A Case Study in Monitoring Hospital-Associated Infections with Count Control Charts Shreyas S. Limaye 1 , Christina M. Mastrangelo 1 , Danielle M. Zerr 2 1 Industrial Engineering, University of Washington, Seattle, Washington 2 Children’s Hospital, Seattle, Washington ABSTRACT Hospital-associated infections are a major concern in the medical community due to the potential loss of life and high costs. Monitor- ing the incidences of infections is an established part of quality maintenance programs in hospitals. However, traditional methods of analysis are often inadequate since the incidences of infections are infrequent. In order to address this issue, techniques such as the cumulative sum (CUSUM) chart for counted data and the g-type control chart have been suggested. This article demonstrates how these charts may be applied to infection control surveillance data from Children’s Hospital and makes recommendations for a control chart most suitable for monitoring hospital-associated infec- tions. KEYWORDS g-type control charts, hospital-associated infections, Poisson control charts, SPC, statistical process control, surveillance INTRODUCTION Hospital-associated (HA) infections are any infections that are acquired or spread as a direct result of a patient’s hospital stay. The Centers for Disease Control and Prevention (CDC) estimates that about 2 million people acquire HA infections each year and that about 90,000 of these patients die as a result of their infections. Vulnerable populations such as children or ICU patients are even more susceptible (Health Protection Agency, 2003). The most common HA infections are central line–associated blood stream infections (BSI), ventilator-associated pneumonia (VAP), and catheter- associated urinary tract infections (UTI). Approximately 80,000 BSIs occur in ICUs each year in the United States, and these infections may prolong a hospital stay by 7 to 21 days (Champion and Mabee, 2000). Some of the pro- spective studies indicate up to a 35% increase in mortality due to these infections. The attributable cost per infection is estimated to be $34,000–$56,000, and the annual cost of caring for patients with BSIs ranges from $296 million to $2.3 billion. The incidence of VAP varies greatly, ranging from 6 to 52% of intubated patients depending on patient risk factors (AHRQ, 2001). The cumulative incidence is approximately 1–3% Address correspondence to Shreyas S. Limaye, 5505 15th Ave. NE, Apt. 306, Seattle, WA 98105. E-mail: [email protected]Quality Engineering, 20:404–413, 2008 Copyright # Taylor & Francis Group, LLC ISSN: 0898-2112 print=1532-4222 online DOI: 10.1080/08982110802334120 404
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A Case Study in MonitoringHospital-Associated Infections
with Count Control ChartsShreyas S. Limaye1,
Christina M. Mastrangelo1,
Danielle M. Zerr2
1Industrial Engineering,
University of Washington,
Seattle, Washington2Children’s Hospital, Seattle,
Washington
ABSTRACT Hospital-associated infections are a major concern in the
medical community due to the potential loss of life and high costs. Monitor-
ing the incidences of infections is an established part of quality maintenance
programs in hospitals. However, traditional methods of analysis are often
inadequate since the incidences of infections are infrequent. In order to
address this issue, techniques such as the cumulative sum (CUSUM) chart
for counted data and the g-type control chart have been suggested. This
article demonstrates how these charts may be applied to infection control
surveillance data from Children’s Hospital and makes recommendations
for a control chart most suitable for monitoring hospital-associated infec-
tions.
KEYWORDS g-type control charts, hospital-associated infections, Poisson
control charts, SPC, statistical process control, surveillance
INTRODUCTION
Hospital-associated (HA) infections are any infections that are acquired or
spread as a direct result of a patient’s hospital stay. The Centers for Disease
Control and Prevention (CDC) estimates that about 2 million people acquire
HA infections each year and that about 90,000 of these patients die as a
result of their infections. Vulnerable populations such as children or ICU
patients are even more susceptible (Health Protection Agency, 2003).
