received testing for C. difficile cytotoxin during a 3-month period in 1994. ..... Sox H Jr, Blatt M, Higgins M, Marton K. Medical Decision Making. Boston, Mass: ...
Predicting Clostridium difficile Stool Cytotoxin Results in Hospitalized Patients with Diarrhea David A. Katz, MD, MSc, David W. Bates, MD, MSc, Eve Rittenberg, MA, Andrew Onderdonk, PhD, Kenneth Sands, MD, Laurie A. Barefoot, RN, David Snydman, MD OBJECTIVE: To validate a model for the prediction of Clostridium difficile cytotoxin assay results, and to identify a subgroup of patients with a very low likelihood of C. difficile–associated disease in whom the yield of routine cytotoxin testing is low. DESIGN: Prospective cohort study. Relevant clinical symptoms, signs, and antibiotic exposure were recorded before reporting of assay results. Each predictor was assigned a score based on regression coefficients, and patients were stratified according to their total score. SETTING: Two urban, tertiary care, university hospitals. PATIENTS: A total of 609 consecutive adult inpatients who received testing for C. difficile cytotoxin during a 3-month period in 1994. MEASUREMENTS AND MAIN RESULTS: The prevalence of positive cytotoxin assays was 8% in the validation set, compared with 14% in the derivation set. Defining patients without both prior antibiotic use and at least one symptom predictor (significant diarrhea or abdominal pain) as a low-risk subgroup, the misclassification rate was 2.8% (5/177) for assay results; of the five misclassified cases patients, only one was judged to have C. difficile–associated disease. Use of this rule to identify low-risk patients could have potentially averted 29% of all cytotoxin assays. CONCLUSIONS: Patients without a history of antibiotic use and either significant diarrhea or abdominal pain are unlikely to have positive C. difficile cytotoxin assays and may not require cytotoxin testing. KEY WORDS: Clostridium difficile; routine diagnostic tests; quality of care; utilization. J GEN INTERN MED 1997;12:57–62.
Received from the Division of Clinical Decision Making (DAK) and the Division of Infectious Diseases (LAB, DS), Department of Medicine, New England Medical Center and Tufts University School of Medicine, Boston, Mass.; the Division of General Internal Medicine and Primary Care, Department of Medicine (DWB, ER) and the Division of Microbiology (AO), Brigham and Women’s Hospital and Harvard Medical School, Boston, Mass.; and the Division of Infectious Diseases, Beth Israel Hospital and Harvard Medical School, Boston, Mass. (KS). Presented at the Society of General Internal Medicine 18th annual meeting, San Diego, Calif., May 4–6, 1995. Supported in part by research grant R01-HS07107-02 from the Agency for Health Care Policy and Research. Dr. Katz was supported by a National Library of Medicine research fellowship at the time of this work. Address correspondence and reprint requests to Dr. Katz: Section of General Internal Medicine, Suite 100, 2870 University Ave., Madison, WI 53705.
C
lostridium difficile accounts for substantial morbidity in hospitalized patients and has been recognized to be the most common cause of antibiotic-associated colitis.1 Disease associated with C. difficile is also costly. In one study, C. difficile colitis during hospitalization accounted for $2,000 in additional charges per patient.2 Testing for C. difficile accounted for nearly 68% of total hospital charges associated with the laboratory evaluation of diarrhea in another study.3 The most widely used method for diagnosis of C. difficile–associated disease is based on detecting a cytotoxin produced by C. difficile by means of a tissue culture assay.4,5 The yield of C. difficile toxin testing in hospitalized patients with diarrhea has been reported to be approximately 20% in patients hospitalized longer than 3 days.6–9 Other investigators, however, have suggested that approximately 40% of testing for C. difficile could be eliminated by using selective criteria.10 Moreover, the yield of serial (repeated) tests for C. difficile toxin has been reported to be low.11 In a retrospective study of consecutive patients who underwent first-time testing for C. difficile cytotoxin at a university medical center, where the overall prevalence of positive cytotoxin assays was 14%, Katz et al. developed a simple three-variable decision rule.5 This rule identified a clinical subgroup at low risk of positive cytotoxin assay results (patients without the combination of antibiotic use within 30 days prior to testing and either a significant diarrhea or abdominal pain). This subgroup accounted for 37% of all cytotoxin assays performed. Although the subjects of this earlier study had a broad range of symptoms and exposure histories, they were predominantly white and were treated at a rural, tertiary-care medical center and an affiliated veterans hospital. The testing patterns identified in the earlier study may differ from those in other practice settings. As prospective validation of prediction rules is advisable (ideally at other sites) before they are used in clinical practice,12,13 we undertook the present study to validate the above decision rule for identifying a patient subgroup with a very low likelihood of positive cytotoxin test results and C. difficile–associated disease.
