MONDAY, APRIL 21, 2014
Font Size
Methodology Statistics in Everyday Life

Statistics in Everyday Life

The internet has brought us so much more information. Statistics are often used to support our opinions or views, and they show up in all types of media, including the internet. But are we more informed and educated on issues as common and important as health care? Let’s take a look at some numbers that appeared in a recent US News article “Health Reform Takes Aim at Hospital Readmission Rates.”

Here is one.

A study of heart failure patients published in June revealed that as hospital stays shortened between 1993 and 2006, the readmission rate jumped by 3 percentage points.

So how much does “jumped by 3 percentage points” inform us about the change? Is it too much or too little? Compared to what baseline? It could be a lot if the rate was at 1% and is now 4%, or it could be insignificant if it was 90%. What is the variation or measurement error?

Here is another example:

The latest data show, for example, that Florida Hospital in Orlando has a rate of 23.0 percent for heart attack patients, compared to a much-better-than-average 15.9 percent at Sarasota Memorial Hospital. In Iowa, a pneumonia readmission rate of 20.8 percent at Trinity Regional Medical Center in Fort Dodge compares with 14.6 percent at Mercy Medical Center in Cedar Rapids.

Is 23.0% really different from 15.9%? How do you define “much-better-than-average?” Are 20.8% and 14.6% different? What is the variation within each hospital year to year? What is the variation among all the hospitals? How can you determine if the two are different if you don’t know these variations?

There is a table showing 10 hospitals with the Highest Readmission Rates (31.6-32.4%) and 10 with the lowest (17.3-19.3%), with an introductory statement that “Medicare payments to hospitals that readmit too many patients within 30 days of discharge will be trimmed. “ How do you know whether the difference observed is due to statistical variation in the system or a particular hospital? Or in other words, the best/worst ones could show up as the worst/best the next time you measure, without any change in their own practices?

The following two statistics come under the subtitle Better Stats. (Really?)

Three months after the program was adopted at a 30-bed unit within St. Mary’s last year, the 30-day readmission rate had dropped to 7 percent from 12 percent.”

Jerry Penso, the group’s quality director, says the “all cause” readmission rate for older patients is 13.8 percent, compared with the 20 percent national average.

How do you know if the drop from 12% to 7% is not due to statistical variation? What would be the value if we measured in another 30 days? And again, is 13.8% really different from the average (20%)? Or is it just statistical variation?

These questions are obvious to people familiar with Edwards Deming’s management teaching. His books Out of the Crisis (1982) and The New Economics for Industry, Government, Education (1994) beautifully illustrated the importance of statistical thinking and the devastating effects of incorrect use of data and measures.

The world is rapidly changing, and we know its impact. However, there is still a lot that hasn’t changed for decades, including the need to learn and continuously improve our ability to understand the change around us. If not, we will continue to be misled by the wrong information and decisions.

Register Now

  • Stop this in-your-face notice
  • Reserve your username
  • Follow people you like, learn from
  • Extend your profile
  • Gain reputation for your contributions
  • No annoying captchas across site
And much more! C'mon, register now.

Leave a Comment


Comments

GJB 12-08-2010, 07:42

Nicely put. It certainly explains my frustrations in the variance in my golf scores.

Reply
Michael Toomey 12-08-2010, 10:47

Common? Yes. Unfortunate? Most certainly…and healthcare is not unique. But you would think that an operation with PhD Epidemiologists in the offices down the hall from decision support would know better than to spout spurious correlation. Yet, I see these stats all the time and when I challenge them I usually find the denominator <100 and confidence intervals that overlap. The upside is that most people are coachable and quickly understand the error when it is detailed out.

For an ice-breaker at the start of your next Six Sigma class, take a copy of the day’s USA-Today. There will be an article somewhere in the edition that demonstrates this problem and it will make for a fun discussion.

Reply
Fang Zhou 14-08-2010, 09:52

Thanks for the idea on using USA-Today for examples of such problems.

I wish more Six Sigma training will include useful examples and excercises to show people how easily we can be misled by information if we don’t develop critical thinking.

Many people are not aware that the data reported are often observations or measurements of convenient samples from the system of interest. Drawing conclusions about the system from the sample data requires statistical training.

I am sure a lot of statistics in reports are showing significant differences. But we live in a world where data are often incomplete and poorly collected or reported. Therefore, we have to be careful how to use them to support our decisions, and understand the potential risk if we did make decisions based on such data.

Reply
James E. Carroll 17-08-2010, 09:28

Bingo. You have shone the light of learning on the lay press’ interpretation of numbers.

Reply

Login Form