The most common HA infections are central line–associated blood stream
infections (BSI), ventilator-associated pneumonia (VAP), and catheter-
associated urinary tract infections (UTI). Approximately 80,000 BSIs occur
in ICUs each year in the United States, and these infections may prolong a
hospital stay by 7 to 21 days (Champion and Mabee, 2000). Some of the pro-
spective studies indicate up to a 35% increase in mortality due to these
infections. The attributable cost per infection is estimated to be
$34,000–$56,000, and the annual cost of caring for patients with BSIs ranges
from $296 million to $2.3 billion. The incidence of VAP varies greatly,
ranging from 6 to 52% of intubated patients depending on patient risk
factors (AHRQ, 2001). The cumulative incidence is approximately 1–3%
Address correspondence to Shreyas S.Limaye, 5505 15th Ave. NE, Apt. 306,Seattle, WA 98105. E-mail:[email protected]
Quality Engineering, 20:404–413, 2008Copyright # Taylor & Francis Group, LLCISSN: 0898-2112 print=1532-4222 onlineDOI: 10.1080/08982110802334120
404
per day of intubation. Overall, VAP is associated with
an attributable mortality of up to 30%. The average
cost per episode of VAP is estimated at $3000 to
$6000, and the additional length of stay for patients
who develop VAP is estimated at 13 days (Warren
et al., 2003). Urinary tract infections (UTIs) account
for about 40% of the total number of HA infections
in some reports and affect an estimated 600,000
patients per year (MMWR, 2002). The average cost
of one hospital-associated UTI is estimated to be
between $680 and $1,875 per patient infection, and
additional hospital days per patient due to UTIs
range between 1 to 4 days.
The problem of HA infections is quite significant
in terms of affecting patient lives, adding to the econ-
omic cost of the healthcare, and putting additional
strain on the hospital resources. Effective monitoring
of infection rates can alert clinicians to a change of
infection rates, prompt the quality improvement
teams to identify causes behind the abnormal
increase, and stimulate efforts to look for effective
interventions to reduce them. A control chart is an
effective tool for this.
The control chart is a running record of behavior
over time (Carey, 2002) and usually has one of three
goals: (1) reduce variation, that is, process improve-
ment; (2) signal the need for a process adjustment;
and (3) demonstrate stability. Control charts can
directly address goals 2 and 3. In order to accomplish
goal 1, they can provide useful pointers if used
appropriately. The use of control charts is increas-
ingly being suggested for a variety of applications
in healthcare in an effort to improve the quality of
healthcare delivery. Components of variability exhib-
ited by healthcare data make them attractive candi-
dates to apply control charting techniques (Matthes
et al., 2007). Woodall (2006) gives an excellent sum-
mary of various types of control charts in healthcare
monitoring and in public health surveillance, as well
as discussing the issues related to these charts.
The use of control charts is also widely used for
monitoring infections in an effort to improve patient
safety. Benneyan (1998) reasons that the use of SPC
in other fields that is, understanding current process
performance, achieving a consistent level of process
quality, monitoring for process deterioration and
reducing process variation, is very much applicable
to the case of monitoring infections as well. In order
to address some of the concerns of traditional control
charts in this setting, alternate charts have been
suggested. Gustafson (2000) suggests the use of
risk-adjusted control charts based on a standardized
infection ratio calculated by dividing the observed
number of infections by the expected number of
infections during a particular period. Benneyan
(2001a) develops the g-type and h-type control
charts based on inverse sampling from geometric
and negative binomial distributions for evaluating
number of cases or number of days between
HA infections as they can exhibit greater detection
power over conventional binomial-based approa-
ches. Morton et al. (2001) demonstrate use of
counted-data EWMA and CUSUM charts for effective
monitoring of hospital-associated infections.
This article compares different control charting
techniques for monitoring hospital-associated infec-
tions. It demonstrates a u-chart, a CUSUM chart for
low count data, and g-type control chart. The next
section describes the data, the application of control
charts to the data, and the merits of these charts. The
following section concludes by providing recom-
mendations regarding use of a suitable control chart-
ing technique for monitoring hospital-associated
infections based on this case study.
DATA DESCRIPTION
Five years of infection surveillance data (from
2002 to 2006) from the pediatric ICU of the Children’s
Hospital, Seattle, is used. The data represents the
total number of HA infections as well as ‘‘patient
days,’’ which is the total number of days patients
spent in the ICU that quarter. Quarterly data on three
of the main infection types, that is, catheter-related
where �xx¼ the average number of days between infections, p¼ therate of infection (if known), and k¼ the number of standard deviationsused in the control limits.
