Use of International Classification of Diseases, Ninth Revision, Clinical Modification Codes and Medication Use Data to Identify Nosocomial Clostridium difficile Infection
Objective. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) code for Clostridium difficile infection (CDI) is used for surveillance of CDI. However, the ICD‐9‐CM code alone cannot separate nosocomial cases from cases acquired outside the institution. The purpose of this study was to determine whether combining the ICD‐9‐CM code with medication treatment data for CDI in hospitalized patients could enable us to distinguish between patients with nosocomial CDI and patients who were admitted with CDI. The primary objective was to compare the sensitivity, specificity, and predictive value of using the combination of ICD‐9‐CM code for CDI and CDI treatment records to identify cases of nosocomial CDI with the sensitivity, specificity, and predictive value of using the ICD‐9‐CM code alone.
Design. Validation sample cross‐sectional study.
Setting. Academic health center.
Methods. Administrative claims data from July 1, 2004, to June 30, 2005, were queried to identify adults discharged with an ICD‐9‐CM code for CDI and to find documentation of CDI therapy with oral vancomycin or metronidazole. Laboratory and medical records were queried to identify symptomatic CDI toxin–positive adult patients with nosocomial CDI and were compared with records of patients whose cases were predicted to be nosocomial by means of ICD‐9‐CM code and CDI therapy data.
Results. Of 23,920 adult patients discharged from the hospital, 62 had nosocomial CDI according to symptoms and toxin assay. The sensitivity of the ICD‐9‐CM code alone for identifying nosocomial CDI was 96.8%, the specificity was 99.6%, the positive predictive value was 40.8%, and the negative predictive value was 100%. When CDI drug therapy was included with the ICD‐9‐CM code, the sensitivity ranged from 58.1% to 85.5%, specificity was virtually unchanged, and the range in positive predictive value was 37.9%–80.0%.
Conclusion. Combining the ICD‐9‐CM code for CDI with drug therapy information increased the positive predictive value for nosocomial CDI but decreased the sensitivity.
Received February 27, 2009; accepted May 20, 2009; electronically published October 2, 2009.
Clostridium difficile infection (CDI) is the most common cause of nosocomial infectious diarrhea; the incidence, case severity, rate of recurrence, and number of treatment‐refractory cases are all increasing.1‐4 Much of the increased incidence and severity of this disease is attributed to a new epidemic strain of C. difficile with greater virulence.4‐7
The Centers for Disease Control and Prevention and other professional organizations have recommended that hospitals monitor cases of nosocomial CDI to identify changes in rates and severity of CDI disease.8‐10 To our knowledge, there is no national surveillance system for healthcare‐associated CDI, but International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) codes obtained from administrative databases have been used to identify and track cases of CDI.3,11‐14 Furthermore, the ICD‐9‐CM code for CDI can be assigned as either a principal diagnosis (the condition established after discharge to be chiefly responsible for occasioning the admission of the patient to the hospital for care) or a secondary diagnosis (a condition that coexists at the time of admission or develops subsequently during the hospital stay).15 A limitation in using ICD‐9‐CM codes to identify CDI is the inability to determine which cases are nosocomial, because ICD‐9‐CM codes are assigned to all patients with CDI at any time during hospitalization, including both those admitted with CDI and those who develop CDI as a consequence of antimicrobial treatment during hospitalization (nosocomial CDI). While it would seem that cases in which CDI is a secondary diagnosis are most likely to be nosocomial, this has not been investigated that we are aware of.
Nearly all confirmed cases of CDI are treated with metronidazole or oral vancomycin.16 By combining records of antimicrobial treatment for CDI with ICD‐9‐CM codes for CDI, it may be possible to improve the ability to distinguish nosocomial cases from cases in which patients were admitted with CDI. Specifically, inpatients with nosocomial CDI will most likely receive therapy for CDI later in their hospitalization compared with patients who are admitted with CDI, who will likely receive CDI‐specific therapy at admission or shortly thereafter. Therefore, we hypothesized that the start date of CDI therapy would correlate with the date that CDI was suspected on the basis of clinical signs and symptoms, including diarrhea and positive results of a C. difficile toxin assay. Since many symptomatic patients are treated empirically for a few days with CDI therapy until C. difficile toxin assay results are available, we also hypothesized that patients who are treated longer are more likely to have CDI.
