The COVID-19 pandemic profoundly impacted the City of Chicago, particularly communities of color on the South and West sides. As of October 20th 2021, COVID-19 has claimed 6024 lives, mostly in majority Black and Hispanic neighborhoods and among people 60 years old or older. This analysis uses City of Chicago Department of Public Health data to generate a high-level overview of pandemic trends and assess the effectiveness of vaccine deployment in preventing death among the most vulnerable populations.

Key Findings

  • Covid deaths are varied over time, characterized by major spikes

  • 41% of covid deaths were Black, 32% were Latinx

  • 79% of covid deaths were 60 years old or older

  • Although vaccine deployment was done in phases, the initial introduction of the vaccine to healthcare workers was most strongly correlated with reducing covid death rates among vulnerable populations

  • The first phase of the vaccine deployment resulted in 1.12 fewer covid deaths per day among Black Chicagoans, and 2.07 fewer for Latinx Chicagoans, and 3.16 fewer for people 60 and older

Data

I retrieved this data from the City of Chicago open data portal1. This dataset contains the rolling average of cases, deaths, and hospitalizations for every seven-day period ending on each date. Every row is a single day corresponding to the end of a seven-day period, and 97 columns comprise various demographic categories. For this analysis, I focus on death counts and rates (per 100,000 people) by age and race.

This dataset treats both age and race as categorical variables. Age variables are divided into brackets: 0-17, 18-29, 30-39, etc. It has been well documented that older adults are most vulnerable to serious illness and death after getting infected with COVID-19. To simplify the analysis, I grouped age brackets into two categories: under 60 and 60 & over.

Exploratory Data Analysis

Let’s start with a broad overview of total cases, hospitalizations, and deaths over the whole pandemic. The data starts on March 13th, 2020.

Cases spike dramatically at different times, while hospitalizations and deaths remain relatively more stable. There are few possible inferences to draw from the divergence between the three rates. First, mitigation efforts were effective in preventing most cases from developing into severe illness. Second, it’s possible that cases during the later peaks were concentrated among less vulnerable people. Lastly, we improved somewhat in treatment and behavioral change to reduce exposure to vulnerable populations.

Next, let’s focus in on death rates by race.

Non-Black and non-Latinx races are shaded green to highlight the racially disproportionate impacts of the pandemic. Although there is some variability over time, the first and second-most covid deaths during the pandemic belong to Black and Latinx racial categories.

Now, let’s examine death rates by age split between two age categories.

These charts illustrate the leading demographic predictors of death following a COVID-19 infection. Below is a summary table of these findings.

Covid Deaths in Chicago:

Total Deaths Proportion
Full Pop. 6024 100%
Black 2443.1 41%
Latinx 1951.5 32%
Over 60 4754.7 79%

Statistical analysis of vaccine efficacy

Near the end of 2020, the City of Chicago began distributing vaccines to frontline healthcare workers. Over the next several months, the City introduced a phased vaccine distribution campaign that prioritized older residents and younger people with co-morbidities2.

Phase 1B provides a clear marker in time to assess vaccine efficacy among vulnerable populations. Although it does not match perfectly with age brackets in our data, it’s close enough to evaluate the impact of vaccination on the 60 and over population, including Black and Latinx folks who may fall into the same age category. Since distribution takes time and immunization is delayed about two weeks after treatment, we should consider an appropriate range of time around vaccine deployment to properly measure population health trends. To capture pre and post trends before and after Jan 25, I subsetted the data to include only death counts that fall within December 1, 2020 and March 31st, 2021.

Phase 1B (Dec. 1, 2020 - Mar. 31, 2021) Deaths by Demographic
Dependent variable:
Black Latinx Over 60
(1) (2) (3)
Av. Deaths per Day -0.07*** -0.08*** -0.18***
(0.002) (0.002) (0.003)
Constant 1,288.34*** 1,492.55*** 3,391.35***
(35.16) (45.42) (64.80)
Observations 121 121 121
R2 0.92 0.90 0.96
Adjusted R2 0.92 0.90 0.96
Residual Std. Error (df = 119) 0.72 0.94 1.33
Note: p<0.1; p<0.05; p<0.01
dataset::CDPH
lm() univariate



The regression table shows the coefficients for each demographic of interest. Each has a statistically significant and negative value indicating an average reduction in deaths: 0.06 fewer per day for Black Chicagoans, 0.07 fewer for Latinx, and 0.18 for 60 and over. The chart below shows the same data in weekly totals to aid visual clarity.