S. S. Limaye et al. 408
2005–2006 are approximately the same, infections
occur at more regular time intervals after December
2004 (as evidenced by the lower number of spikes).
A g-type control chart has an advantage in that
each data point can be plotted immediately, and the
overall effects can be seen quickly as well. Figure 8
shows g-type charts for BSI, UTI, VAP, and total
infections from 2003 to 2006. The plots indicate a
slight decrease in the number of BSIs in 2006 and
an increase in the days between infections. For
2006, UTIs and VAPs show a decrease in frequency
and increased days between occurrences.
CONCLUSIONS
This case study illustrates how control charting is
used currently in the pediatric ICU at the Children’s
Hospital and then demonstrates the use of different
techniques for the infection surveillance data. The
u-chart monitors the number of infections per 1000
FIGURE 7 G-type control chart for number of days between HA-infection incidences.
FIGURE 8 G-type control charts for overall infections, BSI, UTI, and VAP infections.
409 Monitoring Hospital-Associated Infections
patient days. The counted data CUSUM chart does
not use denominators and monitors the number of
infections per month directly. The g-type control
chart monitors the number of days between interven-
tions. Each chart has a set of advantages and
disadvantages.
The u-chart is simple to construct, capable of hand-
ling low-count data, and easy to interpret. It also takes
into account the change in the sample size which in
case is the varying number of patient days, line days
or ventilator days. In Figure 9, the quarterly based
u-chart did not detect a change in infection rates as
would be expected since the control chart was
developed from the data being monitored. A u-chart
for monthly data may provide more information.
However as Figure 10 shows, such a chart for rare
infections like UTI or VAP would not be very useful
since many months would have infection rates of 0.
The counted data CUSUM chart does provide an
early warning and can track small changes effec-
tively. However, it is slightly more complicated to
build, understand, and explain. Moreover, it is very
sensitive to the selection of k and h parameter values,
which affects the usability of this chart. Also, as one
can see in Figure 5, the CUSUM chart shows out-of-
control signals only in February and April 2006 while
actually the months preceding February 2006 have
higher infection rates. This is counterintuitive to the
user and not the preferred option.
The g-type control chart is simple to construct; it
can quickly point out the years in which the number
of infections is low; and it can quickly indicate long
periods having no infection occurrences. However,
the g-charts are not very helpful in detecting
increased rates of infection.
Children’s Hospital can certainly improve their
infection monitoring practices. The simplest and
effective way to achieve that would be the use of
u-chart as it would take into account the changing
number of patient, line, and ventilator days. For
FIGURE 9 U-type control charts for quarterly rates of BSI, UTI, and VAP infections.
S. S. Limaye et al. 410
monitoring overall infection rates and BSI, a u-chart
based on monthly infection rates would be more
effective, whereas for UTI and VAP, a u-chart based
on quarterly infection rates would be more suitable.
One of the disadvantages for the control charts
discussed is that they do not account for the differ-
ences between patients. Patients differ on their sever-
ity of illness, so it is expected that this variability
affects a patient’s likelihood of contracting a hospi-
tal-acquired infection. Risk-adjusted control charts
aim to address this issue by accounting for the vul-
nerability of a patient to infections (Alemi and Oliver,
2001). The vulnerability is established by key clinical
findings, diagnosis codes, or by consensus among
clinicians. The expected rate of infection is calcu-
lated using logistic regression to determine the
expected probability of an infection for each individ-
ual patient and then averaging these to calculate the
expected rate of infection for the period. The control
chart monitors the observed rate of infections versus
their expected rate.
In practice, an index to account for patient con-
dition is difficult to determine because the existing
methods are subjective. Unless a credible system to
compute the vulnerability of a patient for the specific
infection is developed, it may not be practical to use
risk-adjusted control charts. The data used in this
article do not provide information regarding the
vulnerability of each patient for different type of
infection so it is not possible to include this chart
in this case study or use it in most healthcare applica-
tions either.