METHODS Study Population All patients, 18 years or older, who had C. difficile cytotoxin assays at the Brigham and Women’s Hospital and at New England Medical Center from August 1 through November 18, 1994, were considered eligible for the study. Both hospitals are large, urban, tertiary-care teaching hos57
58
Katz et al., Predicting Clostridium difficile
pitals that provide primary care and subspecialty services in the Boston (Mass.) metropolitan area. Of the 839 patients screened for inclusion in the study, 230 were excluded for at least one of the following reasons: diagnostic testing for C. difficile on an outpatient basis prior to admission (107 patients); treatment for presumed C. difficile–associated disease within 3 months prior to the test date (31 patients); high suspicion of recurrent or relapsing disease, defined by history of a positive cytotoxin assay within the previous 3 months (19 patients); age less than 18 years (66 patients); missing or incomplete medical records (4 patients); and final cytotoxin test result already reported at the time of data collection (3 patients). The decision to order a diagnostic assay for C. difficile was made by the patient’s clinician.
Stool Cytotoxin Assays All stool specimens at both hospitals were promptly refrigerated at 48C. Assays for stool cytotoxin were generally set up within 24 hours of receipt in the microbiology department of each hospital. Cytotoxin testing was based on the procedure of Chang et al.4 Positive test results were defined by typical cytopathic changes (i.e., rounding of cells, stringing of cellular processes) in the cellular monolayer found over a period of 48 to 72 hours. All positive test results were verified by demonstrating neutralization of cytopathic effect by antitoxin.
Data Collection The dependent variable of interest was the final C. difficile cytotoxin assay result. All test results were obtained directly from the department of microbiology at each hospital. The unit of evaluation was the diarrheal episode, the onset of which was defined as the time that a change in bowel habit was first noted. The index date refers to the submission date for the first stool sample during this period. Subsequent stool samples were not considered in our analysis, as the yield of positive assays was much lower for this group of specimens compared with first submitted specimens.11,14 Demographic data included age and gender. Other descriptive data included primary admitting diagnosis, duration of hospitalization up to the date of testing, hospital service (e.g., medical vs surgical), and level of care (e.g., intensive care unit vs general care unit) at diarrhea onset. Laboratory data included the white blood cell count obtained closest to the time of sample submission. Leukocytosis was defined by the presence of an elevated white blood cell count (. 10 3 109/L), and was coded as “missing” if the test had not been ordered within 48 hours before the index date. Significant diarrhea was defined as the new onset (within 30 days) of at least three partially formed or watery
JGIM
stools, 1 L of colostomy output, or more than 200 ml of watery rectal bag output,15 documented by nursing or physician staff within a 24-hour period. In cases for which nursing notes indicated “multiple” or “several” loose stools without further quantification, we assumed that the patient had significant diarrhea. Abdominal pain was defined as discomfort during the diarrheal episode that was crampy or aching in quality (in contrast to postoperative incisional pain, dyspepsia, or nausea), and was coded as “unknown” for patients who were unresponsive, aphasic, heavily sedated, or otherwise impaired in their ability to communicate pain. Data regarding antibiotic therapy included prior antibiotic therapy (use of any antibiotic within 30 days of the index date, including antibiotics for which no association with C. difficile–associated disease has been reported), and class of antibiotic (e.g., penicillin, aminoglycoside, cephalosporin, sulfa, quinolone). In assessing prior antibiotic use, we included exposure to chemotherapeutic agents with antimicrobial properties (e.g., doxorubicin).16 Preadmission antibiotic use was obtained from the admitting history, records from referring hospitals, and outpatient clinic notes. Data regarding other potential causes of diarrhea included use of magnesium-containing antacids, as well as laxatives, stool softeners, or phospho-soda enema. These items were considered contributory causes of diarrhea if present within 24 to 48 hours prior to or concurrent with the index episode of diarrhea. The hospital charts of all adults reported to have had C. difficile cytotoxin assays were reviewed on the date of specimen submission or shortly thereafter by one of the investigators. All evaluations were performed before the cytotoxin test results were reported. Data forms were spot checked for accuracy; missing data or questions regarding specific items prompted review of the medical record by one of the other investigators. Disagreements between reviewers were noted and resolved by consensus.
