INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Bacterial pneumonia is a common cause of morbidity in human immunodeficiency virus (HIV)-infected patients, being an indicator of immunosuppression (1). For this reason, recurrent bacterial pneumonia was included as a defining condition for acquired immunodeficiency syndrome (AIDS) (5). Bacterial pneumonia also constitutes a major cause of death in patients with HIV infection (6), with a mortality rate, in-hospital or after 4 wk, that ranges from 2.6% to 27% (1, 7).

In the past decade, several investigators have evaluated prognostic criteria and mortality-associated factors in community acquired pneumonia (CAP) in the general population (12- 17). These investigations led to a consensus on severity criteria for CAP in immunocompetent patients (18). However, in patients with HIV infection and bacterial CAP, the factors that predict mortality have not been thoroughly studied, and it is not known whether the severity criteria used for immunocompetent patients are valid for those who are HIV infected. The knowledge of these prognostic factors would help clinicians to identify patients with a higher probability of death, and to define a strategy of treatment, including antimicrobial and supportive therapy and the need for hospitalization.

The objective of the present study was to evaluate the prognosis and mortality-associated factors in bacterial CAP in HIV-infected patients, and to validate the severity criteria for CAP in immunocompetent patients in the HIV-infected population.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Population

From February 1, 1996 to January 31, 1997 we conducted a prospective, observational, hospital-based study of consecutive cases of bacterial CAP in 12 hospitals in Andalucia (Southern Spain), serving a population of 3,342,329 inhabitants. A total of 355 patients were enrolled in the study. All met the following inclusion criteria: age over 18 yr, HIV infection (by enzyme-linked immunosorbent assay and Western blotting), and hospital admission because of bacterial CAP.

A uniform protocol was designed and followed by all of the participating hospitals, and all recorded data were validated by two of the authors (E.C. and J.P.). The following data were collected from every patient: (1) demographic data consisting of age, sex, type of housing, drug-use habits; (2) HIV infection-related data including risk factors for HIV infection, stage of HIV infection (5), CD4⁺ lymphocyte count, concomitant opportunistic diseases, antiretroviral therapy (ART), Pneumocystis carinii prophylaxis; (3) symptoms, signs, and analytical findings at admission, including fever, cough, expectoration, chest pain, dyspnea, hypotension (22), septic shock (22), septic extrapulmonary complications, leukocyte count, arterial oxygen tension (Pa_O2), and ratio of Pa_O2 to fraction of inspired oxygen (Pa_O2/ FI_O2); (4) radiologic findings (radiologic pattern, number of lobes involved, cavities, pleural effusion, and pneumothorax) at admission and 48 h thereafter; and (5) initial antimicrobial therapy and course. Treatment was considered appropriate when antimicrobial agents were active against the microorganism isolated (23).

Microbiologic Studies

The collection and processing of microbiologic specimens were done in the microbiology laboratory of each participating hospital according to standard criteria. The specimens were processed for bacteria, mycobacteria, parasites, and fungi. Blood cultures were made in cases with fever. Fiberoptic bronchoscopy and pulmonary needle aspiration were ordered by the clinician in charge of the patient in cases without a diagnosis by noninvasive techniques and/or cases failing to show improvement with empirical treatment. Serologic testing for Legionella pneumophila, Coxiella burnetii, and Mycoplasma pneumoniae was done initially in all patients and again after 21 d in 163 (45.9%) cases.

Diagnosis of Bacterial CAP

Patients were considered as having a definite bacterial pneumonia if they met the following three criteria: (1) presence of fever (temperature > 38° C), dyspnea, cough and/or expectoration; (2) a new pulmonary infiltrate in a chest radiograph, for which noninfectious causes were excluded; (3) an etiologic diagnosis based on at least one of the following samples: blood, pleural fluid, sputum with culture, bronchial aspirate ( 10⁶ cfu/ml), bronchoalveolar lavage fluid ( 10⁴ cfu/ml), protected brush specimen ( 10³ cfu/ml), transbronchial biopsy (24), or pulmonary needle aspirate. Sputum samples were considered acceptable if they contained > 25 polymorphonuclear leukocytes and < 10 epithelial cells per high-power field (25). Presumed bacterial pneumonia was defined by the presence of the first two criteria and by cure with antibacterial treatment (other than with cotrimoxazole) or, in those patients who died, by the prescription only of antibacterial treatment by the clinicians in charge of the patient.