ABOUT THE AUTHORS
Shreyas S. Limaye is a doctoral candidate at the
University of Washington, Seattle, and Advanced
Process Control Engineer at the Hitachi Global
Storage Technology, San Jose, CA. His professional
and research interests include applications of com-
plex system modeling, statistical analysis and process
control in healthcare, semiconductor manufacturing
FIGURE 10 U-type control charts for monthly rates of BSI, UTI, and VAP infections.
411 Monitoring Hospital-Associated Infections
and transportation logistics. He is a member of
INFORMS and IIE.
Dr. Christina M. Mastrangelo is an Associate
Professor of industrial engineering at the University
of Washington. She holds a BS, MS and PhD
degrees in Industrial Engineering from Arizona State
University. Prior to joining UW in 2002, she was an
Associate Professor of systems and information
engineering at the University of Virginia.
Dr. Danielle M. Zerr is an Associate Professor of
pediatric infectious diseases at the University of
Washington and the Medical Director of infection
control at Seattle Children’s Hospital.
REFERENCES
AHRQ. (2001). Making Health Care Safer: A Critical Analysis of PatientSafety Practices. Rockville, MD: Agency for Healthcare Research andQuality. Evidence Report=Technology Assessment: Number 43, AHRQPublication No. 01-E058. Available at http://www.ahrq.gov/clinic/ptsafety/, accessed August 2008.
Alemi, F., Oliver, D. W. (2001). Tutorial on risk-adjusted P-charts. QualityManagement in Health Care, 10:1–9.
Alt, F. (2006). Multivariate Quality Control. Encyclopedia of StatisticalSciences, 1–12. Available at http://www.mrw.interscience.wiley.com/emrw/9780471667193/ess/titles?filter¼M, accessed August 2008.
Benneyan, J. (1998). Statistical quality control methods in infection con-trol and hospital epidemiology, part I: Introduction and basic theory.Infection Control and Hospital Epidemiology, 19(3):194–214.
Benneyan, J. (2001a). Number-between g-type statistical control chartsfor monitoring adverse events. Health Care Management Science,4:305–318.
Benneyan, J. (2001b). Performance of number-between g-type statisticalcontrol charts for monitoring adverse events. Health Care Manage-ment Science, 4:319–336.
Benneyan, J. (2001c). Design, use, and performance of statistical controlcharts for clinical process improvement. Available at http://www1.coe.
neu.edu/~benneyan/papers/intro_spc.pdf, accessed August 2008.Carey, R. G. (2002). How do you know that your care is improving? Part II:
Using control charts to learn from your data. Journal of AmbulatoryCare Management, 25:78–88.
Champion, H., Mabee, M. (2000). The coalition for American trauma care.Washington Report, 2(6):2–3. Available at http://www.traumacoalition.org/Washingtonreport.pdf, accessed May 2006.
Gustafson, T. L. (2000). Practical risk-adjusted quality control charts forinfection control. American Journal of Infection Control, 28(6):406–414.
Health Protection Agency. (2003). General information—Healthcare-associated infections. Available at http://www.camr.org.uk/webw/HRAweb/HPAweb/&HPAwebStandard/HPAweb_C/1195733802872?p¼1191942126527, accessed August 2008.
Lucas, J. (1985). Counted data CUSUM’s. Technometrics, 27(2):129–144.
Matthes, M., Ogunbo, S., Pennington, G., Wood, N., Hart, M., Hart, R.(2007). Statistical process control for hospitals: Methodology, usereducation, and challenges. Quality Management in Health Care,16(3):205–214.
Montgomery, D. (2005). Introduction to Statistical Quality Control.New York: Wiley.
Morbidity and Mortality Weekly Report. (2002). Guideline for preventionof catheter-associated urinary tract infections, Centers for DiseaseControl and Prevention. Available at http://www.cdc.gov/ncidod/dhqplgl_catheter_assoc.html, accessed August 2008.
Morton, A., Whitby, M., McLaws, M., Dobson, A., McElwain, S., Looke, D.,Stockelroth, J., Sartor, A., (2001). The application of statistical processcontrol charts to detection and monitoring of hospital-acquired infec-tions. Journal of Quality in Clinical Practise, 21(4): 112–117.
