M A J O R A R T I C L E
The Impact of ICD-9-CM Code Rank Order on
the Estimated Prevalence of Clostridium difficile
Erik R. Dubberke,1Anne M. Butler,1Humaa A. Nyazee,1Kimberly A. Reske,1Deborah S. Yokoe,2Jeanmarie Mayer,3
Julie E. Mangino,4Yosef M. Khan,4Victoria J. Fraser,1and Centers for Disease Control and Prevention Epicenters
1Department of Medicine, Washington UniversitySchool of Medicine, St Louis, Missouri;2Department of Medicine, Brigham and Women's Hospital and
Harvard Medical School, Boston, Massachusetts;3Department of Medicine, University of Utah Hospital, Salt Lake City, Utah; and4Department of
Medicine, The Ohio State University Medical Center, Columbus, Ohio
Classification of Disease, NinthRevision,ClinicalModification (ICD-9-CM) diagnosis codes. Whether ICD-9-CM code
rank order affects CDI prevalence estimates is important because the National Hospital Discharge Survey (NHDS)
and the Nationwide Inpatient Sample (NIS) have varying limits on the number of ICD-9-CM codes collected.
Methods.ICD-9-CM codes for CDI (008.45), C. difficile toxin assay results, and dates of admission and
discharge were collected from electronic hospital databases for adult patients admitted to 4 hospitals in the United
States from July 2000 through June 2006. CDI prevalence per 1000 discharges was calculated and compared for
NHDS and NIS limits and toxin assay results from the same hospitals. CDI prevalence estimates were compared
using the v2test, and the test of equality was used to compare slopes.
Results. CDI prevalence measured by NIS criteria was significantly higher than that measured using NHDS
criteria (10.7 cases per 1000 discharges versus 9.4 cases per 1000 discharges; P , .001) in the 4 hospitals. CDI
prevalence measured by toxin assay results was 9.4 cases per 1000 discharges (P 5 .57 versus NHDS). However, the
CDI prevalence increased more rapidly over time when measured according to the NHDS criteria than when
measured according to toxin assay results (b5 1.09 versus 0.84; P 5 .008).
Conclusions.Compared with the NHDS definition, the NIS definition captured 12% more CDI cases and
reported significantly higher CDI rates. Rates calculated using toxin assay results were not different from rates
calculated using NHDS criteria, but CDI prevalence appeared to increase more rapidly when measured by NHDS
criteria than when measured by toxin assay results.
US estimates of the Clostridium difficile infection (CDI) burden have utilized International
The incidence and severity of Clostridium difficile in-
fection (CDI) havebeen increasingin recentyears [1–7],
but national surveillance efforts and interhospital
comparisons have been limited by the lack of a standard
CDI surveillance system. As a result, the International
Classification of Disease, Ninth
Modification (ICD-9-CM) codes assigned at hospital
discharge have been used as a proxy to estimate CDI
prevalence in the United States [8–12].
Using administrative data (ICD-9-CM codes) to
advantages. Administrative discharge data are in-
expensive to obtain, are systematically collected, and
utilize a single ICD-9-CM code to designate CDI
(008.45), thus potentially providing a nationally repre-
sentative method for tracking CDI rates [8, 13, 14]. Two
administrative databases have been used to estimate CDI
prevalence in the United States: the National Hospital
Received 29 November 2010; accepted 15 March 2011.
Presented in part: 47th Annual Meeting of the Infectious Diseases Society of
America; Philadelphia, PA, 29 October–1 November 2009. Abstract number 421.
