Drug risk factors associated with a sustained outbreak of Clostridium difficile diarrhea in a teaching hospital

A case-control study was undertaken to identify and quantify antimicrobial and nonantimicrobial drug risk factors associated with a sustained outbreak of Clostridium difficile diarrhea on two medical (teaching and nonteaching) units and an oncology unit. In total, 80 cases associated with an endemic clone of toxigenic C difficile were compared with controls. Eighty controls were selected from a group of 290 controls randomly chosen from the outbreak period. The controls were matched to cases according to age, admitting diagnosis and unit of admission. Seventy (88%) patients in the case group received at least one antibiotic before diarrhea, compared with 37 (46%) patients in the control group. Major risk factors implicated in the development of C difficile diarrhea in hospitalized patients were the following antimicrobial agents: ceftazidime (adjusted odds ratio [aor]=26.01, 95% ci 5.67 to 119.19, P=0.0001); cefuroxime (aor=5.17, ci 1.86 to 14.36, P=0.005); ciprofloxacin (aor=3.81, ci 1.05 to 13.79, P=0.04); and clindamycin (aor=15.16, ci 2.93 to 78.44, P=0.004). This is the first time that the use of ciprofloxacin has been linked to the development of C difficile diarrhea. Use of gastrointestinal drugs (ranitidine, famotidine, cimetidine, omeprazole and sucralfate) was also an added risk (aor=3.20, ci 1.39 to 7.34, P=0.01); however, antineoplastic therapy was not significant (P<0.53). Recognition of the specific high risk drugs may spur more restricted use of these agents, which may help in controlling C difficile diarrhea in hospitalized patients.

C LOSTRJOIUM OIFFICILE IS A MAJOR CAUSE OF GASTROINTES- tinal infections associated with a wide spectrum of clinical manifestations.from asymptomatic inles lina l carriage to mild self-limited diarrhea to severe, potentially fatal pseudomembranous colitis (1).The diagnos is of C dij[icile diarrhea rests on the detection of the specific cytotoxin in stool filtrates and/or lhe demonstration of lhe toxigenic organism in culture, accompanied by the clinical symptoms of diarrhea (2) .
Since early 1990 there h as been an apparent increase in th e number of hospita lized pa tients wilh diarrhea due to toxigenic C dij[icile in our inslilution.While cases were initia lly sporadic in various nu rsing unils of lhe hospital, in creases in the frequency of cases have s ubsequently b een documented in lhe medical and oncology service areas.Using the numerical analysis of whole cell protein patterns, we have deduced lhe epidemiology of an evolving C dilficile outbreak in our institution (23).We have a lso observed the presis lenl transmission of a toxigenic clone of C dij[icile among our in-patients, predominantly on four unils (24).In th e present investigation we have undertaken a controlled study to id entify and quantify s p ecific antimicrobia l and nonantimicrobial drugs predisposing patien ts lo C dijjicile diarrhea.

