TYPE I ERROR FOR PATIENT CONTROL CHARTS OF NEWBORN VITAL SIGN DATA


Steven Ringer, MD, Ph.D., Director of Newborn Services, Brigham and Women's Hospital, Harvard Medical School, Boston MA

Marjorie L. Icenogle, Ph.D., Assistant Professor of Management, University of South Alabama, Mobile AL
Steven M. Zimmerman, Ph.D. VP or Research, Biomedical Quality Control of America, Inc., Mobile, Al

Abstract

This study examines the rules for recalculating control limits when Shewhart-type patient control charts with control limits adjusted for autocorrelation are used and the data are collected with a heartbeat-to-heartbeat data collection monitor.  According to process control guidelines, when a series of data points fall beyond the control limits, the limits should be recalculated.  In a healthcare setting, the assumption is that the recalculation of control limits provides a signal for the caregiver to examine the patient in order to identify the cause of the change and to determine if a change in treatment is necessary.  A type I error occurs when the caregiver searches for a reason for a change, when no change has occurred, while a type II error occurs when the caregiver does not search for a reason for a change when a significant change has occurred.  From a clinical perspective, a pertinent research question is, “can caregivers identify statistically significant changes in patients’ vital signs that are also clinically significant so that important changes in vital signs are not missed?” This study investigates the application of various rules for recalculating control limits in a clinical setting so that clinically significant changes are not missed.  The study shows how computer monitoring affects the timing of patient examinations and aids caregivers in finding reasons for changes in a timely manner.

Introduction

Autocorrelation is an important issue in applying quality control charts to patient monitoring.  Autocorrelation occurs when an observation is a function of the prior observation. Patient vital sign data go in and out of autocorrelation by self-correcting (observations exceed the control limits and then come back within the control limits without any caregiver intervention). Control charts tracking patient vital sign data help caregivers identify changes that are statistically significant (observations that are beyond the control limits for a specified period).  However, a clinically significant change is a change in a vital sign that helps the caregiver identify the need to change the patient’s treatment.

In clinical monitoring a false indicator of a change (type I error) is an important issue. When a type I error occurs, the caregiver tries to identify the reason for the change in vital signs although no change has occurred.  Another false indicator of concern is a type II error.  When a type II error occurs, the caregiver does not look for a change when in fact the vital sign has changed.  In clinical monitoring, the objective is to balance type I and type II errors.  Therefore, the research question is:

How should vital sign data be collected so that clinically significant changes in vital signs are recognized without looking for changes when no changes have occurred?

This study examines a series of data files collected from premature babies during a period of time when there were no clinical actions taken by the caregiver.  The study focuses on the rules that are used to identify statistically and clinically significant changes in the data.  A type I error is defined as the average number of times the control chart system indicates that the caregiver should examine the patient looking for the reason for a change when no change has occurred.  Procedures that require the caregiver to examine the patient even when the change is a non-clinically significant change once an hour or less is considered acceptable.

The monitor’s data collection method is critical to this study; therefore, the researchers used a Nellcor 200 monitor, which collects data at the rate of one observation per heartbeat. It is important to note that the results of this study cannot be applied to another type of monitor that uses a different sampling methodology.  Sequential observations were collected in samples of 20 heartbeats.  As the 20 observations were collected, a computer connected to the monitor calculated the average and the standard deviation of each sample. The rate at which the control limits are reset is determined by the caregiver. The caregiver is expected to examine the patient whenever the control limits are reset.  Therefore, the caregiver may choose to reset the control limits when the average of one subgroup of observations is outside the control limits or the caregiver may prefer to recalculate the limits after the averages of several subgroups are beyond the control limits.  This study examines the differences in the number of times the caregiver would examine the patient if the following rules are applied: when a run of 1, 3, 5, 10, 15, 20, 25, or 30 subgroups are outside the control limits.

Experimental Approach

The real-time patient control chart in this study is based on the Shewhart concept and includes an adjustment of the control limits for autocorrelation and data truncation. Autocorrelated data are data that are anchored in the past. For example, it is not possible to change from an overweight individual to an underweight individual over night. The change takes time because the weight of an individual is anchored to the past.

The data collection rate was one observation per heartbeat and the subgroup size was based on 20 observations. The rule that required control limits to be reset after a single subgroup beyond the control limits is not practical because:

1. A newborn may have a heart rate of 160 beats per minute. During one hour 480 subgroups of size 20 are generated. Given a probability of 0.00135 of being beyond the three sigma limits then 0.648 subgroups may be expected to be beyond a limit on the average chart and approximately the same on the sigma chart.
2. There are an unknown number of self-correcting patient changes per hour.

When a reset rule of one point beyond the control limits was applied, the researchers detected about 5 observations beyond the control limits per hour, one every 12 minutes by chance alone, which created an unacceptable rate of type I errors.  Initially, the control chart program allowed the user to select: 5, 10, 15, 20, 25, or 30 subgroups in a row as the indicator that the vital sign had changed beyond the control limits.  When an outlier indicated occur the system was reset by recalculating the patient’s control charts control limits and center lines while at the same time changing the background color on the monitor from white to yellow.

