Abstract

Improvement of health care are needed

to become different in processes of service. The process performance is

complicated by the existence of natural variation if it determined these

changes are having the beneficial effects. Such as same measurements are

repeated will get different values and some of it will indicate a better or

worst performance. Traditional statistical analysis methods will delay the time

of decision making due to a lots of measurements over time but it also consider

method that account for natural variation. Statistical process control (SPC) as

a tool for research and healthcare improvement because it is a kind of

statistics that combines presentation of data with time series analysis method.

The tool make the data more quickly and in a understand way to decision makers.

There are 7 main tools that mentioned in the statistical process control (SPC).

In this paper, we mentioned the primary tool which is the control chart. This tool

make a better understanding and communicating data from the healthcare

improvement efforts to all the researchers and public. This paper also provides overview of statistical

process control and few examples of the healthcare applications which applied

primary tool of SPC—the control chart.

Introduction

Nowadays,

health care is quite important for everyone include younger, teenagers, and

also elder because they know body health are quite important and necessary to

them. Without health, they cannot do anything that they want like travel, enjoying

life and so on. Maybe they just stay at the hospital or lying on the bed only

without do anything. Last time, people only busy work for earn money without

care their health and pollution level of world is not so much serious compared

to now. In last few years, people noticed that health care is quite important

for them to maintain a good and healthy lifestyle. They not only busy their

work and also care about health care like eating a healthy meal, buy a health

care insurance, always make appointment for body checkup and so on. Not only

that, they also care about that due to pollution nowadays quite serious

compared to last time. But, still a lot of people don’t know what is the

healthcare and why is it important. And nowadays health care should make some

improvement to benefit for everyone.

Health

care also called healthcare is a kind of action to maintain or make improvement

of health via the prevention, diagnosis, treatment and other physical and

mental impairments inside human body. Health professionals delivered healthcare

in some occupation like physicians, optometry, audiology, pharmacy, psychology,

physician associates, dentistry, midwifery, nursing, medicine, and other health

professions.

Economic

and social condition mostly always affect the health care when the healthcare

may vary across individuals, groups and individuals. Every country have different

policies and plans to related personal or grouped health care within their

social. Due to meet the health needs for populations, health care system should

be prepared organization. Their exact

configuration will make a difference between national and subnational entities.

In some country, planning occurs more centralized among government and private

organization while in some country, the plan is distributed among market

participants. According to the World Health Organization (WHO), a

well-functioning healthcare system requires a robust financing mechanism; a

well-trained and adequately paid workforce; reliable information on which to

base decisions and policies; and well maintained health facilities and

logistics to deliver quality medicines and technologies.1

Healthcare can

contribute to a large part of a country’s economy. It is important to promote

physical and mental health to people around the world. Example, in 1980, the

first disease in human history to be completely eliminated by health care

interventions was declared by WHO. 4

Even thought,

all improvement needed change, but not all change will carry out in the

improvement. The major components of measurement are:

(1) Determine and define key

indicators

(2) Collect data

(3) Analyze and interpret data

The third component which is analysis

and interpretation of data by using statistical process control (SPC) was

focused by this paper. To determine whether changes in processes to make a real

difference in outcome, statistical process control (SPC) can do so. In this

paper, we discuss an overview of statistical process control (SPC) theory, explain

control charts which is a major tool of SPC, and provide examples of their

application to common issues in healthcare improvement.

The basic theory of statistical

process control was developed in the late 1920s by Dr Walter Shewhart,3 a

statistician at the AT&T Bell Laboratories in the USA, and was popularised

worldwide by Dr W Edwards Deming.4 Both observed that repeated measurements

from a process will exhibit variation—Shewhart originally worked with

manufacturing processes but he and Deming quickly realized that their

observation could be applied to any sort of process. If a process is stable,

its variation will be predictable and can be described by one of several

statistical distributions.

One such model of random variation is

the normal (or Gaussian) bell shaped distribution which is familiar to most

healthcare professionals. While repeated measurements from many processes

follow normal distributions, it is important to note that there are many other

types of distributions that describe the variation in other healthcare

measurements such as Poisson, binomial, or geometric distributions. For

example, the random variation in the number of wound infections after surgery

will follow a binomial distribution since there are only two possible

outcomes—each patient either did or did not have a postoperative infection with

about the same probability (assuming that the data are adjusted for patient

acuity, surgical techniques, and other such variables).