Case definition:
Henderson General Hospila l is a 465bed tertia ry care university-affiliated teaching hospital primarily serving adu lts.The h osp ita l h as a n outpatient can cer clinic on camp u s and houses lwo oncol- ogy (medical a nd h ematological) nursing units for cancer patients from the region.Th e hospital a lso has an active medical leaching unil.Demograph ic characlerisitics of lhe patient popula tion on the medical and oncology units h ave nol changed over the p ast l 0 years.Laboratory-based surveillance of infections is routin ely performed hospital-wid e. Th is routin e surveillance.initialed before lhe C dijficile outbreak, involves an ongoing, systematic recording, analysis and interpretation of accu rate microbiological resu lts essential lo lhe planning, implementation and eval uation of lhe infection control practice.and is closely integrated wilh lhe timely dissemina tion of th ese data lo the infection con trol committee .
The endemic spread of a toxigenic clone of C dij[icile (electrophoretic type 1) in th e hospital units has been previously repo rted (24).There h as b een a marked increase in electrophoretic type 1 cases since October 1991.There was a n apparent decrease in lhe number of n ew cases beyond May 1993, which was the peak month during this outbreak.The present case-control investigation covers a period of 20 months, from October 1991 through May 1993.C dilficile diarrhea cases were evalu ated based on lh e medical charts and microbiolgical results of stool examinations undertaken on in-patients.A C diJficiLe diarrhea case included lhe following criteria: first, an adult patient hospita lized for more than three days; second.a patient identified by a health care provider as h aving at least fi ve episodes of d ia rrhea in 24 h for three or more days: third , a patient wilh loose stools in which cytotoxin (C dij[icile toxin B) a nd enterotoxin (C dij[icile loxin A) were detected: and fourth .toxigenic C dij[icile was the only enteric pall1ogen isolated as the causative agent of diarrhea.
The dale of lhe first detection of toxin or isolation of toxigenic C dij[icile was considered to be the dale of disease onset b eca use il was often impossible to determine lhe onset of diarrhea from a retrosp ec tive review of lh e patient chart.Therapy for suspected C dij[icile diarrhea or colitis included discontinua tion of any antimicrobial agents a nd gastrointestinal drugs used b efore the onset of diarrh ea and inslilution of oral melronid azole.There were no follow-up s igmoidoscopic or colonoscopic exa min a tions performed on elderly inpatients with C dij[icile diarrh ea.Case-control study: A retrospective case-control analysis was undertaken using 80 cases associated with the endemic type 1 C difficile.The cases were chosen from a sampling frame of 170 patients hospital-wide.The cases were on a medical teaching unit.a nonteaching medical unit and a hematological oncology unit.and were in the age group of 58 lo 88 years (mean age ±15).1\vo hundred and ninety controls without a previou history of fecal toxin detection and/or toxigenic C difficile isolation.and a negative stool culture of C difficile and its toxins between October 1. 1991 and May 31.1993 were randomly selected from all adult patients admitted lo these three units.Eighty of these controls were matched lo the cases according lo age.admitting diagnosis and un it of admission.
The patient charts and laboratory-ba ed surveillance records were reviewed.and the follo\ving information was recorded for both ca es and controls: age: sex: diagnosis on admission: dales of admission and discharge: and dale of first detection of C difficile toxin or first isolation of toxigenic C difficile (in cases).Pharmacy records were reviewed lo examine the medication profiles of the cases and matched control .The names and date of antimicrobial agents prescribed before and after the onset of diarrhea: antineoplastic drugs: and other medications known lo affect the ga lroinlestinal system (H2-blockers.antacids.cyloprolective agents and laxatives) were recorded.Antimicrobial agents used in fewer than four (5%) case compared were excluded from the analyses.Antineoplastic drugs analysed were: cylarabine.cyclophosphamide.doxorubicin, vincristine and etoposide.The gastrointestinal drug included in the analyses were: ranilidine.famolidine.cimetidine.omeprazole.alginic acid.sucralfale and aluminium/magne ium ails.No other gaslro-intestinal agents were studied.For the case group.all medications given six weeks before the dale of onset of diarrhea were recorded.For the control group.all medications and the length of treatment during lhe patient's slay on the affected unit were recorded.Statistical analyses: All statistical analyses were performed using SAS statistical software (version 6.07: SAS Inst.ilule.Inc. North Carolina).Univariate analysis was performed initially and the variables were subsequently entered in the stepwise logistic function regression (SLFR) model.Demographic and exposure variables were coded as dichotomous (0 'absent' and 1 'present') variables.Crude odds ratios (COR) and Cl for odds ratios were calculated by lhe standard methods.Fisher's exact lest was used lo measure lhe statistical ignificance.For the SLFR analysis.variables were entered into and removed from lhe model according to the following s leclion proce .First. the adju led x 2 stati lie for all the vari- ables not yet in the model were analyzed.The variable corresponding lo lhe most significant x 2 statistic was entered into the model if its significance level wa less than 0.4.Second.a variable was removed from the model if il was the least ignificanl variable in the model or lhe level of significance of the maximum likelihood estimate was greater than 0.2.The selection process terminated if more variables could not be entered by the forward selection process or if the last variable entered inlo the model was the only variable subsequently removed.CORS from the univariate analysis and adjusted odds ratios from the SLFR model were con idered significant if the 95% Cl did not.include 1.0.

