Jerry Langley is a statistician, author, and consultant. In his role with Associates in Process Improvement (API), Jerry helps the management teams of many organizations integrate quality into their business strategies. In addition, he guides teams and individuals in making changes that result in improvement. Clients include companies in the healthcare, pharmaceutical, electronics, automotive, financial, food, construction, and chemical/energy industries. Before joining API in 1985, Jerry worked as Statistical Services Manager for a Colorado division of Hewlett-Packard.

New York Time Berwick Op Ed - April 2, 2017

Don discussed why the ACA will persist despite attempts to repeal.


Continue reading
340 Hits

Understanding Variation - Request for Examples and Discussion

Provide examples and discussion for applying Shewhart's and Deming's ideas for understanding variation as described in the article:




Continue reading
654 Hits

Analyze Data by Provider Using Control Chart

Many organizations are involved in improvement activities aimed at cancer screening. There are known good ideas to improve screening percentages (use of CIS reminders, huddles, max-packing, etc.). However, there is another source of ideas for improvement that may already be in the organization, but hidden.

Consider these data for breast cancer screening for 60 providers in an organization:


Provider Percent
Provider Percent
Provider Percent
Provider1 58.1 Provider21 53.3 Provider41 55.0
Provider2 42.5 Provider22 50.0 Provider42 54.5
Provider3 28.0 Provider23 34.5 Provider43 30.6
Provider4 41.9 Provider24 74.9 Provider44 42.9
Provider5 32.3 Provider25 45.2 Provider45 24.3
Provider6 57.6 Provider26 43.7 Provider46 53.6
Provider7 64.5 Provider27 50.0 Provider47 41.7
Provider8 37.3 Provider28 25.8 Provider48 100.0
Provider9 41.7 Provider29 55.4 Provider49 42.9
Provider10 61.0 Provider30 0.0 Provider50 43.3
Provider11 50.0 Provider31 62.1 Provider51 48.6
Provider12 77.3 Provider32 43.2 Provider52 36.8
Provider13 48.6 Provider33 43.8 Provider53 30.0
Provider14 56.0 Provider34 42.4 Provider54 46.6
Provider15 32.4 Provider35 58.2 Provider55 51.1
Provider16 51.0 Provider36 40.0 Provider56 29.2
Provider17 43.0 Provider37 54.3 Provider57 36.4
Provider18 54.4 Provider38 41.8 Provider58 43.4
Provider19 38.4 Provider39 43.1 Provider59 45.3
Provider20 63.0 Provider40 50.0 Provider60 45.9

Of course, the data vary. But is there variation is these data that might be useful for guiding improvement efforts? When one thinks about variation within the concepts from Shewhart about common causes of variation and assignable causes of variation, and apply that kind of thinking to our provider data, the following questions arise:

Are all the providers part of the same system? Are some of the providers outside the system? If so, what is different about the way screening is done with the providers who are outside the system.

How can we answer those questions? Shewhart gave us the operational definition of "outside the system" with the control chart methodology that he developed.

A run chart alone does not tell us who is in the system and who is outside (see first figure)

By using P-chart calculations, control limits can be calculated (this requires the raw data of count of patients who are up-to-date with their screening and the total number of patients eligible for screening within the providers panel). The figure below shows a control chart for these data.

Note that there are several provider outside the system (outside of the control limits). Note that the provider with the highest percent of patients screened is inside the system (100%). To help see these differences, we can highlight the data points that indicate a provider who is outside the system (see the figure below);

Red dots show the providers who are outside the system high and blue dot show the providers who are outside the system low.

What can you do with this analysis? Find out why these 7 providers are outside the system. What are they (their care teams) doing differently? The answers to this question will often result in effective ideas for changing the system.

The method described above for looking at variation among providers needs a cautionary note here at the end. These data are a snapshot in time. To look at variation among providers over time, the best tool that I have found is small multiples. Look for another posting on analyzing variation among providers over time where I will post some examples of small multiples.


Even Better Way to View to View Data by Provider (added May 18, 2017)

The funnel plot is a method takes advantage of the idea that the data being analyzed do not have a time order. So, when there is not an inherent order for a set of data, the investigator can order the data in a way that makes it easier to see who is in the system and who is out.

Below is a Shewhart chart of breast cancer screening percentages for 144 providers in a large health system. They are ordered by the size of each provider's panel of women who should be screened for breast cancer, based on age. The providers with the largest panel are on the right-most end of the horizontal axis and those with the smallest panels are at the left-most end of the horizontal axis. Note that most of the providers are "inside the system", meaning that their percentages fall within the Shewhart upper and lower limits. However, there are clearly some of the providers are outside the system high and some of the providers outside the system low.

Adding color-coded circles makes it clear which providers we should talk to about why they are outside the system high (blue in chart below). Knowledge about how to do a better job of screening could be learned from this group of providers and then tested and adapted for the rest of the providers.

Knowledge about why the providers in the red circles are not able to screen at the "system" rate would likely also prove very useful for improving screening rate.

Continue reading
2258 Hits
1 Comment