Correspondence: Erik R. Dubberke, MD, MSPH, Division of Infectious Diseases,
Washington University School of Medicine, Box 8051, 660 S Euclid, St Louis, MO
Clinical Infectious Diseases
? The Author 2011. Published by Oxford University Press on behalf of the Infectious
Diseases Society of America. All rights reserved. For Permissions, please e-mail:
d CID 2011:53 (1 July)
d Dubberke et al
Discharge Survey (NHDS) and the Nationwide Inpatient Sample
(NIS). ICD-9-CM code data are collected differently in each da-
tabase; neither data set collects all potential ICD-9-CM codes
assigned at hospital discharge. The NHDS is collected annually by
the NationalCenterforHealthStatisticsat the Centers for Disease
Control and Prevention (CDC), and about 90% of a panel of 500
hospitals participate . The NIS, on the other hand, is collected
from all states that participate in the Healthcare Cost and Utili-
zation Project and includes information from about 1000 US
hospitals . The NHDS collects up to the first 7 ICD-9-CM
codes assigned per patient, and the NIS collects up to the first 9–
15 codes assigned per patient (the number of codes captured
varies by state).
Despite their advantages, a recent study suggests that ICD-9-
CM codes may not be acceptable for hospital-onset CDI sur-
veillance. A multicenter study performed at the CDC Pre-
vention Epicenters hospitals compared the incidences of
hospital-onset CDI as measured byICD-9-CM codes with toxin
assay results. ICD-9-CM codes overestimated the incidence of
hospital-onset CDI, compared with toxin assay results, in-
dicating that ICD-9-CM codes are not an acceptable surrogate
for hospital-onset CDI surveillance . The degree to which
ICD-9-CM codes overreported the incidence of hospital-onset
CDI varied by year and by hospital, indicating that ICD-9-CM
codes would not have been useful for intrahospital or inter-
hospital CDI surveillance.
No previously published CDI prevalence study using ICD-9-
CM codes has differentiated between the NHDS and NIS
criteria. The data collection discrepancies between these 2 data
sets may account for the conflicting results of previous studiesof
CDI prevalence. It is not known how the differences in the
number of ICD-9-CM codes available for prevalence estimation
affect the estimated CDI prevalence and how CDI burden
estimates based on ICD-9-CM codes compare with toxin assay
results. Therefore, we investigated how the NHDS and NIS
criteria affect CDI prevalence and how these definitions com-
pare with toxin assay results at multiple healthcare facilities
during a 6-year study period.
The study population included all adult patients admitted to 4
US hospitals participating in the CDC Epicenters Program from
1 July 2000 through 30 June 2006. These hospitals included
Barnes-Jewish Hospital (St Louis, Missouri), Brigham and
Women’s Hospital (Boston, Massachusetts), Ohio State Uni-
versity Medical Center (Columbus, Ohio), and University
Hospital (Salt Lake City, Utah). Patients aged $18 years were
included in our analyses. During the study period, 3 of the 4
laboratories at the study hospitals rejected formed stool speci-
mens for C. difficile testing.
Dates of hospital admission and discharge, rank order of CDI
ICD-9-CM code, and positive C. difficile toxin assay results were
obtained from hospital databases. The toxin test assays used at
the 4 hospitals during the study period are as follows: Hospital A
used the cytotoxcity assay until June 2002 and then switched to
an enzyme immunoassay; hospital B used an enzyme immu-
noassay; and hospitals C and D used the cytotoxicity assay. For
upto thefirst 7diagnosesandthe NIScollects data upto thefirst
15 diagnoses. The number of ICD-9-CM diagnoses collected for
the NIS varies by state and year (eg, 15 in Massachusetts, Mis-
souri, and Ohio; 9 in Utah from 2000 to 2002 and 10 in Utah
from 2003 to 2006). The institutional review boards at the CDC
and all participating hospitals gave approval for this study.
Rates of CDI cases per 1000 discharges were calculated and
compared for the overall administrative data set, NHDS and NIS
criteria, and toxin assay results. Discharges, rather than patient-
days, were used for the denominator because the date of CDI
onset was not known and ICD-9-CM codes are assigned at dis-
and slopes were compared using the test of equality. Calculation
of j statistics was performed to measure the agreement between
C. difficile toxin assay results and ICD-9-CM codes. Statistical
analyses were performedwithEpiInfo, version6(CDC);SPSSfor
Windows, version 17.0 (SPSS), and Stata, version 9.2.