A bacterial species was considered to have caused pneumonia when it was isolated as a single or multiple pathogen in sterile fluids, or when it was the only organism isolated in respiratory samples.

Pneumonia was considered to have been community-acquired when the symptoms occurred in a patient who had not been hospitalized in the previous 7 d. Patients with right-sided endocarditis and septic pulmonary emboli were excluded. Patients with pulmonary mycobacteriosis were also excluded.

Severity Criteria

Severe pneumonia was defined by the presence of at least one of the following modified American Thoracic Society (ATS) criteria (18): (1) Pa_O2 < 60 mm Hg, Pa_O2/FI_O2 < 250 mm Hg, or requirement for mechanical ventilation; (2) a chest radiograph showing involvement of multiple lobes, cavities, or pleural effusion extending beyond occupation of the lateral costophrenic angle; (3) an increase in the size of a chest opacity by 50% or more within 48 h following admission; (4) septic shock (22); (5) septic extrapulmonary complications; and (6) acute alteration in mental status.

Statistical Analysis

Descriptive data for contiguous variables are presented as means ± SD or medians with ranges, when appropriate. Statistical comparisons of categorical variables were made by chi-square analysis or Fisher's exact test, and statistical comparisons of contiguous variables were made by Student's t test or analysis of variance. Homogeneity was assessed with Bartlett's test. Nonparametric tests were used when the conditions for parametric tests were not fulfilled. Statistical significance was defined as a two-tailed value of p < 0.05 (26).

Attributable mortality was calculated as the proportion of patients who died during hospitalization and in whom bacterial CAP was judged to be the cause of death.

To determine independent predictors of mortality from bacterial CAP, we used a modeling procedure with a stepwise multiple logistic regression, to adjust for possible confounding effects. The factors included were those related to death in the bivariate analysis and those that, despite having a value of p  0.05, were considered clinically relevant. For those variables with more than two categories, a cutoff point was selected according to the results of the bivariate analysis.

A multiple logistic regression model was used to determine factors present at the patient's admission that were independently associated with radiologic progression of disease within 48 h of treatment. A prognostic rule was then constructed, using the variables independently related to death that were available at admission plus the independent baseline predictors of radiologic progression (26).

The following operative indices were determined in order to compare ATS criteria and the new prognostic rule: sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The posttest NPV and PPV were obtained with Bayes' theorem (26).

Statistical analyses were done with software from the Statistical Package for the Social Sciences (version 8; SPSS, Chicago, IL) and Epi Info (version 6.0; CDC, Atlanta, GA).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

During the 1-yr study, 3,269 HIV-infected patients were admitted to the participating hospitals, 759 (23.3%) because of CAP. Among these cases, 398 (12.2%) were nonbacterial CAP, 355 (10.8%) were bacterial CAP, and six were unclassified. Among the cases of bacterial CAP, 190 were considered definite and 165 presumed. The main characteristics of patients with bacterial CAP are shown in Table 1.

Streptococcus pneumoniae was the most common agent of CAP (74 cases; 38.9%), followed by Pseudomonas aeruginosa (34 cases; 17.9%) and Haemophilus influenzae (26 cases; 13.7%) (Table 2). Blood cultures were done in 329 patients and were positive in 75 (22.8%). The proportions of bacteremic pneumonia according to the microbiologic diagnosis are stated in Table 2. Among the 190 patients with definite CAP, 152 patients (80%) initially received appropriate therapy and 38 patients (20%) received inappropriate therapy.

The mortality attributable to bacterial CAP was 9.3% (n = 33). By ATS criteria, 214 patients had severe bacterial CAP and 141 had nonsevere bacterial CAP. The patients with severe bacterial CAP had a higher mortality (13.1% versus 3.5%; odds ratio [OR] = 4.09; 95% confidence interval [CI]: 1.5 to 13.89; p = 0.02) and a longer hospital stay (15.3 ± 12.3 d [mean ± SD] versus 12.1 ± 11.7 d; p = 0.01), and took longer for desfervescence (5 ± 5.6 versus 2.7 ± 2.4 d; p < 0.001). No difference was found in terms of etiology between the severe and nonsevere CAP groups.

Prognostic Factors

The bivariate analysis identified the following factors associated with mortality: C stage of HIV infection, CD4⁺ cell count < 100/µl, hypotension, septic shock, dyspnea, Pa_O2 < 60 mm Hg, etiologic diagnosis, bacteremia, and radiologic progression after 48 h of treatment; the OR, 95% CI, and level of significance of these factors are summarized in Table 3. Etiology, comorbid conditions, concomitant opportunistic diseases, and inappropriate initial antimicrobial therapy unrelated to mortality.

