INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Diarrhea still remains a common complaint of both hospitalized patients and patients seen in an outpatient setting. Standard tests for determining diarrheal etiology, such as stool culture (SCx) and the Clostridium difficile cytotoxin detection assay (CDTA), are accurate but generally too slow to be helpful to the clinic or emergency room physicians who are diagnosing and treating the patient. This has led to the use of several more rapid tests (fecal leukocyte [FL] stain, fecal occult blood test, fecal lactoferrin test) that are currently used to suggest disease with a bacterial pathogen while awaiting stool culture results. Unfortunately, these tests are not predictive in all situations (4-6, 10-11).

The methylene blue stain for FLs represents a widely used adjunct to slower but more accurate tests of diarrheal etiology, such as SCx or CDTA. Willmore and Shearman (15) first described the FL stain in 1918 to help identify bacillary and amebic dysentery. In 1972, Harris et al. (4) found that the presence of FLs indicated a bacterial cause of acute diarrhea in 89% of cases. FLs were also found in up to 50% of cases of C. difficile infection (1).

Despite the widespread use of FLs to predict a positive SCx, recent studies have found conflicting results (5-6, 10, 12). Similarly, the performance characteristics of FLs for predicting C. difficile infection have varied widely (2, 7). One reason for the conflicting results may be that these studies were performed in a variety of settings, both inpatient and outpatient, with various criteria for ordering the tests. No studies have directly compared the utility of FLs for predicting CDTA and SCx results by patient location. To define the most appropriate clinical setting for ordering these tests, we examined the utility of FL for predicting CDTA and SCx results among inpatients and outpatients at our hospital. Furthermore, published threshold values define a "positive" test range from 1 white blood cell/high-power field (WBC/HPF) (12, 17) to >50 WBC/HPF (13). Since a prior comparison of diagnostic threshold values did not identify a superior cutoff (10), we used both 1 and 5 WBC/HPF values as FL threshold values to determine which is more clinically useful.

(These results were presented, in part, as an abstract at the 1999 annual meeting of the Infectious Diseases Society of America [abstr. 259, p. 84].)

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Study population. We conducted a retrospective cohort study of all patients who submitted a diagnostic stool specimen to the Stanford Hospital Clinical Microbiology Laboratory from 1 May 1998 through 30 April 1999. Patients who submitted a stool sample for FL analysis and either CDTA, SCx, or both methods were included in the study. The decision to order a stool study was made by the patient's clinician. A maximum of one stool sample per patient (the first sample collected) was analyzed. Stool samples that were described as "formed stool without blood" by the laboratory were excluded. Patient location (inpatient or outpatient status) was assessed, but the acuity of illness and length of stay at the time of sample collection was not. Outpatient location included specimens received either from the emergency department or from hospital-associated outpatient clinics. Testing characteristics (sensitivities, specificities, and likelihood ratios [LRs]) for FL were compared with the presence of C. difficile toxin and SCx results for both inpatients and outpatients.

FL method. One drop of fresh (<3 h since collection) stool was mixed with an equal proportion of Loeffler's methylene blue stain (0.2% methylene blue). The suspension was examined under high-dry power (HPF) (×400) for the presence of polymorphonuclear leukocytes (WBCs). Positive results were analyzed at either 1 or 5 WBC/HPF.

SCx method. All stool specimens submitted either fresh (<3 h since collection) or in Cary-Blair preservative were inoculated onto Trypticase soy agar with 5% sheep blood, MacConkey agar, Hektoen Enteric, MacConkey sorbitol (if stool was either unformed or formed with blood, for Escherichia coli O157:H7), and brucella blood agar with trimethoprim, vancomycin, and polymyxin (for Campylobacter) plates. If Vibrio sp. was suspected clinically, TCBS agar was added. Plates were incubated in air at 35°C except for the Campylobacter plates, which were incubated in a microaerophilic atmosphere at 42°C. All plates were examined daily for 3 days before discarding them as negative. Sweep oxidase testing from the blood agar was used to detect Aeromonas, Plesiomonas, and Vibrio spp. Suspicious colonies were identified using standard methods (9). This method is able to detect Salmonella, Shigella, Aeromonas, Plesiomonas, Campylobacter, Vibrio, or Yersinia species and E. coli O157:H7 (in unformed or formed bloody stools).

