Real-time decision-making requires real-time information, which requires real-time data analysis and calculation procedures, while after-the-fact calculations require that all data are collected before any calculations or decisions are made. Real-time decision-making presents special programming challenges for collecting data, but the truly significant problems are presented in the statistical quality control (SPC) calculations and making decisions based on the calculations. There are no standard procedures for performing real-time calculations. Real-time calculations involve the calculation or adjustment of control limits and averages, as subgroup data are generated and making immediate decisions concerning whether the process is within or beyond acceptable limits. One important concern is the influence of custom calculation procedures on the final decisions compared to the customary after-the-fact calculation methods. This paper compares the results of real-time calculations with after-the-fact calculations for patient data.
Introduction
Real-time decision-making requires real-time data analysis and calculation procedures. Real-time calculations suggest that as a subgroup of data is generated, the control limits and average calculations are performed and the decision regarding whether the observations are within or outside the control limits is made. On the other hand, after-the-fact calculations require that all data are collected before any calculations or decisions are made. Real-time decision-making presents special programming challenges for collecting data, but the truly significant problems are presented in the statistical quality control (SPC) calculations and making decisions based on the calculations because there are no standard procedures for performing real-time calculations.
Real-Time Calculations
In the ICU, patient vital sign data are generated from clinical monitors and then transmitted to computers which calculate averages and control limits and chart the observations. Some monitors collect and transmit data for every heart beat, while other monitors collect the data and transmit them a some fixed interval such as one datum per second or one datum per 15 seconds and so on. The amount of autocorrelation in the data changes as the data sampling rate changes, but the calculation procedures do not change.
Some clinical monitors send data as soon as the data are available while others require some type of request for data instruction from the computer. Some clinical monitors are easy to send instructions to, while others have a complex procedure that must be followed.
Figure 1 shows the general flow of data and the steps needed to perform real-time SPC calculations. The steps are simple, but they must be followed exactly because each step depends on the prior step. In addition, the control chart produced is different from control charts created using after-the-fact procedures.
Figure 1 Real-time calculation procedure
Figure 2 shows a portion of a control chart (average and standard deviation charts) calculated using the real-time procedure. Identify the region of the chart where the background color is yellow. The yellow area shows when the control limits are being reset, that is the centerlines and the control limits are being recalculated. Notice how the control limits change in the yellow area. The idea is that the control limits are established during a base period (yellow background) and then the vital signs are monitored for changes relative to the base period.
Figure 2 Real-time patient control chart
The normal procedure for performing an after-the-fact analysis is to collect all the data needed, then using all the data, the control limits and centerlines are calculated and then the graph is drawn. Figure 3 illustrates a control chart for averages calculated using after-the-fact procedures. Notice that the upper and lower control limits for xbar and the centerline, x double bar are straight. Since all the calculations are performed after the fact, all the data were used for the calculations and there is no base period such as required in the real-time calculation example. The same data set is used for the control charts displayed in Figures 2 and 3.
Figure 3. After-the-fact control chart for averages
Fixed Decision-making Rules
A real-time program to generate control charts requires decision-making rules, while manual or spreadsheet control charts allow the user to decide what rules to use at any time as the charts are created. The program’s decision-making rules are based on the numbers displayed on the bottom of Figure 1. The number of subgroups before the program automatically recalculates may be set at 1, 3, 5, 10, 15, or 20. The default value is 10, which is the number displayed in Figure 1. Default values for graphing are part of the program. When collecting data, the default values sometimes do not center the output on the screen. If the autoscale function is on in the program, then the program will calculate the graphing limits and center the output on the screen. The number of standard deviations is usually set at 3. The program allows the user to change the number of standard deviations to 2, 2.5, 3, 4, or 6. The subgroup size is determined when the data are collected.
When collecting data from a heart-beat to heart-beat clinical monitor for newborns, the subgroup size was changed to 20 to increase the elapsed time on the display. The desired elapsed time and the response rate of the screen depend on the nature of the patient and the time horizon associated with the current treatment. For example, in the operating room, a short elapsed time is acceptable, and a high reaction rate is needed. In an ICU, a longer elapsed time is needed and a slower reaction rate is acceptable.
Observations-Conclusions
Using the real-time based period calculation procedure results in a different timing of actions than when using the after-the-fact calculation method. Since the objective of SPC analyses is to identify change and to signal the caregiver to look for the reason for the change, the researchers’ concern was to determine which calculation procedure places the quality control outlier nearest to the event for identification purposes. We could identify no measurement system to determine which calculation method resulted in the outlier being closer to the event.
When the data collected in the base period was stable, the real-time calculation method seemed to have an advantage. Generally, calculations based on a stable base period resulted in a smaller standard deviation, tighter control limits, and a better decision-making tool.
We studied the real-time versus the after-the-fact calculation methods using many simulation data sets, as well as a collection of real vital sign data. There is no way to determine which calculation method is better; however, the results of this study indicate that there is a minimum penalty for using the real-time calculation method and in some situations, the real time method may be better.
BIBLIOGRAPHY