We identified a total of 10,832 cases of CDI, of which 2925
(27%) had an ICD-9-CM code alone, 1643 (15.2%) had a posi-
tive toxin assay result alone, and 6264 (57.8%) had both the
ICD-9-CM code and positive toxin assay result. The overall CDI
prevalence of all ICD-9-CM codes (ie, ICD-9-CM code in any
position) was 10.9 per 1000 discharges, and the median rank-
order of the ICD-9-CM code was 4. Compared with the total
CDI cases captured by ICD-9-CM codes in any position, the
NIS and NHDS criteria captured 99% (n 5 9056) and 87%
(n 5 7978), respectively (Figure 1).
The overall CDI prevalence calculated with the NIS criteria
(10.7 per 1000 discharges) was significantly higher than the CDI
prevalence calculated with the NHDS criteria (9.4 per 1000
discharges; P , .001) but was no different from the CDI prev-
alence of all ICD-9-CM codes (10.9 per 1000 discharges;
P 5 .33). The CDI prevalence measured by means of toxin assay
results (9.4 per 1000 discharges) was no different from the CDI
prevalence measured by NHDS (P 5 .57). The agreement be-
tween the NHDS criteria and toxin assay was good, with an
overall j value of 0.638, and hospital-specific j values ranging
from 0.560 to 0.702.
The Impact of ICD-9-CM
d CID 2011:53 (1 July)
Figures 2 and 3 presentannualCDI prevalence by surveillance
definition, overall and stratified by hospital, respectively.
Overall, the CDI prevalence by means of the NIS criteria was the
highest across the study period (Figure 2). The median rank
order of the ICD-9-CM code for CDI was 3 at hospitals A and B
and 4 at hospitals C and D. Hospital B is the only hospital where
the annual CDI prevalence was highest by means of toxin assay
results during every year of the study (Figure 3). The toxin assay
rate was the highest rate only at Hospital B (Figure 3). Com-
pared with the NHDS criteria, the toxin assay rate was higher at
B (toxin assay rate, 12.0 cases per 1000 discharges versus NHDS
rate, 9.4 cases per 1000 discharges; P , .001), whereas the toxin
assay rate was lower at hospital C (toxin assay rate, 4.3 cases per
1000 discharges versus NHDS rate, 5.3 cases per 1000 discharges;
P , .001) and hospital D (toxin assay rate, 6.9 cases per 1000
discharges versus NHDS rate, 8.7 cases per 1000 discharges; P ,
Overall, ICD-9-CM codes overestimated the number of cases
of CDI relative to use of the toxin assay (Figure 2). While the
overall annual rates increased almost every year of the study
period regardless of the surveillance definition, the annual
increase in the prevalence of CDI varied by definition. The
annual increase in prevalence according to the NIS criteria was
greater than that according to the NHDS criteria (b 5 1.34
versus b 5 1.09; P 5 .003). The annual increase in prevalence
according to the NHDS criteria was greater than that according
to the toxin assay (b 5 1.09 versus b 5 0.84; P 5 .008).
To our knowledge, this is the first study to examine how current
methods used to estimate the prevalence of CDI in the United
States compare with the use of toxin assay results to estimate the
prevalence of CDI. The results of this multicenter study suggest
that the CDI prevalence measured using all ICD-9-CM codes
was higher than the CDI prevalence measured using the NIS,
which, in turn, was higher than the CDI prevalence measured
using the NHDS. These results were not surprising, given that
the NIS captures more ICD-9-CM codes than the NHDS but
does not capture all ICD-9-CM codes. In this study, the overall
CDI prevalence measured by the NHDS criteria was the same as
the CDI prevalence measured by the positive toxin assay results.