National Nosocomial Infections Surveillance Systems Report. (2004). http://www.cdc.gov/ncidod/dhqp/pdf/nnis/2004NNISreport.pdf, accessedAugust 2008.
Smyth, E. T. M., Emmerson, A. M. (2000). Surgical site infection surveil-lance. Journal of Hospital Infection, 45:173–184.
Warren, D. K., Shukla, S. J., Olsen, M. A., Kollef, M. H., Hollenbeak, C. S.,Cox, M. J., Cohen, M. M., Fraser, V. J. (2003). Outcome and attribu-table cost of ventilator associated pneumonia among intensive careunit patients in a suburban medical center. Critical Care Medicine,31(5):1312–1317.
Woodall, W. (2000). Controversies and contradictions in statistical pro-cess control. Journal of Quality Technology, 32(4):341–350.
Woodall, W. (2006). The use of control charts in health-care and public-health surveillance. Journal of Quality Technology, 38(2):89–104.
APPENDIX
TABLE A1 Total Number of Infections and ICU Patient Days
Quarter
# of
Infections
Patient
days
Infections per 1000
patient days Quarter
# of
Infections
Patient
days
Infections per 1000
patient days
Q1 2002 8 1451 5.51 Q3 2004 17 1526 11.14
Q2 2002 4 1207 3.31 Q4 2004 18 1543 11.67
Q3 2002 10 1372 7.29 Q1 2005 20 1629 12.28
Q4 2002 6 1413 4.25 Q2 2005 24 1453 16.52
Q1 2003 21 1481 14.18 Q3 2005 20 1552 12.89
Q2 2003 12 1256 9.55 Q4 2005 20 1519 13.17
Q3 2003 17 1275 13.33 Q1 2006 19 1934 9.82
Q4 2003 22 1348 16.32 Q2 2006 18 1950 9.23
Q1 2004 13 1432 9.08 Q3 2006 16 2066 7.74
Q2 2004 16 1596 10.03 Q4 2006 15 1857 8.08
S. S. Limaye et al. 412
TABLE A3 Total Number of UTI and ICU Foley Catheter Days
Quarter # of UTI
Foley
catheter days
UTI per 1000
Foley catheter days Quarter # of UTI
Foley
catheter days
UTI per 1000
Foley catheter days
Q1 2002 1 778 1.28 Q3 2004 1 718 1.39
Q2 2002 1 471 2.12 Q4 2004 3 848 3.53
Q3 2002 3 625 4.80 Q1 2005 2 921 2.17
Q4 2002 1 657 1.52 Q2 2005 4 844 4.73
Q1 2003 5 801 6.24 Q3 2005 6 873 6.87
Q2 2003 2 659 3.03 Q4 2005 4 768 5.20
Q3 2003 2 689 2.90 Q1 2006 1 752 1.32
Q4 2003 5 757 6.60 Q2 2006 3 624 4.80
Q1 2004 2 599 3.33 Q3 2006 1 808 1.23
Q2 2004 2 708 2.82 Q4 2006 0 512 0
TABLE A4 Total Number of VAP and ICU Ventilator Days
Quarter # of VAP
Ventilator
days
VAP per 1000
ventilator days Quarter # of VAP
Ventilator
days
VAP per 1000
ventilator days
Q1 2002 0 586 0 Q3 2004 2 713 2.80
Q2 2002 0 405 0 Q4 2004 1 766 1.30
Q3 2002 1 692 1.44 Q1 2005 0 1023 0
Q4 2002 0 594 0 Q2 2005 0 807 0
Q1 2003 1 840 1.19 Q3 2005 0 913 0
Q2 2003 0 480 0 Q4 2005 0 739 0
Q3 2003 1 602 1.66 Q1 2006 1 749 1.33
Q4 2003 2 703 2.84 Q2 2006 0 987 0
Q1 2004 1 571 1.75 Q3 2006 1 973 1.02
Q2 2004 3 724 4.14 Q4 2006 2 722 2.77
TABLE A2 Total Number of BSI and ICU Central Line Days