SPC theory uses the phrase “common

cause variation”to refer to the natural variation inherent in a process on a

regular basis. This is the variation that is expected to occur according to the

underlying statistical distribution if its parameters remain constant over

time. For example, the random variation between body temperatures within a

population of healthy people is a result of basic human physiology, while the

random variation in week to week wound infection rates is a result of factors

such as training, sources of supplies, surgical and nursing care practices, and

cleanliness procedures. Processes that exhibit only common cause variation are

said to be stable, predictable, and in “statistical control”, hence the major

tool of SPC is called the “statistical control chart”.

Conversely, the phrase “special cause

variation”refers to unnatural variation due to events, changes, or

circumstances that have not previously been typical or inherent in the regular

process. This is similar to the concept in traditional hypothesis tests of data

exhibiting statistically significant differences, a key distinction being that

we now test for changes graphically and over time using small samples. For

example, heavy demand for A&E services brought on by an influenza epidemic

may create special cause variation (statistically significant differences) in

the form of increases in A&E waiting times. As another example, suppose

that the daily mean turn around time (TAT) for a particular laboratory test is

64 minutes with a minimum of 45 minutes and a maximum of 83 minutes; this mean

has been observed for several months. One day the mean jumps to 97 minutes

because a major power outage caused the computers to go down, the lights to go

out, and the pneumatic tube system to become inoperative. On this particular

day the process is said to be “out of control”and incapable of performing as it

had in the past due to the “special cause”of the power outage.

Note that special cause variation can

be the result of either a deliberate intervention or an external event over

which we have little control. Special causes of variation can also be transient

(being short staffed in A&E one day due to illness of a key person) or can

become part of the permanent common cause system (eliminating a staff position

through a budget cut).

Interventions in a research study or

change ideas in a quality improvement project are deliberate attempts to

introduce special causes of variation. Statistical tools are therefore needed

to help distinguish whether patterns in a set of measurements exhibit common or

special cause variation. While statistical process control charts and

hypothesis tests are both designed to achieve this goal, an important

difference is that SPC provides a graphical, simpler, and often faster way to

answer this question. The basic principles of SPC are summarised in box 1.

Box 1 Basic principles of SPC

Individual measurements from any

process will exhibit variation.

If the data come from a stable common

cause process, their variability is predictable within a knowable range that

can be computed from a statistical model such as the Gaussian, binomial, or

Poisson distribution.

If processes produce data with

special causes, measured values will deviate in some observable way from these

random distribution models.

Assuming the data are in control, we

can establish statistical limits and test for data that deviate from

predictions, providing statistical evidence of a change.

These observations lead to two

general approaches for improving processes. Because processes that exhibit

special cause variation are unstable and unpredictable, they should be improved

by first eliminating the special causes in order to bring the process “into

control”. In contrast, processes that exhibit only common cause variation will

continue to produce the same results, within statistical limits, unless the

process is fundamentally changed or redesigned.

Moreover, if a process remains in

control, future measurements will continue to follow the same probability

distribution as previously—that is, if a stable process produces data that

follow a normal distribution and it is not further disturbed by special causes,

we can expect about 95% of future measurements to fall within ±2 standard

deviations (SD) around the mean. We can make similar statements about

prediction ranges associated with any other statistical distribution. In

general, regardless of the underlying distribution, almost all data will fall

within ±3SD of the mean if the underlying distribution is stable—that is, if

the process is in statistical control.

The control chart therefore defines

what the process is capable of producing given its current design and

operation. If a different level of performance is wanted in the future, we must

intervene and introduce a change in the process—that is, a special cause. If we

simply want to sustain the current level of performance, special causes of

variation must be prevented or eliminated. Control charts can often help to

detect special cause variation more easily and faster than traditional

statistical methods, and therefore are valuable tools for evaluating the

effectiveness of a process and ensuring the sustainability of improvements over

time.