Previous research indicates that ICD-9-CM codes overestimate
CDI prevalence [16–18]. NHDS, by limiting the data set to the
first 7 ICD-9-CM codes in each discharge record, may eliminate
patients who do not truly have CDI. However, in this study, the
annual increase in CDI prevalence as measured by the NHDS
criteria was greater than that revealed by positive toxin assay
results, indicating at some point that the CDI prevalence iden-
tified by the NHDS criteria may become greater than the CDI
prevalence identified by toxin assay results.
When comparing the CDI prevalence measured using the
NHDS criteria with the toxin assay results at the hospital level,
the CDI prevalence was higher by toxin assay results at hospitals
A and B, whereas the prevalence was higher by the NHDS cri-
teria at hospitals C and D. Hospital A’s results may be due in
part to the fact that this institution became more vigilant to
ensure that all medical records were adequately reviewed by
medical coders in a timely fashion midway through the study
period. Hospital B’s results may reflect laboratory practices at
this institution. During the study period, this hospital’s
Classification of Diseases, Ninth Revision, Clinical Modification code for
Clostridium difficile infection identified by National Hospital Discharge
Survey (NHDS) and Nationwide Inpatient Sample (NIS) limits.
Proportion of patients who were assigned the International
surveillance definition. ICD-9-CM, International Classification of Dis-
eases, Ninth Revision, Clinical Modification; NHDS, National Hospital
Discharge Survey; NIS, Nationwide Inpatient Sample.
Yearly overall Clostridium difficile infection rates by
d CID 2011:53 (1 July)
d Dubberke et al
microbiology laboratory tested formed stools for C. difficile
toxin.Thispracticeis discouragedbecause testing asymptomatic
patients may falsely elevate the CDI prevalence by 2 mecha-
nisms: asymptomatically colonized patients without CDI can
have positive toxin assay results, and testing for C. difficile in
low-prevalence populations will increase the number of false-
positive test results [19, 20]. This may explain the higher CDI
prevalence by toxin assay results than that by the NHDS criteria
at this institution. Last, the NIS and NHDS databases do not
exclude on the basis of stool consistency either. Therefore, it was
unknown what effect this might have on our results.
There are limitations to the use of administrative data for
disease surveillance purposes. The ICD-9-CM codes are assigned
by medical coders. Not all medical coders have the same level of
training and certification, which may result in variable coding
to receive an ICD-9-CM code, the diagnosis must be clearly
stated in the medical records by a treating physician. Additional
Patients who receive the ICD-9-CM code for CDI but who do
not have laboratory confirmation frequently have a history of
CDI but lack ongoing symptoms of CDI [17, 18]. Furthermore,
ICD-9-CM codes areassigned after discharge, creatinga time lag
in the availability of data, and ICD-9-CM codes do not provide
any information about date or place of onset of CDI. Therefore,
ICD-9-CM codes alone are not ideal for CDI incidence sur-
Despite the limitations of ICD-9-CM codes, there are lim-
itations to the use of laboratory results on C. difficile toxin
tests for CDI surveillance as well. The ‘‘gold standard’’ to
detect pathogenic C. difficile from stool, toxigenic culture, is
labor and resource intensive and takes several days until re-
sults are final. As a result, there are an increasing number of
methods and algorithms to detect C. difficile or its toxins in
stool, all of which differ in sensitivity and specificity. Stool
handling and processing can also affect the sensitivity and
of Diseases, Ninth Revision, Clinical Modification; NHDS, National Hospital Discharge Survey; NIS, Nationwide Inpatient Sample.
