Recurrent C. diff Infection Among Hospitalized Patients
This study was approved by the Washington University Institutional Review Board, and its conduct was in compliance with the Helsinki Declaration.
We conducted a retrospective cohort study of all adult (age ≥18 years) patients with an inpatient episode of iCDI at Barnes-Jewish Hospital (BJH) between January 1, 2003, and December 31, 2009. An episode of CDI was defined as a positive toxin assay (C. DIFFICILE TOX A/B II from Techlab, Blacksburg, VA, USA) for C. difficile. Because the hospital laboratory performs a test for C. difficile only if the treating physician suspects CDI and if the stool is unformed, all patients with positive toxin results were considered to be CDI case patients. The first episode of CDI during the study period in the absence of any CDI in the prior 60 days was defined as the iCDI, and patients were included only once. Patients were excluded if they died during or were discharged to hospice from the iCDI hospitalization.
All included patients were followed for 42 days from the date of the end of iCDI treatment or until rCDI onset, defined as a repeat positive toxin within this time frame. Initial CDI cases were categorized according to published surveillance definitions as community-onset healthcare facility-associated (CO-HCFA) (indeterminate CDI cases were grouped with CO-HCFA), healthcare facility-onset (HCFO), and community-associated (CA).
Demographic and clinical data were derived from BJH Medical Informatics databases and the BJH electronic medical records. The data available from the Informatics databases included C. difficile toxin assay results and date of stool collection; patient demographics; dates of admission and discharge; discharge disposition; admission location; ICD-9-CM diagnosis (used to define underlying comorbidities in the year prior and during the index hospitalization) and procedure (assessed only during the index hospitalization) codes; dates of ICU stays; start and stop dates of all inpatient CDI treatments, gastric acid suppressors and antimicrobials; and white blood cell count, hemoglobin, serum creatinine, and serum albumin levels on admission and at the time of positive C. difficile toxin assays from the index admission and all readmissions in the 42 days after iCDI treatment end. The BJH medical records included data on antimicrobials and CDI treatments the patient received as an outpatient within the BJH system, and whether a readmission was for CDI. In addition, admission and discharge summaries were reviewed for all included hospitalizations to help determine whether the patient had a history of CDI at another healthcare facility or as an outpatient.
The exposure interval was divided into three periods: 1) from hospital admission until diagnosis of iCDI, 2) from the time of diagnosis of iCDI until the end of its treatment, and 3) from the end of iCDI treatment until the onset of recurrence or until the end of the 42-day monitoring period for recurrence. We compared patients with rCDI to those without rCDI based on their characteristics in these time periods. Cox proportional hazards modeling was used to determine variables associated with at least one episode of rCDI on univariate analysis. Antimicrobials were categorized based on association with CDI as high-risk (cephalosporins, aminopenicillins, and clindamycin), low-risk (aminoglycosides, beta lactamase inhibitors, carbepenems, daptomycin, doxycycline, linezolid, macrolides, rifampin, rifaximin, and tigecycline), fluoroquinolones (>90% was ciprofloxacin), and intravenous vancomycin. Gastric acid suppressors (histamine receptor 2 blockers [HR2B] and proton pump inhibitors [PPI]), choice and duration of iCDI treatment, and iCDI severity, as defined by the SHEA/IDSA Clinical Practice Guidelines for CDI, were also assessed as potential risk factors for rCDI.
We employed extended Cox proportional hazards modeling to determine independent risk factors for at least one episode of rCDI, with variable selection according to the methodology of Hosmer-Lemeshow. Variables eligible for inclusion in the multivariable models were those associated with increased risk of rCDI from the literature or those with clinical or biologic plausibility, and those with p-values <0.20 in the univariate analyses. Antimicrobial exposures from the end of CDI treatment until rCDI or 42 days were analyzed as time-dependent variables. Backward stepwise selection was used to arrive at the best-fitting and most parsimonious model. All relevant 2-way interactions were tested after selection of the main effects, and included in the final models only if they were significant at the alpha ≤0.05. The proportional hazards assumption was verified by assessing the parallel nature of curves in log-log plots. The appropriate functional formats of continuous variables were determined by examining nonparametric regression (smoothing) plots with a restricted cubic spline function. To facilitate interpretation of results, the hazard ratios for the piecewise linear spline variable (fluoroquinolone exposure while on CDI treatment) compared the hazards of developing CDI for values between the 75th and the 25th percentiles of the variable. To assess the importance of time dependency for antimicrobial exposures that occur after CDI treatment, these exposures included in the final model were also analyzed in a time-independent fashion.
All analyses were performed in SAS version 9.3 (SAS Institute, Cary, NC) and R (R Foundation, Vienna, Austria). All statistical testing was two-tailed with significance set at the alpha level ≤ 0.05.