Interpretation of Positive C. difficile Test Results For cytotoxin-positive patients, we presented a composite sketch (including reason for hospital admission, antibiotic exposure, gastrointestinal symptoms, other medical causes for diarrhea, results of diagnostic workup, and response to specific therapy, if any) to a panel of three experts on infectious disease who were blinded to model predictions. Guided by a modified definition of C. difficile– associated disease previously used in clinical studies,17 each of these experts independently judged each positive test to represent either a true-positive or a false-positive result. If all three specialists agreed on a result, it was classified as a definite true-positive or definite false-positive result. If only one or two specialists considered the assay result to correspond to true C. difficile–associated disease, the assay was considered a possible true-positive result.
JGIM
59
Volume 12, January 1997
Data Analysis Each predictor was assigned a score based on the previously derived model. Specifically, the b coefficient for each predictor was divided by 0.63 (the coefficient for abdominal pain, which was the weakest predictor), and rounded to the nearest integer. (The intercept and regression coefficients in this logistic regression model are intercept [24.10], abdominal pain [0.63], significant diarrhea [0.80], and previous antibiotic use [2.20].) The risk score for each patient was calculated by assigning points for each factor present and summing the results. Model accuracy in discriminating between patients with and without a positive assay in the validation set was compared with that in the derivation set using receiver-operating characteristic (ROC) curve analysis.18,19 Calibration curves
were constructed to assess the correlation between observed and predicted probabilities of a positive cytotoxin assay over the range of predicted probabilities.
RESULTS Of all eligible adult inpatients whose stool was tested one or more times during the study period, 8% had one or more positive tests. Of the clinical criteria conventionally associated with C. difficile colitis, prior antibiotic use and significant diarrhea were documented more frequently in test-positive patients, but there was no significant difference in the proportion of test-positive patients and testnegative patients with abdominal pain (Table 1). Moreover, test-positive patients were more likely to have received
Table 1. Comparison of Patients with Positive and Negative Clostridium difficile Cytotoxin Assays (n 5 609) Variable* General descriptors Mean age (6SD), years Male, n (%) Intensive Care Unit residence on date of diarrhea onset, n (%) Hospital service on date of diarrhea onset, n (%)† Medical Surgical Obstetrics-gynecology Duration of hospitalization, (IQR)‡ Primary admitting diagnosis, n (%) Cardiovascular Malignancy-related Gastrointestinal Immunocompromised Infection Other Clinical Variables Maximum bowel movements per day, (IQR) Significant diarrhea, n (%)§,i Abdominal pain, n (%)¶,i Peripheral leukocyte count (3109/L), (IQR)# Potential contributing causes of diarrhea Antibiotic use within 30 days, n (%)i Cephalosporin use Antibiotic use prior to admission or transfer Antacid use, n (%) Laxative and stool softener use, n (%)
Assay Positive (n 5 49)
