|
Understanding
Variation in the Time Series
As
most researchers are aware, random or externally-produced
variations make it difficult to interpret time series data
in the absence of a control, particularly when the aim is
to evaluate the effect of a single intervention. Since controls
are often not feasible for large system improvement, skilled
practitioners of quality improvement use sophisticated analysis
of time series to protect against incorrect causal inference.
From a quality improvement perspective, the patterns in
the data for the measures of system performance provide
an opportunity to learn about the system, what affects its
performance, and how the cause of the trend might interact
with the changes. Understanding variation in the measures
of performance is central to the learning process.
The
theory for understanding variation in system performance
was developed by Shewhart (1986).
At times he observed patterns of variation that were of
similar magnitude over time and that were free of trends
and outlying values, such as these data on admissions for
congestive heart failure:
| Figure
4.2 Percent Admissions of Patients with Congestive
Heart Failure
|
|---|
 |
|
Source: Chronic II Collaborative,
CHRISTUS St. Frances Cabrini Hospital, Alexandria, LA. Reprinted
with permission from Suzanne Edwards.
Shewhart
concluded that these patterns reflected a system that was
stable with respect to the performance measures. He further
concluded that the multitude of causes that interacted to
produce the variation must be continuously present in the
system. He called them "common causes."
Shewhart
also observed other patterns that were not so regular. Trends,
abrupt shifts of average performance levels, and outlying
values inflated the variation. Shewhart concluded that these
unpredictable patterns were caused by some specific circumstances--a
"special cause" of variation. He hypothesized that a vigilant
and frequent observer of the time series, such as a nurse
providing pain relief daily to patients, could identify
the special cause of variation. He developed a method of
informed observation and action that has been called Statistical
Process Control (Wheeler, 1995).
This method argues that as the team identifies and removes
special causes, variation is reduced, the system becomes
stable, and the investigator is more reliably able to determine
the effect of subsequent changes.
In
sophisticated quality improvement, work systems are first
stabilized and variation is reduced, usually by instituting
changes such as standard protocols or guidelines. Only then
are more fundamental changes made in an attempt to raise
system performance. This sequence of stabilization, improved
performance, and re-stabilization at the new level of performance
allows practitioners of quality improvement to make sustainable
improvement and provide reliable evidence of the effectiveness
of the changes even when controls are not feasible.
In
addition to providing the concepts for understanding variation,
Shewhart provided the control
chart method, an effective way of finding a signal of
changed performance from repeated small samples.
|