| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
* From the Division of Infectious Diseases (Dr. Gupta) and the Division of Gastroenterology (Dr. Imperiale), Indiana University School of Medicine (Dr. Sarosi), Indianapolis, IN.
Correspondence to: Samir K. Gupta, MD, Indiana University School of Medicine, Whishard Hosptial OPW-430, 1001 W 10th St, Indianapolis, IN 46202; e-mail: sgupta1{at}iupui.edu
 |
Abstract |
|---|
 TOP Abstract IntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
Design: Retrospective case-control study.
Setting: An urban county hospital and a tertiary-care Veterans Affairs hospital.
Patients: Thirty-seven patients with Legionella pneumonia (diagnosed by a positive result of a urinary Legionella antigen test) and 31 patients with bacteremic pneumococcal pneumonia. A subgroup of patients with all required laboratory criteria were studied further.
Results: The WUH criteria correctly identified 29 of 37 patients with Legionella pneumonia (sensitivity, 78%; 95% confidence interval [CI], 61 to 90%), while successfully excluding legionellosis in 20 of 31 patients with bacteremic pneumococcal pneumonia (specificity, 65%; 95% CI, 45 to 80%). The positive and negative predictive values, adjusted for a relative prevalence of 1:3 between Legionella and Streptococcus pneumoniae bacteremia, were 42% (95% CI, 25 to 61%) and 90% (95% CI, 74 to 97%), respectively. In the subgroup analysis, the WUH criteria were successful in identifying 20 of 23 patients with Legionella pneumonia (sensitivity, 87%; 95% CI, 65 to 97%), while excluding legionellosis in 9 of 18 patients with bacteremic pneumococcal pneumonia (specificity, 50%; 95% CI, 27 to 73%). The adjusted positive and negative predictive values for a 1:3 relative prevalence were 37% (95% CI, 20 to 59%) and 92% (95% CI, 62 to 98%), respectively. The predictive values were changed in the directions expected for an increased relative prevalence of 1:1. The areas under the receiver operating characteristic curves were 0.72 ± 0.06 for the entire study group and 0.68 ± 0.09 for the subgroup.
Conclusions: Although the WUH criteria discriminated fairly well between cases (mean ± SE) and control subjects, the sensitivity is not high enough to exclude legionellosis confidently. These results suggest that empiric therapy for Legionella pneumonia should be included in the initial antibiotic regimen for hospitalized patients with CAP.
Key Words: atypical • clinical decision aids • community-acquired infections • Legionella • legionellosis • pneumococcal pneumonia • pneumonia
|
Introduction |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
In 1998, Cunha15 published a weighted point evaluation scale, the Winthrop-University Hospital (WUH) criteria, to identify legionellosis based on clinical criteria. This comprehensive set of criteria incorporates most clinical symptoms and signs that previously were believed to be associated with legionellosis16 17 18 19 20 and arbitrarily assigns points that are weighted more heavily for extrapulmonary manifestations that are thought to be more specific for this etiology of pneumonia. Negative points are given for signs or symptoms classically associated with other pathogens, such as pneumococcus or Mycoplasma. Cunha15 successfully tested this scale with three case reports of legionellosis from the medical literature, but no systematic evaluation has been performed.
We hypothesized that this scale would be neither highly sensitive in diagnosing Legionella pneumonia nor satisfactorily discriminatory against another lethal etiology of CAP, Streptococcus pneumoniae. Legionellosis is common in our patient population and may be comparable in presentation to pneumococcal pneumonia.7 21 22 23 24 25 Therefore, we attempted to evaluate the utility of these clinical criteria to identify Legionella pneumonia at the time of hospital admission for CAP by performing a retrospective case-control study between groups of hospitalized patients known definitively to have either Legionella or pneumococcal pneumonia.
|
Materials and Methods |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
Inclusion and Exclusion Criteria
Patients admitted to the hospital between January 1994 and
November 1999 were screened for inclusion in the study by examining the
laboratory records for positive results of urine Legionella antigen
tests and for blood cultures positive for S pneumoniae. All
patients included in the study had initial clinical presentations and
chest radiograph findings that were consistent with CAP. Patients with
positive results of urine Legionella antigen tests but no evidence of
coinfection were included in the Legionella case group. The control
group with pneumococcal pneumonia consisted of patients with at least
two serial blood cultures that were positive for S
pneumoniae but negative for the urinary Legionella antigen.