The primary objective of this study was to compare the sensitivity, specificity, and predictive value of using the combination of ICD‐9‐CM code for CDI and CDI treatment records to identify cases of nosocomial CDI with the sensitivity, specificity, and predictive value of using the ICD‐9‐CM code alone. Furthermore, since CDI listed as a secondary diagnosis is theoretically more likely to represent nosocomial CDI than when it is listed as a principal diagnosis, we hypothesized that CDI treatment records would further distinguish nosocomial CDI among those patients with a secondary diagnosis of CDI.
Methods
Data Sources
There were 2 data sources for this project: the University HealthSystem Consortium's (UHC's) Clinical Resource Manager (CRM) database and the medical records of patients at Virginia Commonwealth University Medical Center (VCUMC) for the same time period. The UHC (http://www.uhc.edu) is an alliance of 103 US academic health centers and 206 of their affiliated hospitals. The UHC CRM database combines patient encounter, charge master, and line item transactional data from participating institutions to provide resource utilization and patient outcome reports. UHC maps members’ drug descriptions from billing transactions into a common pharmacy lexicon, creating standardized descriptions to achieve reporting at the drug level. Additionally, demographic, procedure, and diagnosis data are derived from discharge abstract summaries. Additional details of the UHC CRM database were recently published.17
Identification of CDI Cases from the CRM Database and from Review of Medical Records
We queried the CRM database to identify all adult patients (age, ⩾18 years) discharged from VCUMC with a CDI diagnosis code (ICD‐9‐CM code 008.45, intestinal infection due to C. difficile) from July 1, 2004, through June 30, 2005. For each patient identified from the CRM database, we conducted a manual review of the patient’s medical record to compare selected patient variables with those reported in the CRM database. We also obtained from the hospital’s laboratory a list of all adult inpatients with a positive C. difficile toxin assay result (Meridian Bioscience C. difficile Premier Toxins A and B assay) who were discharged within the same time period.
Comparison of Cases from Medical Record Review with Cases Identified from the CRM Database
We determined from medical record review (1) whether the cases of CDI identified in the CRM database were true cases of CDI by using clinical and laboratory criteria (below), (2) whether CDI cases were nosocomial or nonnosocomial, and (3) whether the cases identified in VCUMC medical and/or laboratory records were identified in the CRM database.
To address the first 2 issues, we reviewed the medical records of all patients identified in the CRM database with an ICD‐9‐CM code for CDI. Patients were considered to have true cases of CDI if there was medical record documentation of diarrhea and a positive C. difficile toxin assay result or if there was documentation of a diagnostic imaging (computed tomographic scan, colonoscopy, endoscopy) result consistent with C. difficile disease and clinical symptoms such as diarrhea but no positive toxin assay result. A nosocomial case of CDI was defined a priori as a true case if the above criteria were met on day 5 or later of the patient's hospital stay (see below). In contrast, nonnosocomial infections were defined as those that occurred in patients who met the case criteria within the first 4 days of admission or who had a positive toxin assay result during a hospital stay within the previous month. Preadmission data were obtained from the admitting history, outpatient visit records, and/or records from referring hospitals. The criterion of requiring a positive toxin assay result on day 5 of the hospital stay or later was chosen to allow time for a nosocomial case to develop. This included time for onset of diarrhea, for stool sample collection, for laboratory testing for C. difficile, and for clinicians to respond to the positive result of a toxin assay.
To address the third issue, we determined what proportion of true and nosocomial cases of CDI identified from the hospital laboratory data were also identified in the CRM database. One investigator (M.S.) was responsible for medical record review and initial decisions regarding whether the cases represented CDI and whether CDI was nosocomial; these designations were reviewed by another study member (A.P.).