RESULTS
Description of the out break: An epidemic curve representing lhe monthly number of new C difficile ca es and the cumulative total over the period from October 1991 through June 1993 is shown in Figure 1.A lola! of 170 cases of C d([ficile diarrhea were identified during the study period from October 1991 through May 1993.which was the sampling frame for the selection of cases in the pre ent investigation. The hospital-wide attack rate increased from 0.15/100 discharges in 1991 lo 1.36/100 discharges in 1993.The earliest recorded incidence rate. in 1988.was 0.013.Cases were identified from 13 of24 units in the hospital over the study period.Three units had only one patient each wilh C difficile.Among the remaining 10 units were six where at.least two cases were identified within four weeks of each oth r.A sharp increase in attack rate was recorded for the medical leaching and nonteaching medical units and a hematological oncology unit.The increase was most dramatic in the first.five months of 1993.with a rate of 6.09 for lhe nonteaching medical unit: 2.49 for the medical teaching unit: and 1.81 for the hematological oncology unit.To ensure that.this outbreak was not simply a laboratory phenomenon.il was documented that lhe labora- tory procedures of toxin assays or microbiological identification during the period of this study were unchanged.Before October 1991, the cases were sporadic in nature on seven units in the hospital , as described previously (23).An initial outbreak of diarrhea with a cluster of nin e cases on the medi cal teaching unit within a period of eight w eeks was believed to have started on November 20. 1991.From this lime.s ubsequent cases were identified more frequently on lh e same unit.Once established on this unit, toxigenic C difficile spread to cause further secondary outbreaks successively through a hematological oncology unit, a nonteaching m edical unit and, eventually .the intesive care unit (24).Case-control comparisons: To determine the risk factors for C difficile diarrhea, a case-control investigation was conducted fo cusing on patients housed on the three most frequently affected units: a m edical teaching unit, a nonteaching medical unit and a h ematological oncology unit.The in-patients selected for the case group h ad the following diagnoses on a dmission: 24 patients had pneumonia, 19 h ematological malignancy, 14 carcinoma, 13 h eart disease, five diabetes (cellulitis) and five urinary tract infection.Matching was performed based on age, admitting diagnosis and unit of admission.There were 23 oncology patients with secondary pneumonia or line-related sepsis in the case group who were match ed to 23 oncology patients with s imilar secondary infections in the control group.The cases and m atch ed controls were distributed through out the outbreak period of October 1991 through May 1993.The cases and controls were similar with respect to m ean age (cases, 72.6; controls, 70.9), and mal e:femal e ratio (cases.l: 1.05; controls .1:0.66).There were no s ignificant diffe rences in numbers of previous admissions, length of stay at th e present admission and recent surgical intervention (P>0.05).No diarrhea or any other symptomatic C difficiLe infection was record ed for pa tients in the A more intensive comparison of recent m edicatio n s a dministered was performed.Seventy (88%) patients in the case group received at least one antibiotic.compared with 37 (46%) in the control group.Twenty-one cases received one antibiotic: 30 cases received two antibiotics: 16 cases received three a ntibiotics; a nd three cases received four or more a ntibiotics , compared with n ine.22 , six and none, respectively, for th e controls (Mantel-Haenszel x 2 =23 .975.P<O.OOOl).
All 11 variables were consid ered for entry into the SLFR model; however.the following four variables were removed from the model: vancomycin.piperacillin.aminoglycosides and antineoplastic drugs.Ceftazidime.clindamycin and cefuroxime remained s ignificant risk fac tors in descending order of s ignificance (Table 1) .Ceftazidime had an a dju sted odds ratio (AOR) differing s ubstantia lly from Lhe COR.Ceftazidime had increased odds ra tio of 26.01 with 95% Cl 5.67 to 11 9 .1 9 in th e SLFR mod el.The odds ra tio of cefuroxime was a lso substantially in creased (AOR=5.17. 95% CI 1.86 to NATH eta/ 14.36).In the SLFR model.the use of ciprofloxacin was found to be an additional risk factor (AOR=3 .81,95% CI 1.05 to 13. 79).Use of gastrointestinal drugs (ranitidine, famotidine.cim etidine , omeprazole and sucralfate) also remained a significant risk factor in the SLFR model (AOR=3.20.95% CI 1.39 to 7.34).None of the other variables entered in the multivariate logistic regression analysis reached statistical significance.