Assay Negative (n 5 560)
p Value
58 6 19 21 (43) 5 (10)
56 6 18 277 (49) 106 (19)
12 14 2 4
(5) (9) (11) (2–9)
220 147 16 5
(95) (91) (89) (2–9)
.46
10 5 9 7 6 12
(18) (10) (18) (16) (12) (26)
101 109 115 41 82 112
(18) (19) (21) (7) (15) (20)
.95 .11 .72 .03 .65 .45
.18 .38 .10 .13†
5 (3–5) 39 (83) 13 (30) 12.6 (6.0–13.0)
4 (3–6) 388 (71) 120 (24) 9.0 (7.3–16.2)
.05 .08 .38 .05
48 36 23 10 24
470 277 166 58 237
.009 .001 .009 .04 .33
(98) (73) (51) (20) (50)
(84) (49) (32) (10) (43)
* The median and interquartile range (IQR) are reported for all continuous variables except age. † For comparison between medical and surgical services (including obstetrics and gynecology); this analysis does not include 128 patients who developed diarrhea prior to admission to a study hospital and 67 patients for whom this information was unknown. ‡ Duration of hospitalization is calculated up to the date of stool sample submission. § Two patients with positive assays and 14 patients with negative assays were excluded, because their stool patterns were not documented. i Variables used in previously derived prediction models. ¶ Six and 65 patients with positive and negative assays, respectively, were excluded, because they were unable to communicate pain or because abdominal pain was not documented. # Forty-one patients with negative assays were excluded because leukemia or severe neutropenia was present, and 21 were excluded because no leukocyte count was available; no patients with positive assays were excluded for these reasons.
60
Katz et al., Predicting Clostridium difficile
Table 2. Risk Stratification Based on Predictive Model* Risk Group (Score)†
Assay Positive/Total (%)
0–1 2–3 4–5
1/70 (1.4) 4/107 (3.7) 36/345 (10.4)
*Model includes prior antibiotic use (3 points), significant diarrhea (1 point), and abdominal pain (1 point). † Analysis is based on 522 cases with complete data for model variables.
antibiotics prior to admission or transfer and were also more likely to have reported onset of diarrhea prior to admission to study hospitals (13% vs 7%, p 5 .05).
Model Evaluation The model successfully stratified patients according to likelihood of cytotoxin positivity and disease (Table 2). Based on these data, we considered patients with a total score of 3 or less (i.e., those without both previous antibiotic use and at least one symptom predictor) to be at low risk of cytotoxin positivity; 179 (29%) of 609 study patients were in this category. When ROC curve areas for the model were compared in the derivation and validation sets, the area in the derivation set was 0.69 and the area in the validation set was 0.64; this difference in ROC curve area was not statistically significant. Calibration curves demonstrated excellent agreement between observed and predicted values (Fig. 1). Our model misclassified one patient who was predicted to be at low risk but was judged to have C. difficile– associated disease by at least one expert (1/179, false reassurance rate 5 0.6%). This patient was admitted for management of dilated cardiomyopathy and had received both ceftazidime and clindamycin. While in the hospital, she received a stool softener and developed mild diarrhea, which did not satisfy our definition of significant diarrhea and was not associated with either abdominal pain or leukocytosis.