Toxin assays for C. difficile. The Stanford Hospital Microbiology Laboratory refrigerated fresh stool specimens at 4°C immediately upon receipt. The stool was diluted 1:6, and filtrate was used to inoculate tissue culture tubes containing human diploid embryonic fibroblast (MRC-5) cell culture monolayers. Each sample was tested in duplicate: stool only and stool plus cytotoxin B antitoxin (Tech Lab, Blacksburg, Va.) Tubes were incubated at 35°C and examined microscopically three times daily for 2 days for typical toxin cytopathic effect (CPE) and neutralization of toxicity. The presence of CPE in stool filtrate and the absence of CPE in the corresponding antitoxin-containing tube was considered positive for C. difficile toxin B. If the CDTA showed nonspecific toxicity, the original stool (from a frozen aliquot) was tested for cytotoxin A using the Meridian Immunocard Toxin A Test (Meridian Diagnostics, Kansas City, Mo.). A positive result by either method was considered positive for C. difficile toxin.

Statistical analysis. Results from all stool specimens entered into the database of the Stanford University Diagnostic Microbiology Laboratory were analyzed (Stata 6.0; College Station, Tex.) after expunging personal identifiers. Only the first unformed specimen from each patient was analyzed, and only if the specimen was tested for FLs in conjunction with either or both CDTA and SCx methods. LRs for each test were calculated as follows: LR = [sensitivity/(1  specificity)] (3). The LR is a predictive number that evaluates the utility of the test for evidence-based clinical decision-making. Briefly, the pretest odds of a specific outcome (e.g., being infected with Campylobacter sp.), multiplied by the LR, yields the posttest odds of that outcome. The utility of FL for predicting C. difficile disease or bacterial diarrhea was analyzed separately, despite the fact that some specimens had both the CDTA and the SCx ordered. Results are presented for each subgroup (FL method versus CDTA and FL method versus SCx) separately, resulting in total numbers higher than the number of specimens (see Tables 1 and 2).

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Of 797 stools that were tested by the FL method and at least one confirmatory test, 502 were tested by CDTA, 473 were cultured, and 238 were tested by both methods (Table 1). Of the 502 tested by CDTA, 409 (81%) were inpatients and 93 (19%) were outpatients. Of the 473 tested by SCx, 273 (58%) were inpatients and 200 (42%) were outpatients (Table 1). The prevalence of a positive CDTA result was 6% (29 of 502). The prevalence of a positive SCx in this population was also 6% (27 of 473). The most common bacterial pathogens isolated were Campylobacter spp. (11 isolates) and Shigella spp. (8 isolates), followed by Salmonella spp. (3 isolates), E. coli O157:H7 (2 isolates), and Aeromonas, Vibrio, and Yersinia spp. (1 isolate each). Of all specimens tested by either CDTA, SCx, or both, 12% (97 of 797) were positive for FLs. Overall, 56 (10%) of 542 inpatients and 41 (16%) of 255 outpatients were positive for FLs (Table 1).

FLs for the diagnosis of C. difficile infection. The presence of 1 WBC/HPF was only 14% sensitive and 90% specific for C. difficile and did not predict the 29 positive CDTA results (LR, 1.4; 95% confidence interval [CI], 0.5 to 3.7) (Table 2). The sensitivity and specificity did not vary with patient location (Table 2), nor did it vary significantly when analyzing the subgroup of 234 patients with negative SCx results, from which 11 specimens were CDTA positive and 4 of the 11 had FLs. Of the 223 CDTA-negative specimens, 40 (18%) also had a positive FL result. A stricter threshold of 5 WBC/HPF was less sensitive than the 1 WBC/HPF for predicting a positive CDTA result. The presence of 5 WBC/HPF predicted 2 of 21 (10%) inpatients to be positive for C. difficile, none of 8 outpatients, and 2 (7%) of all 29 patients with positive CDTA results. The LRs for each group were not different from those with only 1 WBC/HPF (Table 2).