Experiment

Fifty-eight (58) hours of patient vital sign data were selected from the database. The number of resets was counted using the reset rules specified above and the number of resets for each rule was compared. Figure 1 indicates a patient control chart for data set: D:\z\med\h\r-pt15\14a21a0.bqc.  The yellow background indicates that the control limits and centerlines have just been recalculated (reset).  The rule used was five (5) points beyond a control limit and three (3) sigma limits.

Figure 1 Patient Control Charts for SaO2 (Oxygen saturation)

When a reset rule of one point beyond the control limits was used the screen looked like it was always being reset.  Figure 2 illustrates the results using the same file as illustrated in Figure 1 with the reset rule of one observation beyond the control limits.

Figure 2 Reset rule - one point beyond control limit


Another way to compare the run rules is to examine the approximate amount of time the patient control chart is out of limits before action is taken.  On the assumption that a newborn’s heart rate is 160 beats per minute and using a run rule of 5 in a row, the amount of time out of limits to action is 0.63 minutes.  Using the ten (10) in a row rule the amount time out-of-limits to action is 1.25 minutes.

Figure 3 illustrates the results in terms of resets per hour based on the number of subgroups in a reset decision.  Using a run of five (5) subgroups beyond the control limits resulted in an average resets rate of 1.07 per hour for the SaO2 (oxygen saturation data) and an average resets rate of 1.01 per hour for the heart rate data.  Using a run of 10 subgroups beyond the control limits resulted in an average reset rate was 0.43 per hour for the SaO2 data and 0.53 resets per hour for the heart rate data. Using a reset rule of 15 observations beyond the control limits resulted in .29 and .38 resets per hour for SaO2 and heart rate data.  As the number of subgroups was increased, the number of resets declined.  As a result of this analysis the software was revised to allow the user to set the reset rules using 1, 3, 5, 10, 15, or 20 subgroups in the reset decision.  The default value was set at 10 subgroups in a reset decision.

Figure 3 Resets per hour based on the number of observations beyond the control limits

When using patient control charts for more than a single parameter, the ten in a row rule is recommended because the estimate of the type I error is calculated based on the total number of control charts needed to monitor the parameters.  Since the objective is to reduce the number of unnecessary trips to examine the patient based on the reset alarms, the ten in a row rule reduces the number of reset alarms, while still allowing for adequate patient monitoring.

Conclusions

Using the rule one subgroup beyond the patient control limits in this setting, results in the caregiver examining the patient every 12 minutes or more, even though no change in vital signs has occurred.  Using the rule of five subgroups in a row beyond the control limits, the patient will be examined approximately once per hour.   Using ten subgroups in a row beyond the control limits, the patient will have to be examined approximately once every two hours. We recommend the ten in a row rule, because of the current importance of the type I error, looking for the reason for a change, when there was no change in patient vital signs. 

The final proof of any rule is in the using.  Caregivers may try each of the rules to determine the most effective rule for the particular monitoring situation.  The results of this study are applicable for newborns only.  When monitoring adults, the analysis should be replicated by varying the sample sizes and varying the reset rules.  A method that may be used to convert these findings to other types of patients or with newborns using a fixed time monitor is to examine the amount of time the subgroups are out of limits rather than to look at the number of subgroups.

Bibliography

1) Laffel, Glenn, Robert Luttman, and Steven M. Zimmerman "Using Control Charts to Analyze Serial  Patient-Related Data", Quality Management in Health Care, 1994 2(1), p.70-77 Volume 3 Number 1 Fall 1994.


2) Plsek (1992), "Introduction to control charts, Quality Manage Health Case. p.65-73.
3. Zimmerman, Steven M, and Steven Ringer, "Issues in Clinical Monitoring," Computers in Industrial Engineering Vol. 31 No ½, pp. 451-454, 1996.


4) Zimmerman, Steven M., Robert N. Zimmerman, Lonnie D. Brown, and Shannon S. Brown, (1992) "Using Moving Average Process Control Charts in Biomedical Applications," Proceedings- Ninth International Conference of the Israel Society of Quality Assurance, 1992, November 1992, p.761-764.


5) Zimmerman, Steven M., Lonnie D. Brown, Shannon S. Brown, and Richard L Goldhamer, M.D.   (1990), "Quality Control Charts for Patient Data." The 8th International Conference of Israel  Society for Quality Assurance Transactions November 26-29, 1990 Jerusalem.


6) Zimmerman, Steven M., Lonnie Brown, Shannon Brown, and Robert N. Zimmerman (1992),        "Using the Theory of Runs in a Biomedical Application," 46th Annual Quality Control Congress Transactions May 18-20, p.903-908.

Home
Articles about BioMedQC material
Dept-Z@BioMedQC.com

Site hosted by Angelfire.com: Build your free website today!