To test all aspects of our hypotheses, we compared the sensitivity, specificity, and predictive values of the following definitions for a case of nosocomial CDI within the CRM database: (1) the presence of any ICD‐9‐CM code (principal or secondary CDI) alone, (2) the presence of any ICD‐9‐CM code plus the receipt of at least 3 days of therapy for CDI (metronidazole or oral vancomycin), (3) the presence of any ICD‐9‐CM code plus start of CDI treatment no earlier than day 5 of hospitalization, and (4) the presence of any ICD‐9‐CM code plus the receipt of at least 3 days of therapy plus start of CDI treatment no earlier than day 5 of hospitalization. We also compared the sensitivity and specificity of the above criteria by using the same definitions but checking for the presence only of secondary ICD‐9‐CM diagnosis codes instead of either primary or secondary codes for CDI. We chose to use the criterion of at least 3 days of CDI therapy because many patients are started on CDI therapy empirically for a few days pending C. difficile toxin assay results before therapy is discontinued on receipt of a negative assay result. We also assessed the correlation between the start day of antibiotic therapy for CDI and the actual day a stool sample was collected that yielded a positive C. difficile toxin assay result.
Since there is more than 1 CDI treatment option and patients may receive more than 1 type of treatment, the start date of oral metronidazole was used first for calculating the correlation to the actual day of identification of a positive C. difficile toxin assay result. If no oral metronidazole was administered, then the start date of oral vancomycin was used. If neither oral metronidazole nor oral vancomycin was administered, then the start date of intravenous metronidazole was used.
Analysis
Descriptive analyses of the frequencies and percentages of CDI case classifications and characteristics of patients in each category were performed. Sensitivity, specificity, predictive values, and their exact binomial 95% confidence intervals were calculated for each definition. All analyses and calculations were performed with Stata SE, release 10 (StataCorp). Unless otherwise stated, a 2‐sided α level of .05 was used. The institutional review board at Virginia Commonwealth University approved this investigation.
Results
Comparison of Cases Identified from Medical Record Review with Cases Identified from the CRM Database
A total of 23,920 adult patients were discharged from VCUMC during the study period, and our review of CDI toxin assay results revealed 118 unique adult inpatients with positive results. Of these, 62 met our clinical definition of nosocomial CDI (ie, diarrhea and a positive toxin assay result no earlier than day 5 after admission) and 56 did not.
The CRM database query returned a total of 147 adult patients who had a CDI diagnosis code: 17 with a principal diagnosis code and 130 with a secondary diagnosis code. Of these 147 cases, 31 did not have a positive C. difficile toxin assay result and were not considered true cases of CDI. All 31 were coded as secondary CDI. Of the remaining 116 cases, 55 patients had a positive toxin assay result either prior to admission or within the first 4 days and were therefore considered to have nonnosocomial CDI (all secondary codes) and 1 patient underwent computed tomography in the emergency department; the results were suggestive of pseudomembranous colitis. Of the remaining 60 patients, all had a positive toxin assay result from a stool sample obtained on hospital day 5 or later and were considered to have nosocomial cases of CDI (17 had principal diagnosis codes, and 43 had secondary codes). Two nosocomial cases were identified by means of hospital laboratory data but did not appear in the CRM database and were therefore missed cases. Table 1 displays patient characteristics according to CDI status.
Of the 56 patients who were considered to have CDI at admission, 54 received metronidazole or oral vancomycin starting at some point in the first 4 days of their hospital stay (Figure 1). Of the 31 patients identified in the CRM database with CDI but without a positive C. difficile toxin assay result, 11 did not receive therapy for CDI and 10 received therapy for no more than 3 days (Figure 2). The day that metronidazole or oral vancomycin therapy was begun was highly correlated to the hospital day on which a stool specimen was collected that yielded a positive C. difficile toxin assay result (
, 
; Figure 3).
Figure 1. Day of hospitalization on which Clostridium difficile infection (CDI) treatment was started for 56 patients who had CDI at admission to the hospital.