Patients were excluded from the study if there was any evidence of
coinfection (ie, by sputum Gram’s stain, culture, or
serology) for any other pathogen. Patients also were excluded from the
study if their clinical history either was unavailable or incompatible
with acute CAP.
Scoring System, Variables Studied, and Definitions
The WUH point scoring scale is shown in Table 1
with three hypothetical examples of its use. A score of < 5 makes
legionellosis unlikely, a score from 5 to 9 makes the diagnosis
probable, and a score of 
10 makes legionellosis highly probable.
These cutoffs were arbitrarily chosen by Cunha.15
Data were considered abnormal if the values were outside the range of
normal for the particular assay used for each hospital at the time of
hospital admission (B. Cunha, MD; personal communication;
December 1, 1999). According to Cunha, any otherwise unexplained
abnormality of laboratory values from the time of hospital admission
through the first 7 days of hospitalization (ie, as the
disease progresses and becomes more apparent) may be used to fulfill
the WUH criteria. Relative bradycardia (ie, an
inappropriately low heart rate for the elevation of temperature) has
been defined previously.15
Increases in serum creatinine,
bilirubin, or transaminase levels or decreases in serum sodium or
phosphorus levels were considered to be positive results by Cunha only
if they were transient, as opposed to persistently abnormal despite
appropriate treatment. This would prevent the inclusion of patients
with non-Legionella causes of these abnormalities. Microscopic
hematuria was defined as more than two RBCs per high-power field.
|
In addition to the information required for the WUH scale, other data collected for each patient included gender, length of hospital stay (ie, the number of days from hospital admission to discharge or death), initial leukocyte count, requirement for ventilatory assistance (either invasive or noninvasive), and mortality. Underlying comorbidities, including HIV status, also were recorded.
Laboratory Analysis
All Legionella antigen testing for both hospitals was performed
in one centralized laboratory. Before March 25, 1996, specimens
underwent a procedure incorporating a direct, sandwich-type,
solid-phase radioimmunoassay using radioactive iodine
labeling.26
Since then, Legionella antigen testing has
been performed by horseradish peroxidase-labeled enzyme-linked
immunosorbent assay. The two methods are equivalent in performance
specifications (J. Smith-Davis, R. Kohler, MD; personal communication;
January 12, 2000). Specimens for blood cultures were obtained
and observed for growth using standard laboratory methods.
Statistical Analysis
All data were analyzed using computer software (Statistical
Analysis System; SAS Institute; Cary, NC). The Student’s t
test was used to compare the continuous baseline characteristics of the
cohorts (ie, age, total leukocyte count, and length of stay)
and the WUH scores obtained at hospital admission and through the
first week of hospitalization. Fisher’s Exact Test or
2 test was used to compare the groups’ binary
clinical characteristics and the proportions of patients that scored in
the highly probable category. All significance testing was two-sided. A
p value of < 0.05 was considered to be statistically significant.
Comparisons between the two groups’ actual laboratory values were not
documented, as only the presence of a laboratory abnormality (and not
the degree of the abnormality) was required. Sensitivities and
specificities were derived for the WUH scale, defining scores in the
highly probable category as positive; scores in the other two
categories were classified as negative. Ninety-five percent
confidence intervals (CIs) were derived for sensitivities,
specificities, and predictive values of the WUH scale. The receiver
operating characteristic (ROC) curve area was measured as an index of
discrimination between pneumococcal and Legionella cohorts.