THE CONTROL CHART: THE KEY TOOL OF

SPC

Shewhart developed a relatively

simple statistical tool—the control chart—to aid in distinguishing between

common and special cause variation. A control chart consists of two parts: (1)

a series of measurements plotted in time order, and (2) the control chart

“template”which consists of three horizontal lines called the centre line

(typically, the mean), the upper control limit (UCL), and the lower control

limit (LCL).

To interpret a control chart, data

that fall outside the control limits or display abnormal patterns (see later)

are indications of special cause variation—that is, it is highly likely that

something inherently different in the process led to these data compared with

the other data. As long as all values on the graph fall randomly between the

upper and lower control limits, however, we assume that we are simply observing

common cause variation.

Where to draw the UCL and LCL is

important in control chart construction. Shewhart and other SPC experts

recommend control limits set at ±3SD for detecting meaningful changes in

process performance while achieving a rational balance between two types of risks.

If the limits are set too narrow there is a high risk of a “type I

error”—mistakenly inferring special cause variation exists when, in fact, a

predictable extreme value is being observed which is expected periodically from

common cause variation. This situation is analogous to a false positive

indication on a laboratory test. On the other hand, if the limits are set too

wide there is a high risk of a “type II error”analogous to a false negative

laboratory test.

For example, for the familiar normal

distribution, in the long run 99.73% of all plotted data are expected to fall

within 3SD of the mean if the process is stable and does not change, with only

the remaining 0.27% falling more than 3SD away from the mean. While points that

fall outside these limits will occur infrequently due to common cause

variation, the type I error probability is so small (0.0027) that we instead

conclude that special variation caused these data. Similar logic can be applied

to calculate the type I and type II errors for any other statistical

distribution.

Although traditional statistical

techniques used in the medical literature typically use 2SD as the statistical

criteria for making decisions, there are several important reasons why control

charts use 3SD. For the normal distribution approximately 95% of the values lie

within 2SD of the mean so, even if the process was stable and in control, if

control limits are set at 2SD the type I error (false positive) rate for each

plotted value would be about 5% compared with 0.27% for a 3SD chart. Unlike one

time hypothesis tests, however, control charts consist of many points (20–25 is

common) with each point contributing to the overall false positive probability.

A control chart with 25 points using 3SD control limits has a reasonably

acceptable overall false positive probability of 1–(0.9973)25?=?6.5%, whereas

using 2SD limits would produce an unacceptably high overall false positive

probability of 1–(0.95)25?=?27.7%! The bottom line is that the UCL and LCL are

set at 3SD above and below the mean on most common control charts.5

In addition to points outside the

control limits, we can also look more rigorously at whether data appear

randomly distributed between the limits. Statisticians have developed

additional tests for this purpose; for example, a common set of tests for

special cause variation is:

one point outside the upper or lower

control limits;

two out of three successive points

more than 2SD from the mean on the same side of the centre line;

four out of five successive points

more than 1SD from the mean on the same side of the centre line;

eight successive points on the same

side of the centre line;

six successive points increasing or

decreasing (a trend); or

obvious cyclic behaviour.

In return for a minor increase in

false positives, these additional tests greatly increase the power of control

charts to detect process improvements and deteriorations. The statistical

“trick”here is that we are accumulating information and looking for special

cause patterns to form while waiting for the total sample size to increase.

This process of accumulating information before declaring statistical

significance is powerful, both statistically and psychologically.

A final important point about the

construction of control charts concerns the mechanics of calculating the SD. As

with traditional statistical methods, many different formulae can be used to

calculate the SD depending on the type of control chart used and the particular

statistical distribution associated with that chart. In particular, the formula

for the SD is not the one typically used to calculate the empirical SD as might

be found in a computer spreadsheet or taught in a basic statistics class. For

example, if we are monitoring the proportion of surgery patients who acquire an

infection, the appropriate formulae would use the SD of a binomial distribution

(much like that for a conventional hypothesis test of proportions); if

monitoring a medication error rate the appropriate formulae would use the SD of

a Poisson distribution; and when using normally distributed data the

appropriate formulae essentially block on the within sample SD in a manner

similar to that used in hypothesis tests of means and variances. Details of

calculations for each type of control chart, when to use each chart, and

appropriate sample sizes for each type of chart are beyond the scope of this

paper but can be found in many standard SPC publications