COVID-19 and Lean Six Sigma

COVID-19 and Lean Six Sigma

Evan Dumas, LSSBB

Since I have no internet at-the-moment due to the storm that just ravaged the East Coast, and I’ve been asked by a bunch of people (read: no one) about my take on the whole flattening the curve thing - here are my thoughts on the matter:

First, let me give you some background.  I work in a health and insurance adjacent field, specifically relating to cost containment and risk assessment & mitigation.  I’m not exactly coming at this from a completely ignorant and purely speculative position.  I am also specifically trained in Lean Six Sigma process methodology, which itself has some applicable tools here that we can use to assess the current COVID situation.

In Lean Six Sigma (“LSS”), there are 2 things that we look for when we analyze and improve a process: Waste and Variation.  As most of you should be aware, these two factors are omnipresent in both the government and healthcare spaces – sadly, two places where it would benefit us most to reduce them.  At any rate, aside from the anecdotal comparison, what I want to focus on here a bit is the Six Sigma portion of the analysis, with a 36,000 ft. view of statistical variation as it relates to COVID-19.

I like my statistical analysis like I like my script fonts:

Source: http://my.ilstu.edu/

CURVY.

Let’s get into the whole “flattening the curve” thing - since I’m seeing a broad, fundamental misunderstanding of the concept.  In this case, flattening the curve doesn’t have anything to do with reducing the total number of people that are at-risk of becoming infected by coronavirus (although studies suggest that said total would be reduced as a by-product of the social distancing method of mitigation).  Rather, flattening said curve spreads out that total number of potential individuals over a longer period of time.  This is done so as to not thoroughly overburden an already overburdened healthcare system which was not built to accommodate a pandemic of this nature or magnitude.

The problem with COVID-19, and why it varies statistically from other viruses like, say, influenza, is because A) it is significantly more virulent than the flu and B) WE DO NOT YET HAVE A VACCINE.  Imagine if you will that a huge portion of the people that will have been infected with COVID, ALL got sick at nearly the same time.  Everyone would be going to hospitals at the same time, putting each facility way beyond its capacity to treat these people, at a level exponentially higher than what we are already seeing - not just for COVID, but for any communicable disease that can be transmitted via hospital cohabitation.  I don’t even want to get into the concept of exponents right now – let’s stick with curves.

When looking at a “normal” curve, we can assess probability by examining its Standard Deviation.  Now, I didn’t take statistics in high school or in college, but it seems like many of you didn’t either – or forgot what you were taught.  I’m coming at this from a semi-fresh (like that produce that’s been sitting in the fridge the last few weeks) perspective, having been trained in LSS within the last few years.  Standard Deviation is a measure of the variation of a set of data. It is important to answer questions such as how many measurements fall within which number of standard deviations from the statistical mean (average) of the curve.  In a predictive model, this is how probability is visualized.

At any rate, what these standard deviations tell us is the statistical likelihood of an event occurring.  In LSS, we call these events “defects”.  If we apply LSS methodology to the COVID problem-at-hand, we can infer some troubling statistics that I don’t think most people take into consideration.

When we think of probability in terms of percentage, we often think of 99% as being a great number as it relates to grades, percent of completion, return on investment, etc.  What statistics tells us is that 99% IS NOWHERE GOOD ENOUGH in this situation.  The Empirical Rule (Central Limit Theorem) tells us that:

• 68.26894921371% of the area under the curve is within one standard deviation of the mean.
• 95.44997361036% of the area is within two standard deviations.
• 99.73002039367% of the area is within three standard deviations.
• 99.99366575163% of the area is within four standard deviations.
• 99.99994266969% of the area is within five standard deviations.
• 99.99999970268% of the area is within six standard deviations.  

These percentages will become important in a bit. LSS also has a calculation called DPMO, or “Defect per Million Opportunity”.  This becomes immensely relevant when it comes to dealing with percentages of population, because said population can be measured in multiples of millions of people.  To date, the mortality rate for COVID-19 has been calculated in “Case Fatality Rates” or CFRs.  As of today (4.14.2020), there have been 587,281 confirmed cases of COVID-19, 23,656 of which have unfortunately resulted in death.  Right now, 12,772 patients are listed as being in serious or critical conditions.

Those statistics aside, let’s focus on the 99% metric.  Most of the stats and memes I see lately refer to a survivability rate of somewhere between 98.3% and 99% of patients, touting those numbers as a basis for reducing or eliminating social distancing restrictions.  I’m all for optimism, but we need to focus more on realism right now.  To that end, let’s work with the most conservative of those estimates (which are likely baseless anyway), our good friend, the ol’ 99%. 

99 out of 100 people that contract COVID-19 will survive. 

Sounds great right?!

We’re talking about human life here. Subsequently, human death serves as the “defect” in our DPMO calculation.  That’s a pretty concerning defect.  Let’s expand to 1000 patients.  For every 1000 patients, 990 patients will survive.  That is 10 human deaths.  10 people have lost their lives in this scenario.  Let’s break it down further:

• 9,900 survivals per 10,000 patients – 100 human deaths
• 99,000 survivals per 100,000 patients – 1000 human deaths
• 990,000 survivals per 1,000,000 patients – 10000 human deaths

Ten thousand human deaths per one million patients.  That’s a telling number, and one that would be considered unacceptable even in a manufacturing environment, calculated as a DPMO - where a defect may be something as trivial as a defective widget.  Again, we’re not talking about widgets here, we’re talking about HUMAN LIFE.  If we were operating at Six Sigma, for example - the goal of most manufacturing environments - that number would be reduced to 3.4… from 10,000.

 

9996 LIVES SAVED.

THAT’S how important statistics are when analyzing and looking to mitigate risk during a pandemic.

When assessing risk, there are a number of factors that must be considered.  I’ll let the actuaries try and calculate all the various probabilities; there are a bunch of models out there as it is.  What I want to illustrate is how many people will be put at-risk, given the simple 99% model we used above. The United States currently has an estimated population of 331,002,651. Let’s use every last person in this calculation.  At a 99% survivability rate, that accounts for 3,310,027 deaths. 

THREE MILLION, THREE-HUNDRED TEN THOUSAND, AND TWENTY-SEVEN HUMAN DEATHS. 

This doesn’t even factor in the number of people that will be significantly affected, medically and financially, by COVID and related treatments.  Those people are at-risk as well. Let’s keep this in terms of mortality though, since it’s the most poignant of metrics.  I'm currently waiting on a legitimate mortality regression analysis, from which we may be able to draw even more conclusions.

I want an open economy too.  I want to go to restaurants and eat pizza ‘til I explode, but it’s important to weigh ALL the factors here. 

Think about this the next time you’re so quick to dismiss that 1% of the population that is at risk of death. 

Take from this what you will.