Because this study only considered pneumonias caused by Legionella and S pneumoniae, relative prevalences between these two pathogens, as determined from US epidemiologic studies of CAP,4 5 6 were used to adjust the calculated predictive values. These studies, the patient populations in which were similar to that in our study in that they did not exclude patients because of disease severity and that they utilized the urine Legionella antigen assay in their hospital admission diagnostic panel, have shown that the relative ratios of Legionella to S pneumoniae as the cause of CAP range from 1:3 to 1:1. We subsequently used prevalence estimates of 25% and 50% for Legionella, when considering only populations consisting of these two etiologies of CAP in calculating the ranges of possible, adjusted predictive values.
|
Results |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
Baseline Characteristics
The baseline characteristics seen in Table 2
were comparable between the two groups. There were more men than women
in both groups, primarily due to the inclusion of patients from the
Veterans Affairs hospital. The only statistical difference between the
groups was the greater proportion of patients who were HIV-positive in
the pneumococcus group than in the Legionella group (19% vs 2.7%,
respectively; p = 0.04). Ventilatory assistance and mortality were
higher in the pneumococcus group, but the differences were not
statistically significant. Due to the small number of end points, no
correlation analysis between WUH score and ventilatory assistance or
mortality could be performed. These three characteristics of the two
groups (ie, HIV positivity, ventilatory assistance, and
mortality) were similar to those seen in other
studies.4
5
6
27
28
29
30
31
32
|
Table 3 summarizes the results applying the WUH scale to the 37 and 31 patients, respectively, with either Legionella or pneumococcal pneumonia. For the Legionella group, the mean hospital admission score of 14.6 was significantly higher than the corresponding score of 7.5 for the S pneumoniae group (p < 0.001). More patients in the Legionella group scored in the highly probable category (29 of 37 patients; 78%) on hospital admission compared to the pneumococcal group (11 of 31 patients; 35%; p < 0.001). These differences persisted through the first week after hospital admission (data not shown).
|
Subgroup Analysis
All patients in the subgroup analysis had complete laboratory data
available, including those for the major extrapulmonary criteria to
support the diagnosis of legionellosis (ie, serum sodium,
phosphorus, aminotransferase, bilirubin, and creatinine levels and
urinalysis results) by the WUH criteria. The most common reason for
failure to meet the full criteria was the lack of data on serum
phosphorus levels. This occurred primarily after June 1998 when
phosphorus level was no longer part of the standardized comprehensive
metabolic panel. Twenty-three patients in the Legionella group and 18
patients in the pneumococcus group remained for analysis. The clinical
characteristics of this subgroup were no different than those of the
overall study group. Lethargy (48% vs 0%, respectively; p < 0.001)
and pleuritic chest pain (61% vs 17%, respectively; p < 0.01) were
significantly more common in the Legionella and pneumococcus
groups, respectively. The trends observed in the main analysis
for extrapulmonary manifestations and in electrolytic, renal, and
hepatobiliary laboratory tests were present in the subgroup analysis.
No differences in laboratory criteria were statistically significant,
except for the presence of microscopic hematuria seen in the Legionella
group compared with that in the pneumococcus group (61% vs 22%,
respectively; p = 0.03).
Table 4 shows the WUH scores at hospital admission for those patients with complete laboratory data. As before, there was a significantly greater proportion of patients in the Legionella subgroup (20 of 23 patients; 87%) than in the pneumococcus group (9 of 18 patients; 50%) who scored in the highly probable category on hospital admission (p = 0.02). The three patients in the Legionella subgroup who did not qualify for the highly probable category all had elevated transaminase levels, but none fulfilled any of the criteria on the 5-point list of criteria (ie, abdominal pain with or without diarrhea, acute renal failure, relative bradycardia, and no response to ß-lactam therapy). These three patients had WUH scores at hospital admission of 3, 3, and 9. The other 20 patients had at least one of the criteria on the 5-point list. Hospital admission sensitivity in this subgroup analysis was 87% (95% CI, 65 to 97%), and specificity was 50% (95% CI, 27 to 73%). The positive predictive value of 69% (95% CI, 49 to 84%) decreased to 37% (95% CI, 20 to 59%) when adjusted for a 25% prevalence of Legionella, and the negative predictive value subsequently increased from 75% (95% CI, 43 to 93%) to 92% (95% CI, 62 to 98%). Using a 50% prevalence for Legionella, the positive and negative predictive values became 64% (95% CI, 44 to 81%) and 79% (95% CI, 46 to 94%), respectively. The ROC curve area of 0.68 ± 0.09 in this subgroup is similar in magnitude to that found in the main analysis.