Interestingly, the downward slope is relatively smooth across the five-month period on either side of the Phase 1B rollout on January 25. This indicates that the downward trend started earlier. Let’s try the same thing, but now looking at the timeframe around Phase 1A when the vaccine was made available to frontline healthcare workers.

It’s also worth noting that a major spike in cases, hospitalizations, and deaths occurred around the holiday season, so the downward slope following that period may be at least partially attributable to trends stabilizing after a local peak.

By pushing back our timeframe by one month (November 1, 2020 - February 28, 2021), we can see a very clear shift in covid deaths among each demographic population subset just before the vaccine was deployed to medical staff. I think there are a few key takeaways evidenced by this chart. First, the downward trend started before vaccines came online. We see a peak between Thanksgiving and Christmas, then a steep decline heading into 2021. Infection and death rates spiked when people gathered for Thanksgiving, but started to drop afterwards. Second, once vaccines were introduced into the population through nurses and other healthcare professionals, the downward trend continued instead of spiking again after Christmas.

There is another potentially disturbing implication in this data, although I cannot comment on its validity without more research and collaboration. The demographics shown in these charts don’t include an occupational subset for the healthcare industry. But, deploying vaccines to them dramatically decreased covid deaths among the most vulnerable populations. One potential way to interpret this data is that frontline healthcare workers are major disease vectors themselves. The first two key takeaways suggest potentially cogent counterarguments. Factors outside of vaccination (like policy and behavior) impact transmission rates, which is why we see a huge peak right before lockdowns went into effect in Spring 2020. Also, introducing vaccinations into the general population at any level may have had a similar effect. Once a virus encounters an inoculated person, it is less likely to spread to another person breaking the chain of transmission within a population.

Regression Discontinuity Design

Finally, let’s estimate vaccination efficacy on the demographics in question using a regression discontinuity design (RDD) around Phase 1A. The regression table below was created by changing the independent date variable into a binary (before and after December 15, 2020), then running a similar regression as before. The dependent variables are death counts among demographic subsets on a given day between December 1, 2020 and March 31, 2021. The RDD specification is defined by the equation below:

\[Y_i = \alpha + \tau V_i(binary) + \epsilon_i\] Where \(i\) represents a given day during Phase 1A, the dependent variable \(Y\) represents deaths counts for a demographic subset, \(\alpha\) represents the constant, \(\tau\) represents the magnitude effect of vaccine availability \(V(binary)\) in the pre and post periods, and \(\epsilon\) represents the error term.

Phase 1A (Nov. 1, 2020 - Feb. 28, 2021) Regression Discontinuity Design
Dependent variable:
Black Latinx Over 60
(1) (2) (3)
Vax Available (Binary) -1.12*** -2.12*** -3.26***
(0.39) (0.45) (0.94)
Constant 6.16*** 6.91*** 15.52***
(0.31) (0.35) (0.74)
Observations 120 120 120
R2 0.07 0.16 0.09
Adjusted R2 0.06 0.15 0.09
Residual Std. Error (df = 118) 2.05 2.37 4.96
Note: p<0.1; p<0.05; p<0.01
dataset::CDPH
rdd() univariate



Model 1 estimates the causal effect \(\tau\) of the initial vaccine deployment in Phase 1A as 1.2 fewer covid deaths among Black Chicagoans on any given day between Nov. 1, 2020 and Feb. 28, 2021, and Model 2 estimates 2.07 fewer for Latinx residents. Model 3 estimates the largest causal effect is for people 60 and over at 3.16 fewer deaths during the 5-month timeframe examined.

This dataset limits the ability to analyze covariates such as the interaction between race and age. Since these data are death counts across demographic categories, there is significant overlap between people over 60 and people who are Black or Latinx. At best, we can identify proportional outcomes for different demographic groups. Without individual level data, estimating the causal effects of belonging to an age and race category would not be a valid statistical measure because the outcome variable \(Y_{over 60}\) and the predictor variable \(X_{race}\) are not independent of each other.

  1. https://data.cityofchicago.org/Health-Human-Services/COVID-19-Daily-Rolling-Average-Case-Death-and-Hosp/e68t-c7fv↩︎

  2. https://www.chicago.gov/city/en/sites/covid19-vaccine/home/vaccine-distribution-phases.html↩︎