JGIM
false-negative case patients, only one was judged to have C. difficile–associated disease by a panel of experts. Other investigators have used a combination of clinical and laboratory criteria to predict C. difficile test results. On the basis of four variables (hospital stay of at least 4 days, loose stool, erythrocytes or leukocytes seen on direct microscopy, and history of exposure to antibiotics or cytotoxic therapy), Bowman and colleagues determined that 17% of first stool samples received from inpatients and outpatients did not qualify for routine testing of C. difficile. Of those cases in which testing was not indicated, 3.6% of cultures were positive.20 In a recent study, Manabe et al. reported that cephalosporin use, the presence of semiformed stool, and the presence of fecal leukocytes were independently predictive of C. difficile toxin results.14 Our study has the important advantage that it used clinical criteria alone to define a patient subgroup at very low risk of C. difficile–associated disease. Using clinical data alone is much more practical and economical than using one or more laboratory tests to decide whether or not to perform others. Our analysis confirms the strong association between cephalosporin use and cytotoxin positivity reported by other investigators.14,17,21–24 Although our local experience revealed cephalosporin use to be predictive of cytotoxin results in both the validation and derivation sets, others have demonstrated associations for different classes of antibiotics,25–27 and have shown reduced rates of C. diffi-
DISCUSSION Application of a simple decision rule to identify lowrisk patients in our study population (order test if both prior antibiotic use and either significant diarrhea or abdominal pain are present) could have potentially averted 29% of all cytotoxin assays in hospitalized patients who had not received previous testing or treatment for C. difficile, compared with 37% in the derivation set. Based on 1993 billing data, this reduction in first-time assays could translate into a yearly savings of approximately $12,000 for each hospital (320 tests at $37.50 each), if current testing patterns remained unchanged. Using this decision rule, only five low-risk patients (3% of our study population) would have been incorrectly predicted to be test-negative, when in fact they were test-positive. Of these five
FIGURE 1. Observed versus predicted probability of cytotoxin positivity for predictive model. The 45-degree diagonal line corresponds to perfect agreement between observed and predicted values. Each solid block represents the mean probability for a subgroup of study patients with one of the following unique combinations of predictors: no prior antibiotic use (score 5 0–2), prior antibiotic use only (score 5 3), prior antibiotic use and one symptom present (significant diarrhea or abdominal pain) (score 5 4), and prior antibiotic use and both symptoms present (score 5 5).
JGIM
61
Volume 12, January 1997
cile–associated disease when the usage of these antibiotics was restricted.28 Thus, to enhance the generalizability of our models to other hospital settings, we did not present models including specific antibiotic variables (e.g., prior cephalosporin use) in the current study. We also found that an increased peripheral white blood cell count was significantly associated with a positive cytotoxin test result, as reported by others.29 In the current study, a model substituting leukocytosis for abdominal pain demonstrated test performance comparable to that of the original predictive model (data not shown). Such an alternative rule for test-ordering decisions may be particularly useful for those patients in whom abdominal pain is unknown or otherwise not evaluable. However, because leukocytosis could not be adequately assessed in the derivation set owing to missing data, this alternative model requires further validation in an independent study sample. Several variables that have been implicated in previous studies, such as fever,25 severity of illness,17 or hospital ward,27 were not included in our predictive model. In the derivation set,5 both maximum temperature (on the index date) and comorbidity score were nearly equivalent for testpositive and test-negative subjects, and thus were not assessed in the current study. Although some hospital units had episodic clusters of cytotoxin-positive cases, utilization of the cytotoxin assay remained relatively constant throughout medical and surgical units of both hospitals. Among patients who developed diarrhea while in the hospital, no clear patterns linking hospital service (e.g., medical vs surgical) or unit (ward vs