|
|
Discussion |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
Olaechea et al35 designed a clinical prediction rule to distinguish atypical from typical etiologies of CAP, but this rule was derived from and designed only for patients requiring intensive care. These patients should automatically receive treatment for Legionella according to current guidelines. Keller and colleagues36 have derived a clinical predictive model for distinguishing patients with Legionnaires disease as the cause of CAP requiring hospitalization; however, neither the model nor confirmatory studies have yet been published. As far as we know, the WUH scale is the only published set of clinical criteria for identifying legionellosis requiring hospitalization that is not limited to intensive care. The criteria seen in Table 1 are primarily based on the initial observational studies of Legionella pneumonia from the late 1970s16 17 18 30 37 38 and permit the WUH scale to be fairly comprehensive. All of these criteria were examined in this study. Some clinical findings were significantly more common in the Legionella group, whereas others were equally prevalent in both groups.
The comparisons of WUH scores should not be affected by confounding baseline characteristics, as the two groups were similar in age, gender, and most comorbid conditions. The only significant difference between the two groups was the larger proportion of HIV-infected patients in the pneumococcus group. However, this should not affect the clinical criteria scores in the pneumococcus group as the presence of HIV infection does not typically affect the presentation or severity of bacteremic pneumococcal pneumonia.39
We tested the ability of these clinical criteria to discriminate between Legionella and pneumococcal pneumonia. The testing modalities of urinary antigen testing for Legionella and bacteremia with pneumococcus were chosen because of their clinical practicality, specificity, reproducibility, and established use in definitively identifying the etiology for CAP on hospital admission.6 28 29 34 Requiring positive sputum cultures or serologic tests for the diagnosis of Legionella pneumonia is impractical and insensitive. The utility of positive sputum cultures for pneumococcal pneumonia is debated due to questionable specificity. Since all patients in the pneumococcus group had negative results of Legionella antigen tests, legionellosis was considered in the initial differential diagnosis, making the comparison between the two groups using the WUH scale more valid clinically.
The WUH system was applied in this study at the time of hospital admission. Cunha15 also uses this scale with data gathered through the first 7 days of hospitalization, but this would not be practical as a screening modality at the onset of hospitalization. We examined the records of the study patients gathered through the first week (data not shown), but this information did not appreciably affect the comparison between the two groups.
Because prior experience has shown that it is difficult to distinguish Legionnaires disease from other pneumonia syndromes,7 21 22 23 24 25 we hypothesized that the WUH system would not prove useful as a screening or discriminatory test for legionellosis at hospital admission for pneumonia. Contrary to our hypothesis, the sensitivities on hospital admission for the WUH system were higher than anticipated at 78% and 87%, respectively, for the main analysis and subgroup analysis. These are comparable to that of the urinary antigen test for L pneumophila serogroup 1. However, the specificities on hospital admission were just 65% and 50%, respectively, for the main cohort and the subgroup. The positive predictive values on hospital admission were low, ranging from 37 to 64% for the subgroup population, whereas the negative predictive values ranged from 79 to 92%. The predictive values changed in the expected directions for a range of relative prevalences between Legionella and S pneumoniae bacteremia. However, the predictive values of the WUH criteria still do not reliably and accurately discriminate between the two pathogens, even for a relative prevalence of 50%. Although the sensitivity of the WUH system was higher than initially suspected, the false-negative rates would be 22% and 13% for the main analysis and subgroup analysis, respectively. Considering the potential lethality of Legionella pneumonia, the inability to identify and subsequently treat 13 to 22% of patients with this disease is clinically unacceptable. Therefore, the sensitivity of the WUH system is not high enough to allow the withholding of antibiotics for legionellosis empirically when the prediction rule indicates that Legionella is not highly probable.
