In this incredibly instructive almost hour-long video from a webinar by ISQua’s (The International Society for Quality in Health Care) Education Channel, see how control charts are being used to analyze the COVID-19 data from states in the United States. The webinar is presented by Rocco Perla and Lloyd Provost.

The presenters always look at two important questions regarding data analysis:

• What can we learn from the variation in the data?
• Is the variation in the data from a stable system or has something changed?

They make the important point that the media and decision-makers are not using the theory of variation to analyze the data and instead make decisions based on ups and downs from data point to data point.

What is needed are answers to the following questions:

• Are things getting better, worse, or staying the same?
• Are the changes we’re making improving things or making things worse?
• Are we in or out of the growth phase of death?

The answers to these questions can come from the theory of variation originally proposed by Walter Shewhart in the early 1930s which defines common causes and special causes and is operationalized in the familiar Shewhart Control Chart.

Rocco and Lloyd focus on analyzing Coronavirus deaths as opposed to cases because the latter depends on the various testing strategies that were used in a particular location or country.

They find that the standard C-chart for counts does not apply in an epidemic environment. As a result, they describe how to model such data through the following progression before settling in on the appropriate approach.

1. I-chart or Individuals Chart of number of deaths
2. I-chart with regression center line
3. I-chart based on log-death regression center line
4. Transform I-chart back to original scale

The presenters introduce MDmetrix, a group of clinicians who use statistical process control based on the model described.

You can visit the MDmetrix site here. There you can see U.S. state by state charts updated every day. MDmetrix uses a basic run chart until growth dictates the use of the log linear regression I-chart.

From those charts, we see that

• States are in different phases of growth
• Many states (but not all) are progressing

Rocco and Lloyd and team determined there are three general phases to a “Death Curve”:

• Phase-1 (Pre-growth)
• Phase-2 (Growth)
• Phase-3 (Post-growth)

Based on their model which doesn’t capture pre-growth, Rocco and Lloyd then combined two charts, the C-chart for Phase 1 and the log linear regression chart to model all three phases in one combined chart. They provide the rules for when to go from a C-chart to the log linear regression I-chart.

Plotting the states by growth phase gives an important overall picture of where the United States is in the COVID-19 epidemic.

The last 10 minutes of the video focuses on the questions from the webinar video attendees.

You can watch the ISQua video here.

Of course, the methodology can and should be applied to all countries.

NOTE: You can view the COVID-19 Data Dashboard on the Institute for Healthcare Improvement (IHI) website here.