intensive care) to cytotoxin assay results emerged. Our finding of high cytotoxin positivity rates in patients who developed diarrhea prior to hospitalization is consistent with reports that colonized patients who are new hospital admissions and transfers are a major source of C. difficile infection.30,31 Although there is no definitive “gold standard” for the diagnosis of C. difficile–associated disease, the cytotoxin assay remains a valid marker of this disease. However, the following limitations of the test should be considered. False-negative tests may reflect undetectable quantities of cytotoxin in patients with early and mild disease, the production of toxins not identified by the assay, inactivation of toxin during transport,32 and decreased toxin sensitivity of certain cell lines.33 Positive tests in the absence of disease may occur in up to 5% of healthy adults and hospital control patients.25,34 The predictive models in this study showed limited ability to discriminate positive from negative outcomes (as indicated by ROC curve areas in the 0.6–0.7 range). However, they did accurately define a patient subgroup at low risk of C. difficile–associated disease, which is clinically most relevant. Moreover, we found excellent agreement between observed and predicted probabilities of C. difficile–associated disease, despite a significant difference in the baseline prevalence of cytotoxin-positive subjects in the two study populations.35 An advantage of the validation set is that it included a more diverse group of pa-
tients than that of the derivation set. However, we emphasize that both the derivation and validation sets were limited to hospitalized adults; thus, the prediction rule was not designed to be utilized for test-ordering decisions in children and outpatients. Choice of a more conservative cutoff for test ordering may be appropriate for some patients, such as the immunocompromised. Clinicians who have low treatment thresholds may still opt to order the cytotoxin assay even in patients predicted to be at low risk of cytotoxin positivity.36 It is unlikely, however, that the benefit of performing large numbers of cytotoxin assays on all patients in this low-risk subgroup to identify a rare positive result justifies the associated cost, especially as an alternative strategy to immediate testing is continued observation, with the option to test for cytotoxin if there is no clinical improvement. In conclusion, inpatients without prior antibiotic use and either significant diarrhea or abdominal pain are unlikely to have C. difficile–associated disease. Elimination of testing for C. difficile in this low-risk group would have reduced the number of cytotoxin assays performed by 29% in the current study. Physicians should remain vigilant in managing patients initially predicted to be at low risk, however, and reconsider testing for C. difficile if these patients fail to improve or worsen despite adequate supportive therapy (including the cessation of all nonessential offending agents). We emphasize that our decision rule was not intended to guide testing for C. difficile during outbreak investigations or related infection-control activities. With these caveats in mind, we believe that our decision rule can be used by physicians to develop more effective strategies for detecting C. difficile–associated disease in hospitalized patients with diarrhea and may result in substantial savings. The authors thank Richard Platt, MD, for his review of the C. difficile cytotoxin-positive cases.
REFERENCES 1. Fekety R. Antibiotic-associated colitis. In: Mandell G, Douglas R, Bennett J, eds. Principles and Practice of Infectious Disease. New York, NY: Churchill Livingstone; 1990. 2. Kofsky P, Rosen L, Reed J, Tolmie M, Ufberg D. Clostridium difficile—a common and costly colitis. Dis Colon Rectum. 1991;34: 244–8. 3. Yablon S, Krotenberg R, Fruhmann K. Clostridium difficile–related disease: evaluation and prevalence among inpatients with diarrhea in two freestanding rehabilitation hospitals. Arch Phys Med Rehabil. 1993;74:9–13. 4. Chang T, Lauermann M, Bartlett J. Cytotoxicity assay in antibiotic-associated colitis. J Infect Dis. 1979;140(5):760–70. 5. Katz D, Lynch M, Littenberg B. Clinical predictors of Clostridium difficile stool cytotoxin results in hospitalized patients with diarrhea. Am J Med. 1996;100:487–95. 6. Siegel D, Edelstein P, Nachamkin I. Inappropriate testing for diarrheal diseases in the hospital. JAMA. 1990;263(7):979–82. 7. Koplan J, Fineberg H, Ferraro M, Rosenberg M. Value of stool cultures. Lancet. 1980;2(8191):413–6. 8. Fan K, Morris A, Reller L. Application of refection criteria for stool
62
9.
10. 11.