The ROC curve areas were only fair to moderate for discrimination between the two groups. Thus, the WUH scale cannot be used clinically to distinguish Legionella from bacteremic pneumococcal pneumonia. It should be remembered that these performance characteristics were based on the inclusion of only two causes of CAP (albeit two of the more morbid and fatal causes) and that the Legionella patients in the study were restricted to those infected with L pneumophila serogroup 1. When applying the WUH criteria to all patients with CAP regardless of the etiologic pathogen, the sensitivity and negative predictive value may decrease further while the specificity and positive predictive value may improve. Therefore, a more refined set of clinical criteria, based only on the more discriminatory parameters found in this study along with other relevant clinical variables, may prove useful. If a rule can be developed with sensitivity higher than the WUH scale, then anti-Legionella therapy could be safely withheld in those patients with low scores. The number of cases with low scores would likely increase if consecutive cases of CAP were then prospectively studied.
A strength of the study is the strict inclusion criteria, as there were no confounding criteria between the two analyzed etiologies of pneumonia, Legionella and S pneumoniae. The study, though, has several limitations. One limitation is the sampling strategy for the study groups. It would have been preferable to test the WUH system on a cohort of consecutive patients with CAP who required hospitalization, all of whom would have undergone extensive diagnostic testing for a specific etiology. However, this methodology was neither feasible nor practical. Thus, we performed a retrospective case-control study and adjusted the test characteristics for the literature-based relative prevalence of Legionella and pneumococcus. However, as the study group is not a consecutive cohort, both cases (patients) and control subjects may be biased with respect to the spectrum of disease in that those patients selected for specific microbiological diagnostic testing may have had more classic manifestations. This type of bias might overestimate the performance of the WUH system.
A second important limitation is the small sample size. Although all patients at two large academic hospitals who received diagnoses of pneumonia were screened over a period of 6 years, a total of only 68 patients (Legionella group, 37 patients; pneumococcus group, 31 patients) met the inclusion criteria. This is explained by the fact that most patients with pneumonia did not have Legionella testing performed, although it was readily available. Also, because of the usually rapid positive culture results for pneumococcal bacteremia, it was not clinically imperative to test for Legionella. A large number of patients never had Legionella tests performed, even in the face of negative blood test and sputum culture results. This most likely was due to a lack of clinical suspicion for legionellosis. Another possibility is that the initial antimicrobial therapy may have included quinolones or macrolides. These agents would treat Legionella empirically and would reduce the need for a precise microbiological diagnosis. Conversely, urine antigen testing was probably pursued more often in those patients who had clinical features that were more classically consistent with Legionella. If this were the case, then the resulting incorporation bias would tend to make the WUH criteria appear better than they truly were. Also, a significant proportion of the overall group of patients did not have full laboratory data available, primarily due to the lack of data on serum phosphorus levels after June 1998. This further limited the number of patients for whom complete data were available. One other possible limitation of this study was the WUH system itself. Any laboratory data outside the range of normal could fulfill diagnostic criteria; the degree of the abnormality was not important.
In summary, the WUH criteria are not adequately sensitive to warrant general use as a type of prediction rule for diagnosing legionellosis on hospital admission. Therefore, it should be assumed that Legionella might be the cause of CAP, and empiric anti-Legionella therapy should be included in the initial treatment regimen for all patients with CAP requiring hospitalization. If a more sensitive rule could be derived, however, it might be possible to withhold treatment for Legionella in patients with low scores. More specific antibiotic therapy then would help prevent antibiotic resistance. Such a prediction rule would require prospective validation on consecutive patients with CAP to determine its utility for guiding decisions about initial antibiotic therapy in this clinical setting.
|
Addendum |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
|
Acknowledgements |
|---|
|
Footnotes |
|---|
This research was supported in part by grant No. DK02756–01 (T.F.I.).
Received for publication October 23, 2000. Accepted for publication May 8, 2001.
|
References |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionAddendumReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. Cramer Legionnaires Disease: A Case Study Am. J. Crit. Care., May 1, 2003; 12(3): 234 - 238. [Full Text] [PDF]
|
|
|
|
 |
|
 J. Roig and J. Rello Legionnaires' disease: a rational approach to therapy J. Antimicrob. Chemother., May 1, 2003; 51(5): 1119 - 1129. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