Patients with septic shock had the highest attributable mortality (63.1%). The bacterial etiology in these cases was P. aeruginosa (n = 5), S. pneumoniae (n = 4), Staphylococcus aureus (n = 2), Escherichia coli (n = 1), and unknown (n = 7). Of the patients with septic shock and inappropriate empirical antimicrobial therapy, 85.7% died, as compared with 25% of those with appropriate antimicrobial therapy (p = 0.09). Death occurred earlier in patients with septic shock (1.5 d, range: 1 to 14 d; versus 13 d, range: 1 to 21 d; p < 0.01).

When the variables selected in the bivariate analysis were introduced in a stepwise logistic regression analysis, it was observed that only septic shock (OR = 28.6; 95% CI: 8.31 to 98.8), radiologic progression (OR = 10.9; 95% CI: 4.3 to 27.6), and CD4⁺ cell count < 100/µl (OR = 4.8; 95% CI: 1.93 to 12.32) were independent indicators of mortality (Table 3).

The attributable mortality for patients who met at least one of these three criteria was 13.7%, whereas in patients who did not meet any of these criteria it was 0% (p < 0.001). Because the degree of radiologic progression is not known at baseline, we conducted an analysis to determine those variables at admission that could predict radiologic progression (Table 4). When we did this, the multivariate analysis selected three independent variables: multilobar infiltrates, cavities, and pleural effusion.

We used the variables selected in both multivariate analyses (septic shock, CD4⁺ cell count < 100/µl, multilobar infiltrate, cavities, and pleural effusion) to construct a prognostic rule. Patients who met one or more of these criteria (n = 282) had a mortality of 11.3%, whereas the group of patients who met with none of them (n = 73) had a lower risk of mortality (1.3%) (OR = 0.11; 95% CI: 0.00 to 0.68; p = 0.008). Attributable mortality increased proportionally to the number of severity criteria, from 7.5% in patients who met with one criterion, to 8.7% in those who met two, 34.5% in those who met three, and 42.8% in those who met four criteria (p < 0.001). No patient met five severity criteria. Using this prognostic rule, an NPV for mortality of 97.3% was found, resembling the 96.4% obtained with the ATS severity criteria. The operative characteristics of both predictive rules are detailed in Table 5. Among the five criteria, the only one with a high PPV (63.1%) was septic shock (Table 6).

The length of stay, excluding patients who died in the first 24 h, was 15.2 ± 11.9 d among patients with severe CAP defined by the proposed prognostic rule, versus 12 ± 11.9 d for patients with nonsevere CAP (p = 0.04).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In the present study we found that the ATS severity criteria for CAP (18) which are generally accepted for immunocompetent patients, are also valid for the HIV-population with bacterial CAP. Moreover, we established a prognostic rule with two main strengths: its capacity to identify patients with low risk of death, and availability of the prognostic criteria at the first patient visit. Both conditions would permit use of the rule to identify patients who may be safely treated on an outpatient basis.

The incidence of bacterial pneumonia in HIV-infected patients ranges from 1.93 per 100 to 19.2 per 100 patients/year (1, 3, 7, 27, 28). With the introduction in 1996 of highly active ART (HAART), a reduction in the number of opportunistic diseases, in HIV-infected patients, including recurrent bacterial pneumonia, has occurred (28, 29). Nevertheless, bacterial pneumonia continues to be an important problem in HIV- infected patients (28), being the pathology with the smallest decrease in incidence of admissions and the chief cause of hospitalization in HIV-infected patients in the HAART era (29).

Mortality resulting from CAP in the general population ranges from less than 1% to 50% (12, 30, 31), depending on the clinical severity of the pneumonia. Different studies have identified prognostic factors associated with mortality from CAP (12, 14). Based on these studies, guidelines defining criteria for hospitalization of patients with CAP have been created (13, 18, 19). However, HIV-infected patients have always been excluded from studies that have analyzed prognostic factors in CAP, and from all recommendations of the different expert panels regarding hospitalization for this condition. Therefore, HIV-infected patients with bacterial CAP, like other immunosuppressed patients, are frequently hospitalized, since there are no guidelines addressing this matter.