DISCUSSION
We have previously reported an evolving C dijfi.cile outbreak in our institution by tracing strains using numerical analysis of sos-PAGE protein patterns (23) .
We have also previously identified a toxigenic C dijfi.cile clone from the majority of our patients and the environment.The same toxigenic clone has been persistently lransmilted for 25 months among the patients on a medical teaching unit.a nonteaching medical unit, a hematological oncology unit and the intensive care unit (24).In the present study, we have identified and quantified specific antimicrobial and nonantimicrobial drug risk factors that may have contributed to the development of C diffi.cile diarrhea during the sustained outbreak.
Antibiotic exposure has been recognized to be of primary importance in the pathogenesis of C diffi.cile infections ( 1. [3][4][5][6][7] .In this study, antibiotic exposure was s ignificantly higher among the cases.The case patients received antibiotic combinations more frequently than controls.Other published reports of case-control studies of C dijfi.cile outbreaks h ave also shown a significant risk of C dijfi.cile infection associated with multiple antibiotics (11)(12)(13)(14)(15) .Four antibiotics were found to be specifically associated with C difficile diarrhea in this patient population.The major risk factors were the use of ceftazidime and cefuroxime, in add ition to clindamycin.in both the univariate and the SLFR model.a finding also reported by others (9 , 14-16).Among the moderate to high level use antibiotics.use of ciprofloxacin was shown to be an added risk by the regression analysis .The contribution of ciprofloxacin to the development of C diffi.cile diarrhea h as not been demonstrated before this study.Although this study involved a considerable proportion of oncology patients.it did not reveal the use of antineoplastic drugs (cytarabine.cyclophosphamide, doxorubicin.vincristine and etoposide) to be a significant risk factor .However. the cases received significantly more gastrointestinal drugs (ranilidine.famotidine.cimetidine.omeprazole and sucralfate) than controls .An earlier report demonstrated the use of ranitidine and cimetidine to be a risk factor for disease (14) .
Our study may have certain limitations.We did not reveal the risk associated with numbers of previous a dmissions or length of stay.Similarly.we did not investigate the risk associated with malignancy and previous gastrointestinal disease or intervention, which has been stud ied by others (14 , 15) .We have not looked at the overall severity of underlying illness (comorbidity or acute physiology and chronic health evaluation !APAC HE] II score) of cases and controls.The role of these variables and other intrinsic patient characteristics in th e development and nosocomial transmission of C dijfi.cile disease could not be established in our institution .The study findings about medications contributing to the development of C difficile diarrhea may also be limited by the fact that our institution has a structured antibiotic formulary containing a single antibiotic for each antibiotic class.Thus, the impact of antibiotics.the specific gastrointestinal drugs and antineoplastic agents other than those maintained on the formulary could not be assessed.

CONCLUSIONS
This case-control study revealed that ceftazidime and cefuroxime were highly associated with C difficile diarrhea.As anticipated.clindamycin was a significant risk factor .The SLFR model , which considered all variables for entry, indicated that ciprofloxacin in combination with other high risk agents (eg, gastrointestinal drugs) may be linked to C difficile diarrhea in our in-patients.We noted that specific second-and thirdgeneration cephalosporins (cefuroxime and ceftazidime) were more likely associated than other cephalosporins with this infection; these data have not been previously demonstrated (16).In addition, this is the first time that the use of ciprofloxacin has been implicated in the development of C diffi.cile diarrhea.The gastrointestinal drugs (ranitidine, famolidine, cimetidine, omeprazole and sucralfate) were also implicated in the development of C difficile diarrhea.
C dijfi.cile diarrhea continues to be a major nosocomial problem.Efforts to contain hospital epidemics of this infection by improving infection control policies have met with variable success.The evaluation and alteration of antibiotic prescribing practices have been proved to be more effective in decreasing the risk of infection (9).Similar studies on larger scales are warranted so that the strategies to curtail the use of high risk drugs to reduce nosocomial C dijfi.cile diarrhea can be tested .Practices such as mandatory infectious diseases consultation before antibiotic use and structured antibiotic order forms may have an impact on reducing this infection.

TABLE 1
Drugs used in cases and matched controls and relative risks in the univariate analysis and SLFR model