Figure 2. Days of Clostridium difficile infection (CDI) treatment for 31 patients who were coded as having CDI but did not meet the study definition of CDI.
Figure 3. The start of oral or intravenous metronidazole or oral vancomycin treatment correlates well with the observed hospital day on which a specimen was collected that yielded a positive Clostridium difficile toxin assay result (
; 
).
The sensitivity, specificity, and predictive values of the various definitions for identifying nosocomial CDI by means of ICD‐9‐CM codes and data regarding medication use are shown in Table 2. The ICD‐9‐CM code alone had the highest sensitivity (96.8%); the specificity and positive predictive value were 99.6% and 40.8%, respectively. The highest specificity (100%) was achieved with the combination of ICD‐9‐CM code with receipt of at least 3 days of CDI therapy starting on or after day 5 of hospitalization, with a sensitivity and a positive predictive value of 71.0% and 80.0%, respectively. Using the definition with only secondary codes combined with receipt of at least 3 days of CDI therapy starting on or after day 5 of hospitalization resulted in a lower sensitivity value, 58.1%, and slightly lower predictive value, 76.7%, than the same definition with any diagnosis code.
Discussion
The use of administrative claims data to determine nosocomial infection rates has great appeal, but obtaining information on the timing of infection onset during the hospital stay is difficult. The results of this study suggest that combining drug therapy data specific for CDI with the ICD‐9‐CM code for CDI increases the positive predictive value for nosocomial CDI compared with using ICD‐9‐CM code data alone. While the specificity and sensitivity of identifying nosocomial CDI by using the ICD‐9‐CM code alone (either primary or secondary diagnosis) were high, the positive predictive value was only 40.8%. In contrast, use of the combined approach (definition 4, the presence of any ICD‐9‐CM code plus the receipt of at least 3 days of therapy plus start of CDI treatment no earlier than day 5 of hospitalization) maintained high specificity and increased the positive predictive value to as much as 80.0%, although it decreased the sensitivity. The decreased sensitivity resulted from the additional requirement that the drug treatment duration criterion be met (ie, therapy for at least 3 days). Many patients with true cases of CDI will likely receive only 1–2 days of therapy in the hospital before being discharged to complete therapy at home. When dropping this condition and using the definition for CDI of presence of a primary or secondary code and the start of CDI treatment no earlier than day 5 of hospitalization, we found that sensitivity increased to 77.4% with a slight decrease in specificity (99.9%); the positive predictive value was 78.7%. Having CDI listed as a secondary diagnosis versus any diagnosis (principal or secondary) actually resulted in a decreased positive predictive value, 76.7% versus 80.0%. Surprisingly, all 17 patients with CDI listed as a principal diagnosis had nosocomial CDI.
A previous study by Dubberke et al evaluated the sensitivity and specificity of CDI surveillance based on ICD‐9‐CM codes at 1 tertiary care hospital.18 The sensitivity and specificity of ICD‐9‐CM codes were 78% and 99.7%, respectively. Another study by Scheurer et al revealed that the sensitivity, specificity, and positive and negative predictive values of surveillance based on ICD‐9‐CM codes were 71%, 99%, 87%, and 96%, respectively, at a 747‐bed nonprofit teaching hospital.19 It was not determined in these 2 studies whether the CDI cases were nosocomial, so direct comparisons of these results to those of our study are not possible. In addition, direct comparisons are not possible because of different source patient populations selected. In the study by Scheurer et al, for example, sensitivity and specificity calculations were based on the number of unique patients who had a positive toxin assay result and not on the total number of hospital discharges, as was done in our study and the study by Dubberke et al.18 Most administrative databases do not include results of laboratory assays, but we found that the initiation date of treatment for CDI was an accurate surrogate for the date of a positive toxin assay result. By identifying the date of CDI treatment, one can distinguish nosocomial CDI cases from community‐acquired cases.