Yearly hospital Clostridium difficile infection rates by surveillance definition at hospitals A, B, C, and D. ICD-9-CM, International Classification
The Impact of ICD-9-CM
d CID 2011:53 (1 July)
specificity of an assay. Testing practices vary in interpretation
of positive results. Although this practice is uncommon, the
diagnoses for some patients are made by means of endoscopy
alone . Indiscriminate repeated testing for C. difficile can
falsely elevate CDI incidence by as much as 27% . Most
importantly, CDI is a clinical diagnosis. Testing stool samples
obtained from patients who do not have clinical symptoms
compatible with CDI will result in positive test results for
patients without CDI. In addition, the NIS and NHDS databases
do not exclude patients with recurrent disease or patients with
repeated toxin assay tests. To keep comparisons consistent, we
did not exclude these patients either. As a result, this may have
overestimated the true CDI prevalence by toxin assay.
An alternative CDI surveillance system already in use is that of
the National Healthcare Safety Network (NHSN), which has
been augmented by mandatory C. difficile public reporting re-
quirements of many states in the United States. Currently, the
NHSN is collecting data on C. difficile using 2 different reporting
methods: (1) infection surveillance and (2) laboratory-identified
events . To date, 166 facilities are participating in the NHSN
C. difficile infection surveillance reporting, 576 facilities are
participating in the C. difficile laboratory-identified event re-
porting, and 36 facilities are participating in both reporting
methods (D. Sievert, PhD, personal communication, 4 June
2010). Review of data from the 36 facilities performing both
methods of surveillance will be important to further our un-
derstanding as to whether use of laboratory data alone in the
absence of clinical information from facilities that do not test
This study indicates that current estimates of CDI preva-
lence in the United States based on ICD-9-CM codes may be
falsely elevated. Fortunately, the NHSN is currently collecting
data for CDI surveillance. The NHSN system provides
a standardized method of CDI surveillance and will be able to
assess the utility of laboratory-based CDI surveillance. Thus,
the NHSN system may represent a substantial improvement in
the quality of data available for hospital-based CDI surveil-
lance, national CDI prevalence estimates, and interhospital
CDI prevalence comparisons.
Findings and conclusions in this report are those of the authors and do
not necessarily represent the official position of the Centers for Disease
Control and Prevention.
Financial support. This work was supported by grants from the
Centers for Disease Control and Prevention (UR8/CCU715087-06/1 and
5U01C1000333 to Washington University, 5U01CI000344 to Eastern
Massachusetts, 5U01CI000328 to The Ohio State University, and
5U01CI000334 to the University of Utah) and from the National Institute
of Allergy and Infectious Diseases (K23AI065806).
Potential conflicts of interest.V. J. F. holds stock options in Express
Scripts. J. M. has received institutional financial support for a lecture on
candidemia from Fallon Medical CME. E. R. D. has received payment from
Optimer,Merck, Pfizer,Steris,BD,and Merck forconsultancy; paymentfor
lectures from Schering-Plough; and payment for development of educa-
tional presentations from the Robert Michaels Educational Institute. His
institution has received grant support on his behalf from Optimer and
Merck. All other authors: no conflicts.
All authors have submitted the ICMJE Form for Disclosure of Potential
Conflicts of Interest. Conflicts that the editors consider relevant to the
content of the manuscript have been disclosed in the Acknowledgments
1. Dallal RM, Harbrecht BG, Boujoukas AJ, et al. Fulminant Clostridium
difficile: an underappreciated and increasing cause of death and com-
plications. Ann Surg 2002; 235:363–72.
2. Muto CA, Pokrywka M, Shutt K, et al. A large outbreak of Clostridium
difficile-associated disease with an unexpected proportion of deaths
and colectomies at a teaching hospital following increased fluo-
roquinolone use. Infect Control Hosp Epidemiol 2005; 26:273–80.
3. Pepin J, Valiquette L, Alary ME, et al. Clostridium difficile-associated
diarrhea in a region of Quebec from 1991 to 2003: a changing pattern
of disease severity. CMAJ 2004; 171:466–72.
4. Pepin J, Saheb N, Coulombe MA, et al. Emergence of fluoroquinolones
as the predominant risk factor for Clostridium difficile-associated di-
arrhea: a cohort study during an epidemic in Quebec. Clin Infect Dis
variant strain of Clostridium difficile. N Engl J Med 2005; 353:2433–1.