Methods
This study was approved by the Washington University Institutional Review Board, and its conduct was in compliance with the Helsinki Declaration.
Cohort Definition
We conducted a retrospective cohort study of all adult (age ≥18 years) patients with an inpatient episode of iCDI at Barnes-Jewish Hospital (BJH) between January 1, 2003, and December 31, 2009. An episode of CDI was defined as a positive toxin assay (C. DIFFICILE TOX A/B II from Techlab, Blacksburg, VA, USA) for C. difficile. Because the hospital laboratory performs a test for C. difficile only if the treating physician suspects CDI and if the stool is unformed, all patients with positive toxin results were considered to be CDI case patients. The first episode of CDI during the study period in the absence of any CDI in the prior 60 days was defined as the iCDI, and patients were included only once. Patients were excluded if they died during or were discharged to hospice from the iCDI hospitalization.
All included patients were followed for 42 days from the date of the end of iCDI treatment or until rCDI onset, defined as a repeat positive toxin within this time frame. Initial CDI cases were categorized according to published surveillance definitions as community-onset healthcare facility-associated (CO-HCFA) (indeterminate CDI cases were grouped with CO-HCFA), healthcare facility-onset (HCFO), and community-associated (CA).
Data Sources
Demographic and clinical data were derived from BJH Medical Informatics databases and the BJH electronic medical records. The data available from the Informatics databases included C. difficile toxin assay results and date of stool collection; patient demographics; dates of admission and discharge; discharge disposition; admission location; ICD-9-CM diagnosis (used to define underlying comorbidities in the year prior and during the index hospitalization) and procedure (assessed only during the index hospitalization) codes; dates of ICU stays; start and stop dates of all inpatient CDI treatments, gastric acid suppressors and antimicrobials; and white blood cell count, hemoglobin, serum creatinine, and serum albumin levels on admission and at the time of positive C. difficile toxin assays from the index admission and all readmissions in the 42 days after iCDI treatment end. The BJH medical records included data on antimicrobials and CDI treatments the patient received as an outpatient within the BJH system, and whether a readmission was for CDI. In addition, admission and discharge summaries were reviewed for all included hospitalizations to help determine whether the patient had a history of CDI at another healthcare facility or as an outpatient.
Statistical Analyses
The exposure interval was divided into three periods: 1) from hospital admission until diagnosis of iCDI, 2) from the time of diagnosis of iCDI until the end of its treatment, and 3) from the end of iCDI treatment until the onset of recurrence or until the end of the 42-day monitoring period for recurrence. We compared patients with rCDI to those without rCDI based on their characteristics in these time periods. Cox proportional hazards modeling was used to determine variables associated with at least one episode of rCDI on univariate analysis. Antimicrobials were categorized based on association with CDI as high-risk (cephalosporins, aminopenicillins, and clindamycin), low-risk (aminoglycosides, beta lactamase inhibitors, carbepenems, daptomycin, doxycycline, linezolid, macrolides, rifampin, rifaximin, and tigecycline), fluoroquinolones (>90% was ciprofloxacin), and intravenous vancomycin. Gastric acid suppressors (histamine receptor 2 blockers [HR2B] and proton pump inhibitors [PPI]), choice and duration of iCDI treatment, and iCDI severity, as defined by the SHEA/IDSA Clinical Practice Guidelines for CDI, were also assessed as potential risk factors for rCDI.
We employed extended Cox proportional hazards modeling to determine independent risk factors for at least one episode of rCDI, with variable selection according to the methodology of Hosmer-Lemeshow. Variables eligible for inclusion in the multivariable models were those associated with increased risk of rCDI from the literature or those with clinical or biologic plausibility, and those with p-values <0.20 in the univariate analyses. Antimicrobial exposures from the end of CDI treatment until rCDI or 42 days were analyzed as time-dependent variables. Backward stepwise selection was used to arrive at the best-fitting and most parsimonious model. All relevant 2-way interactions were tested after selection of the main effects, and included in the final models only if they were significant at the alpha ≤0.05. The proportional hazards assumption was verified by assessing the parallel nature of curves in log-log plots. The appropriate functional formats of continuous variables were determined by examining nonparametric regression (smoothing) plots with a restricted cubic spline function. To facilitate interpretation of results, the hazard ratios for the piecewise linear spline variable (fluoroquinolone exposure while on CDI treatment) compared the hazards of developing CDI for values between the 75th and the 25th percentiles of the variable. To assess the importance of time dependency for antimicrobial exposures that occur after CDI treatment, these exposures included in the final model were also analyzed in a time-independent fashion.
All analyses were performed in SAS version 9.3 (SAS Institute, Cary, NC) and R (R Foundation, Vienna, Austria). All statistical testing was two-tailed with significance set at the alpha level ≤ 0.05.
SHARE