FLs for the diagnosis of enteroinvasive bacterial diarrhea. A total of 1% (4 of 273) of inpatients and 12% (23 of 200) of outpatients had a positive SCx (Table 1). The FL test was positive in 14 (52%) of 27 specimens yielding an enteric pathogen but was also positive for 55 (12%) of 446 with negative cultures (Table 1). The sensitivity of 1 WBC/HPF for predicting a positive SCx result was twice as high (57%) among outpatients as among inpatients (25%) with similar specificities (Table 2). Moreover, the presence of 1 WBC/HPF had predictive value for SCx for outpatients (LR, 5.0; 95% CI, 2.9 to 8.6; P < 0.001) but not for inpatients (LR, 1.9; 95% CI, 0.3 to 10.8; P = 0.5). Among outpatients, using 5 WBC/HPF decreased the sensitivity but not the predictive value (LR, 5.1; 95% CI, 2.5 to 10; P < 0.001) (Table 2). Of 234 culture-negative specimens for which CDTA tests were also ordered, the sensitivity of either WBC threshold for predicting a positive CDTA result was 18% (2 of 11 specimens), and the specificities for 1 and 5 WBC/HPF were 87% (194 of 223) and 95% (212 of 223), respectively.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

This study is the first to compare the utility of FL for predicting CDTA and SCx results by patient location. Prior studies have not stratified patients by admission status and have drawn different conclusions about the utility of the FL method in the evaluation of diarrhea (1-2, 4-6, 10, 12, 17). We also sought to determine which diagnostic threshold, 1 or 5 WBC/HPF, is more clinically useful since previous studies have not answered this question (5, 10).

Based on our data, routine use of the FL method for patients with a clinical suspicion of C. difficile infection is not indicated. The FL method was a poor predictor of C. difficile infection overall, regardless of the patient status. Moreover, the sensitivities and LRs remained poor even in the subgroup of patients with negative stool cultures, a group one might expect to have C. difficile infection as an alternative explanation for their diarrhea. Decreasing the threshold for a positive result from 5 to 1 WBC/HPF did not improve the predictive value of the test. Because the sensitivity of FL for C. difficile was low and because the FL test did not predict C. difficile results in inpatients or outpatients, we would not advocate ordering the FL test in any setting to evaluate for C. difficile infection. Our laboratory no longer accepts such requests.

In this study we found, as have others, that SCx results are rarely positive for inpatients (12, 14). This is not surprising since bacterial diarrheas are rarely nosocomial. Of the small number that was positive, the FL test was a poor predictor. Among outpatients, however, the presence of 1 WBC/HPF was positive in >50% of specimens with positive SCx results and was statistically significantly associated with a positive result (LR, 5.0; 95% CI, 2.9 to 8.6; P < 0.001). Furthermore, the utility of the FL method was similar regardless of the threshold used. Thus, if a clinician strongly suspects a bacterial diarrhea and the FL test is positive, it is more likely that the patient will have a positive stool culture. Moreover, the negative predictive values for either FL threshold are high (94 and 92%, respectively). A negative FL test in an outpatient suspected of having bacterial diarrhea suggests that the suspected etiology is incorrect. Based on such a result, the clinician may opt not to treat the patient with antibiotics while awaiting the culture result or not to order a stool culture at all.

The low prevalence of positive CDTA or SCx results at our institution (6%) may explain in part why the FL test was not more sensitive. These prevalences are similar to those in most U.S. institutions (14). This implies that too many tests are being ordered and that, perhaps, in addition to better screening tests for inpatients, better prediction rules are needed at our institution. For example, prior studies have found that factors such as hospital readmission status, leucocytosis, cephalosporin use, prolonged hospital stay, and the onset of diarrheal symptoms at >6 days after the administration of antibiotics are predictive of a positive C. difficile toxin result (2, 7). Rigorous guidelines incorporating such risk factors could help define patients who warrant further testing for the evaluation of their diarrhea and could increase the cost-effectiveness of such tests. Importantly, the present study is too small to address the relevance of FLs in predicting verotoxigenic E. coli infection, where antibiotics are contraindicated or Salmonella infection, where antibiotics are of limited usefulness (16). In fact, the FL test may be more useful for predicting the course and severity of disease and less useful for identifying the need for antibiotics.

This study is the first to show that differences in the usefulness of the FL test may reflect the type of diarrhea the patient has (whether C. difficile diarrhea or some other type of diarrhea). Specifically, this study showed that the FL test is of minimal use in the detection of C. difficile infection regardless of who is tested or how a positive test is defined. Moreover, SCx results are rarely positive in inpatients, and physicians should not order them. Fortunately, most laboratories appropriately limit such tests to samples from patients within 3 days of admission (9, 14). On the other hand, judicious use of the FL test for the evaluation of diarrhea in the clinic or emergency department, when combined with clinical and laboratory parameters, may help to suggest appropriate treatment strategies. Nevertheless, better screening tests and clinical prediction rules are needed.