6. Archibald LK, Banerjee SN, Jarvis WR. Secular trends in hospital-
acquired Clostridium difficile disease in the United States, 1987–2001.
J Infect Dis 2004; 189:1585–9.
7. Loo VG, Poirier L, Miller MA, et al. A predominantly clonal multi-
institutional outbreak of Clostridium difficile-associated diarrhea with
high morbidity and mortality. N Engl J Med 2005; 353:2442–9.
8. McDonald LC, Owings M, Jernigan DB. Clostridium difficile infection
in patients discharged from US short-stay hospitals, 1996–2003. Emerg
Infect Dis 2006; 12:409–15.
9. Riccardi R, Rothenberger DA, Madoff RD, Baxter NN. Increasing
prevalence and severity of Clostridium difficile colitis in hospitalized
patients in the United States. Arch Surg 2007; 142:624–31.
10. O’Brien JA, Lahue BJ, Caro JJ, Davidson DM. The emerging infectious
challenge of Clostridium difficile-associated disease in Massachusetts
hospitals: clinical and economic consequences. Infect Control Hosp
Epidemiol 2007; 28:1219–7.
11. Zilberberg MD, Shorr AF, Kollef MH. Increase in adult Clostridium
difficile-related hospitalizations and case-fatality rate, United States,
2000–2005. Emerg Infect Dis 2008; 14:929–31.
12. Schmiedeskamp M, Harpe S, Polk R, Oinonen M, Pakyz A. Use of
International Classification of Diseases, Ninth Revision, Clinical Modifi-
cation codes and medicationusedatatoidentifynosocomialClostridium
difficile infection. Infect Control Hosp Epidemiol 2009; 30:1070–6.
13. Iezzoni L. Coded data from administrative sources. In: Iezzoni L, ed.
Risk adjustment for measuring health care outcomes. 3rd ed. Chicago,
IL: Health Administration Press, 2009: 83–138.
14. Klabunde CN, Warren JL, Legler JM. Assessing comorbidity using
claims data: an overview. Med Care 2002; 40(suppl 8):26–35.
15. Healthcare Cost and Utilization Project. Introduction to the HCUP
Nationwide Inpatient Sample (NIS). Available at: http://www.hcup-
us.ahrq.gov. Published 2010. Accessed 30 August 2010.
16. Dubberke ER, Butler AM, Yokoe DS, et al. Multicenter study of
surveillance for hospital-onset Clostridium difficile infection by the use
of ICD-9-CM diagnosis codes. Infect Control Hosp Epidemiol 2010;
17. Scheurer DB, Hicks LS, Cook EF, Schnipper JL. Accuracy of ICD-9
coding for Clostridium difficile infections: a retrospective cohort. Epi-
demiol Infect 2007; 135:1010–3.
d CID 2011:53 (1 July)
d Dubberke et al
18. Dubberke ER, Reske KA, McDonald LC, Fraser VJ. ICD-9 codes and Download full-text
surveillance for Clostridium difficile-associated disease. Emerg Infect
Dis 2006; 12:1576–9.
19. Gerding DN, Brazier JS. Optimal methods for identifying Clostridium
difficile infections. Clin Infect Dis 1993; 16:S439–42.
20. Litvin M, Reske KA, Mayfield J, et al. Identification of a pseudo-
outbreak of Clostridum difficile infection (CDI) and the effect of
repeated testing, sensitivity, and specificity on perceived prevalence of
CDI. Infect Control Hosp Epidemiol 2009; 30:1166–71.
21. Centers for Disease Control and Prevention. Multidrug-Resistant
Organism (MDRO) and Clostridium difficile-Associated Disease
wc_MDRO_CDAD_ISlabID.html. Published 2008. Accessed 8 June
The Impact of ICD-9-CM
d CID 2011:53 (1 July)