There are limitations to this study. We may have misclassified cases as a result of our reliance on C. difficile toxin assay results for the definition of CDI cases. None of the toxin assay tests commonly used to help diagnose CDI are both 100% sensitive and 100% specific. The sensitivity of the enzyme immunassay toxin tests ranges from 70% to 80%.20 Our definition, however, required a positive toxin assay result for C. difficile toxin A and/or B in addition to the presence of diarrhea. Since the VCUMC laboratory performs a C. difficile toxin assay only on unformed, or liquid, stool samples, all patients were likely to have diarrhea at the time of testing. A further limitation is one that is inherent in any investigation using ICD‐9‐CM codes to identify cases—that the accuracy of the coding is reliant on documentation by health‐care providers and on the accuracy of the medical coders. Finally, this evaluation was conducted in a single facility, and findings may not be extrapolatable to other institutions.
The addition of valid medication use data to ICD‐9‐CM coding data to identify nosocomial cases of CDI has advantages and disadvantages. The combination approach will not be a useful strategy to measure the prevalence of CDI or to study trends in its occurrence, since it is possible that true cases may be missed by using ICD‐9‐CM codes or that negative cases may be coded as CDI. Unfortunately, studies that use ICD‐9‐CM codes alone, as is commonly done, are prone to these same errors.3,11‐14 Valid measures of the prevalence of various types of infections in administrative databases await further refinement. On the other hand, a case from a database that is identified by using ICD‐9‐CM codes plus drug treatment for CDI appears to have a relatively high likelihood (almost 80%) of being a true nosocomial CDI case; nearly all patients identified by this means do in fact have CDI. The true prevalence of the disease may not be important in some studies, such as those investigating risk factors for nosocomial CDI or examining the clinical outcomes of CDI; thus, missed cases are not of high importance. What is important is that nearly all cases categorized as nosocomial CDI be in fact nosocomial CDI so that the investigation is centered on true cases. In this study, patients with cases of CDI identified by using an ICD‐9‐CM code and who also began receiving CDI‐specific therapy starting after at least 5 days of hospitalization and for at least 3 days have an approximately 80% chance of having true nosocomial cases of CDI.
Beginning October 1, 2008, hospitals were required by the Centers for Medicare and Medicaid Services to indicate which diagnoses were present on admission (POA).21 While the POA indicator may be useful in identifying hospital‐acquired infections, the method described in this article would be useful to assess CDI cases recorded before October 1, 2008, and to provide a way to compare infection rates before and after the introduction of the POA indicator. Repeating our analysis with 2008–2009 POA data will help determine the validity of the POA indicator compared with using ICD‐9‐CM code and medication use data in identifying cases of hospital‐acquired CDI. The method proposed in this investigation should be useful to help determine the postadmission day that nosocomial CDI became evident.
In summary, claims databases that include ICD‐9‐CM codes and drug administration information, such as the UHC CRM database, seem to be useful in identifying cases of nosocomial CDI. In addition, the postadmission start date of CDI diagnosis (from positive toxin assay result) is highly correlated to the start date for metronidazole or oral vancomycin. Furthermore, the majority of cases that were identified by means of ICD‐9‐CM codes that were not nosocomial, such as those that were present when the patient was admitted to the hospital or those that occurred in patients who did not have a positive toxin assay result, can be eliminated from consideration.
Acknowledgments
We thank Betty Dixon‐Leigh and Betty Forbes for assistance with data collection and Norm Carroll, PhD, for his critical review of the manuscript.
Financial support. Grant from the PhRMA Foundation.
Potential conflicts of interest. A.P. reports that she has received honoraria from Schering‐Plough for an oral presentation and research funding from ViroPharma and Cubist Pharmaceuticals. R.P. reports that he has received research funding from ViroPharma and Cubist Pharmaceuticals and is a member of the Schering‐Plough speakers’ bureau. S.H. reports that he has received research funding from ViroPharma. Both other authors report no conflicts of interest relevant to this article.
References
- 1. Musher DM, Aslam S, Logan N, et al. Relatively poor outcome after treatment of Clostridium difficile colitis with metronidazole. Clin Infect Dis 2005; 40:1586–1590.
