Research Article: Surveillance for Clostridium difficile Infection: ICD-9 Coding Has Poor Sensitivity Compared to Laboratory Diagnosis in Hospital Patients, Singapore

Date Published: January 20, 2011

Publisher: Public Library of Science

Author(s): Monica Chan, Poh Lian Lim, Angela Chow, Mar Kyaw Win, Timothy M. Barkham, Erika Martins Braga. http://doi.org/10.1371/journal.pone.0015603

Abstract: Clostridium difficile infection (CDI) is an increasingly recognized nosocomial infection in Singapore. Surveillance methods include laboratory reporting of Clostridium difficile toxin assays (CDTA) or use of International Classification of Diseases, 9th Revision (ICD-9) discharge code 008.45. Previous US studies showed good correlation between CDTA and ICD-9 codes. However, the use of ICD-9 codes for CDI surveillance has not been validated in other healthcare settings.

We compared CDI rates based on CDTA to ICD-9 codes for all discharges in 2007 from our hospital to determine sensitivity and specificity of ICD-9 codes. Demographic and hospitalization data were analyzed to determine predictors for missing ICD-9 codes.

During 2007, there were 56,352 discharges. Of these, 268 tested CDTA-positive but only 133 were assigned the CDI ICD-9 code. A total of 141 discharges had the ICD-9 code; 8 were CDTA-negative, the rest were CDTA-positive. Community-acquired CDI accounted for only 3.2% of cases. The sensitivity and specificity of ICD-9 codes compared to CDTA were 49.6% and 100% respectively. Concordance between CDTA and ICD-9 codes was 0.649 (p<.001). Comparing concordant patients (CDTA+/ICD9+) to discordant patients (CDTA+/ICD9−), concordant patients were more likely to be over 50 years of age (OR 3.49, 95% CI 1.66–7.34, p = .001) and have shorter time from admission to testing (OR 0.98, 95% CI 0.97–0.99, p = .009). Unlike previous studies in the US, ICD-9 codes substantially underestimate CDI in Singapore compared to microbiological data. Older patients with shorter time to testing were less likely to have missing ICD-9 codes.

Partial Text: Clostridium difficile infection (CDI) is an emerging healthcare-associated problem in Singapore. Among hospitalized patients in our institution, CDI incidence has risen 4-fold from 1.49 cases per 10,000 patient-days in 2001 to 6.64 cases per 10,000 patient-days in 2006 [1]. This increased incidence is comparable to that reported by large hospitals in Canada [2], and mirrors the increases seen in North America and Europe over the past decade. However, national surveillance systems to track CDI rates are relatively less well-developed. Currently, potential surveillance methods include laboratory-based reporting of diagnostic assays or administrative surveillance using International Classification of Diseases, 9th Revision (ICD-9) codes assigned to hospital discharges. Although the primary role of ICD-9 codes is for remuneration, easy accessibility, standardized format across healthcare facilities and consistency over time make ICD-9 codes attractive for surveillance purposes. Previous studies in the United States demonstrated good correlation between toxin assay results and ICD-9 codes, with sensitivity and specificity of ICD-9 codes reported at 71–78% and >99% respectively compared to microbiologic data [3]–[5]. A recent study by Zilberberg and colleagues showed good agreement between pediatric CDI hospitalization rates using administrative (ICD-9) coding from 2 separate databases [6]. However, use of ICD-9 codes for CDI surveillance has not been validated in other healthcare settings including Asia. In this study, we compared CDI rates based on laboratory diagnostic testing to CDI diagnoses captured by ICD-9 codes for hospitalized patients to determine the sensitivity and specificity of ICD-9 codes for use as CDI surveillance.

Ethics Statement: The institutional ethics review board for the National Healthcare Group Domain Specific Review Board (NHG DSRB) approved the study prior to initiation (approval number DSRB E/09/008). A waiver of informed consent was specifically requested and granted by the NHG DSRB because the study methods utilized retrospective medical record review on hospitalization records of patients who had been discharged over 2 years before, and the data would be collected in anonymized unlinked datasets and reported in anonymized aggregate form.

Of 56,352 admissions to Tan Tock Seng Hospital in 2007, 2,212 (3.9%) patients had CDTA requested. Clostridium difficile toxin assay were positive in 268 (12.1%) but only 133 were assigned the ICD-9 code [Figure 1]. An ICD-9 code of 008.45 for CDI was assigned to 141 discharges. Of these, 133 had CDTA-positive and 8 had CDTA-negative results. Review of medical records confirmed symptoms of diarrhea, abdominal pain or cramping in all patients with positive CDTA or ICD-9 code.

Rates of Clostridium difficile infection would be substantially underestimated by ICD-9 codes compared to laboratory testing in our setting. Sensitivity of ICD-9 codes is poor compared to laboratory testing but specificity remains high. A substantial reduction in sensitivity is attributable to coder interpretation of medical records. More than half of the cases with missed ICD-9 codes had the CDI diagnosis written in the text of the medical discharge summary, yet were not coded appropriately. Inclusion of these records would increase sensitivity from 49.6% to 76%, comparable to other published studies [3]–[5]. Training for coders could improve sensitivity and make ICD-9 coding feasible as a surveillance instrument for CDI. Although ICD9 coding system is widely used and should be standardized, differences in institutional structure and awareness of Clostridium difficile may influence coding practices. Public hospitals in Singapore offer government-subsidized healthcare with individual “co-payment” from compulsory medical saving plans. It is possible that differences in payers and billing methods could affect the rigor of medical coding if compared to private hospitals or insurance-based healthcare systems elsewhere. These findings suggest variations in sensitivity that should be validated before use in surveillance.

Source:

http://doi.org/10.1371/journal.pone.0015603