12. 13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
Katz et al., Predicting Clostridium difficile
cultures for bacterial enteric pathogens. J Clin Microbiol. 1993;31:2233–5. Yannelli B, Gurevich I, Schoch P, Cunha B. Yield of stool cultures, ova and parasite tests, and Clostridium difficile determinations in nosocomial diarrheas. Am J Infect Control. 1988;16:246–9. Bowman R, Riley T. Routine culturing for Clostridium difficile? Pathology. 1984;16:240–2. Renshaw AA, Stelling JM, Doolittle MH. The lack of value of repeated Clostridium difficile cytotoxicity assays. Arch Pathol Lab Med. 1996;120:49–52. Centor R, Yarbrough B, Wood J. Inability to predict relapse in acute asthma. N Engl J Med. 1984;310:577–80. Rose C, Murphy J, Schwartz J. Performance of an index predicting the response of patients with acute bronchial asthma to intensive emergency department treatment. N Engl J Med. 1984;310:573–7. Manabe Y, Vinetz J, Moore R, et al. Clostridium difficile colitis: an efficient clinical approach to diagnosis. Ann Intern Med. 1995;123:835–40. Sleisenger M, Fordtran J, eds. Gastrointestinal Disease: Pathophysiology, Diagnosis, Management. 4th ed. Philadelphia, PA: WB Saunders; 1989. Kamthan A, Bruckner H, Hirschman S, Agus S. Clostridium difficile diarrhea induced by cancer chemotherapy. Arch Intern Med. 1992;152(8):1715–7. McFarland L, Surawicz C, Stamm W. Risk factors of Clostridium difficile carriage and C. difficile–associated diarrhea in a cohort of hospitalized patients. J Infect Dis. 1990;162(3):678–84. Hanley J, McNeil B. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology. 1983;148:839–43. Centor R. A Visicalc program for estimating the area under the receiver operating characteristic (ROC) curve. Med Decis Making. 1985;5:139–48. Bowman R, Bowman J, Arrow S, Riley T. Selective criteria for the microbiological examination of faecal specimens. J Clin Pathol. 1992;45:838–9. Bartlett J, Willey S, Chang T, Lowe B. Cephalosporin-associated pseudomembranous colitis due to Clostridium difficile. JAMA. 1979;242:2683–5. Zimmerman R. Risk factors for Clostridium difficile cytotoxin-positive diarrhea after control for horizontal transmission. Infect Con-
JGIM
trol Hosp Epidemiol. 1991;12:96–100. 23. Thibault A, Miller M, Gaese C. Risk factors for the development of Clostridium difficile–associated diarrhea during a hospital outbreak. Infect Control Hosp Epidemiol. 1991;12:345–8. 24. Jobe B, Grasley A, Deveney K, Deveney C, Sheppard B. Clostridium difficile colitis: an increasing hospital-acquired illness. Am J Surg. 1995;169:480–3. 25. Gerding D, Olson M, Peterson L, et al. Clostridium difficile–associated diarrhea and colitis in adults: a prospective case-controlled epidemiologic study. Arch Intern Med. 1986;146:95–100. 26. Bartlett J. Antimicrobial agents implicated in Clostridium difficile toxin-associated diarrhea or colitis. Johns Hopkins Med J. 1981;149:6–12. 27. Brown E, Talbot G, Axelrod P, Provencher M, Hoegg C. Risk factors for Clostridium difficile toxin-associated diarrhea. Infect Control Hosp Epidemiol. 1990;11(6):283–90. 28. Pear S, Williamson T, Gettin K, Gerding D, Galgiani J. Decrease in nosocomial Clostridium difficile–associated diarrhea by restricting clindamycin use. Ann Intern Med. 1994;120(4):272–7. 29. Jensen G, Bross J, Bourbeau P, et al. Risk factors for Clostridium difficile stool cytotoxin b among critically ill patients: role of sucralfate. J Infect Dis. 1994;170:227–30. 30. Clabots C, Johnson S, Olson M, Peterson L, Gerding D. Acquisition of Clostridium difficile by hospitalized patients: evidence for colonized new admissions as a source of infection. J Infect Dis. 1992;166(3):561–7. 31. Samore M, Bettin K, DeGirolami P, et al. Wide diversity of Clostridium difficile types at a tertiary referral hospital. J Infect Dis. 1994;170(3):615–21. 32. Lyerly D, Krivan H, Wilkins T. Clostridium difficile: its disease and toxins. Clin Microbiol Rev. 1988;1(1):1–18. 33. Maniar AC, Williams T, Hammond G. Detection of Clostridium difficile toxin in various tissue culture monolayers. J Clin Microbiol. 1987;25(10):1999–2000. 34. Fekety R. Diagnosis and treatment of Clostridium difficile colitis. JAMA. 1993;269(1):71–5. 35. Poses R, Cebul R, Collins M, Fager S. The importance of disease prevalence in transporting clinical prediction rules: the case of streptococcal pharyngitis. Ann Intern Med. 1986;105:586–91. 36. Sox H Jr, Blatt M, Higgins M, Marton K. Medical Decision Making. Boston, Mass: Butterworths; 1988.