- 2. Pepin J, Alary M‐E, Valiquette L, et al. Increasing risk of relapse after treatment of Clostridium difficile colitis in Quebec, Canada. Clin Infect Dis 2005; 40:1591–1597.
- 3. McDonald LC, Owings M, Jernigan DB. Clostridium difficile infection in subjects discharged from US short‐stay hospitals, 1996–2003. Emerg Infect Dis 2006; 12:409–415.
- 4. 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; 353:2442–2449.
- 5. McDonald LC, Killgore GE, Thompson A, et al. An epidemic, toxin gene–variant strain of Clostridium difficile. N Engl J Med 2005; 353:2433–2441.
- 6. Warny M, Pepin J, Fang A, et al. Toxin production by an emerging strain of Clostridium difficile associated with outbreaks of severe disease in North America and Europe. Lancet 2005; 366:1079–1084.
- 7. Frost F, Craun GF, Calderon RL. Increasing hospitalization and death possibly due to Clostridium difficile diarrheal disease. Emerg Infect Dis 1998; 4:619–625.
- 8. McDonald LC, Coignard B, Dubberke E, Song X, Horan T, Kutty PK, Ad Hoc Clostridium difficile Surveillance Working Group. Recommendations for surveillance of Clostridium difficile–associated disease. Infect Control Hosp Epidemiol 2007; 28:140–145.
- 9. Siegel JD, Rhinehart E, Jackson M, Chiarello L, Healthcare Infection Control Practices Advisory Committee. Management of multidrug‐resistant organisms in healthcare settings, 2006. Atlanta, GA: Centers for Disease Control and Prevention, 2006. Available at: http://www.cdc.gov/ncidod/dhqp/pdf/ar/mdroguideline2006.pdf. Accessed May 14, 2009.
- 10. Dubberke ER, Gerding DN, Classen D, et al. Strategies to prevent Clostridium difficile infections in acute care hospitals. Infect Control Hosp Epidemiol 2008; 29(Suppl 1):S81–S92.
- 11. Ricciardi 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; 142:624–631.
- 12. Chandler RE, Hedberg K, Cieslak PR. Clostridium difficile–associated disease in Oregon: increasing incidence and hospital‐level risk factors. Infect Control Hosp Epidemiol 2007; 28:116–122.
- 13. 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; 28:1219–1227.
- 14. 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; 14:929–931.
- 15. Iezzoni LI. Coded data from administrative sources. In: Iezzoni LI, ed. Risk Adjustment for Measuring Health Care Outcomes, 3rd ed. Chicago, IL: Health Administration Press; 2003: 83–138.
- 16. Gerding DN, Muto CA, Owens RC. Treatment of Clostridium difficile infection. Clin Infect Dis 2008; 46(Suppl 1):S32–S42.
- 17. Pakyz AL, MacDougall C, Oinonen M, Polk RE. Trends in antibacterial use in US academic health centers. Arch Intern Med 2008; 168:2254–2260.
- 18. Dubberke ER, Reske KA, McDonald LC, Fraser VJ. ICD‐9 codes and surveillance for Clostridium difficile‐associated disease. Emerg Infect Dis 2006; 12:1576–1579.
- 19. Scheurer DB, Hicks LS, Cook EF, Schnipper JL. Accuracy of ICD‐9 coding for Clostridium difficile infections: a retrospective cohort. Epidemiol Infect 2007; 135:1010–1013.
- 20. Bartlett JG, Gerding DN. Clinical recognition and diagnosis of Clostridium difficile infection. Clin Infect Dis 2008; 46(Suppl 1):S12–S18.
- 21. Centers for Medicare and Medicaid Services. Hospital‐acquired conditions (present on admission indicator). Baltimore, MD: Centers for Medicare and Medicaid Services. Updated July 20, 2009. Available at http://www.cms.hhs.gov/HospitalAcqCond/. Accessed May 14, 2009.
-
Presented in part: 46th Interscience Conference on Antimicrobial Agents and Chemotherapy; San Francisco, California; September 27–30, 2006.