RE: control charts in health care

[Thomas Raithby <traithby@mhsqa.com>]

Mike,

In my area, ambulance services are delivered under a public utility model -
that is the government awards contracts for specific geographic areas to
ambulance service contractors. These contractors are held to specific
standards, such as the 8 minute response-to-scene / 90th percentile. There
are others also. The 90th percentile is an arbitrary (but industry accepted)
measurement. Contractors who exceed specification are awarded bonuses, while
those who do not meet the standard are penalized. 

Why 90%? I'm not sure where that came from. I think the fact that it is not
100% recognizes the multitude of factors that impact ambulance responses
that are beyond the control of paramedics and contractors:
* challenging weather conditions, such as snow storms
* the large geography, and areas of sparse population
* traffic congestion at selected times of day
* unanticipated events using lots of resources (like an apartment fire, or
transit bus accident)

The 90th percentile may be a compromise measure in some respects. While some
contracts cover urban areas with dense call volumes (to 11,000 calls per
year), other contracts are for remote areas with a single ambulance and an
annual call volume of 100. The 90th percentile may improve the validity of
measure with such small or variable volumes. I would have said sample sizes,
but I think census would be the proper term, given that every event is
measured by a separate entity (the dispatch agency).

I am using X-Bar S chart to monitor response time for the service with
11,000 calls per year. There are some cases of special cause variation, but
the response time is mostly in control (I think). I'm using Minitab to
process the data and plot the control chart. It plots a mean and the UCL
plus LCL and the daily means.

I'm wondering if there is another way to approach this, where I plot the
90th percentile each day, and look for the range of expected variation in
this measure (UCL and LCL). Is this a reasonable approach? Would a run chart
be more appropriate to plot daily 90th percentiles?

Perhaps another service industry example would be helpful? How long is an
acceptable wait in a lineup at a supermarket? Would it be reasonable to
expect you would wait less than 5 minutes 90% of the time? That it would not
cause you to boycott the store if you waited more than 5 minutes one trip in
ten?

I am new to SPC, and equally new to Minitab. If these topics were a car, I
know how to put gas in and change the oil, but I don't understand most of
the operation of the vehicle.

And I welcome any thoughts.

Thanks,

Tom

Thomas Raithby, BSc EMT-2 Consultant and Field Coordinator 
tel: (506) 649-2597 
fax: (506) 649-2529 
Mobile Health Services Quality Agency  www.MHSQA.com 
133 Prince William Street, Suite 802 
Saint John, NB E2L 2B5