To our knowledge, only two studies assessing mortality-associated factors in bacterial pneumonia in HIV-infected patients have been previously reported. Falco and colleagues (10) conducted a study of 68 episodes of bacterial pneumonia in HIV-infected patients at a single institution. They found that pleural effusion and a lymphocyte count < 1,000/µl were the only variables independently associated with mortality. Tumbarello and coworkers (11) conducted a retrospective study, also at a single center, including both community- and nosocomially-acquired bacterial pneumonia. They identified four factors related to death: hypoxemia, CD4⁺ cell count < 100/µl, neutropenia < 1,000/µl, and Karnofsky score < 50. The present study was based on a large multicenter population, the data were recovered prospectively, the study included a large number of episodes of bacterial CAP, and the study was restricted to CAP.

In this study, the factor with the strongest relation to mortality was septic shock, such as occurs in immunocompetent patients (12). Early, appropriate antimicrobial therapy is an important factor in the outcome of septic shock in the general population (32, 33). Although the difference was not significant, we found a trend toward higher mortality among those patients with septic shock who recieved inappropriate antimicrobial therapy. This fact, and the chronology of death in patients with septic shock, make it advisable to promptly begin antimicrobial therapy that is active against the most frequently observed microorganisms in this situation (P. aeruginosa, S. pneumoniae, and S. aureus). The lack of association between initial inappropriate therapy and prognosis in the overall patient population of the study may have been due to the later occurrence of death in patients without septic shock, making it possible to change their antimicrobial treatment after an etiologic diagnosis has been made.

With respect to other factors analyzed, no specific microorganism was related to death. However, a trend toward higher mortality and more frequent presentation of severe pneumonia in cases of S. aureus and P. aeruginosa infection must be noted. A negative relationship between CD4⁺ cell count and the incidence of bacterial pneumonia has previously been found (1). We have observed in the present study that severe cellular immunodeficiency is also a factor for poor prognosis, and is strongly related to death.

Our results suggest a good relationship between the clinical severity of pneumonia defined by the ATS criteria (18) and outcome. These criteria are associated with mortality in bacterial CAP in HIV-infected patients. They also are related to other clinically relevant medical outcomes, such as time to defervescence or length of hospital stay.

However, because the ATS criteria include an event occurring in the course of CAP (radiologic progression), they cannot be used as a basis for deciding to hospitalize a patient. To solve this problem, we constructed a new prognostic rule based on five variables: septic shock, CD4⁺ cell count < 100/µl, pleural effusion, cavities, and multilobar infiltrates. The first two criteria are independent mortality-related factors, and the last three criteria are predictors of radiologic progression.

The prognostic rule we propose offers the possibility of use at the first patient evaluation. It has a very high NPV, we therefore believe that it may be used to determine the need for hospitalization. The absence of these criteria selects a group of patients with a low risk of death, who may be treated on an outpatient basis, providing there is close subsequent clinical follow-up. Also, this prognostic rule may lead to a reduction in the use of hospital resources. Considering that the median stay of patients with CAP in our study was 11 d, our rule would result in a reduction of 803 beds/d at the participating centers over a 1-yr period. Both the ATS rule and our own prognostic rule identify patients with a low risk of dying. However, two rules' low PPV precludes their use in selecting a group of patients with a high risk of mortality.

Our data were obtained from a patient population with intravenous drug use as the most frequent risk factor for HIV infection, reflecting the characteristics of epidemic HIV in Spain. The possibility of generalizing our results to geographic areas with different predominant HIV risk factors should be evaluated.

Although this study provides evidence about prognostic factors in bacterial CAP in HIV-infected patients, some limitations may be noted. First, although the spectra of etiologies and proportions of microbiologic diagnoses were similar to those in previous studies (2, 7, 27, 34, 35), serologic results for Chlamydia pneumoniae, an important pathogen in one study of bacterial pneumonia in HIV-infected patients (7), could not be determined in our study. However, because C. pneumoniae usually causes a nonsevere pneumonia, we think that it may not modify the conclusions of our study. Also, the previous use of antimicrobial treatment for our patients was unknown, which might have reduced the etiologic yield in our study. Second, since the study was conducted before the standardization of HAART in clinical practice, we could not determine the influence of this variable on the mortality from bacterial CAP. Additionally, as in other studies in immunocompetent patients (13), it must be stressed that the proposed rule should be regarded as a guideline to be used for an individual patient concomitantly with the judgment of the clinician.

In summary, our study provides a prognostic rule that identifies HIV-infected patients with bacterial CAP who may be treated on an outpatient basis. The introduction of the rule in clinical practice would result in better management of a still prevalent disease in the HIV-infected population. However, the safety and effectiveness of our proposed prognostic rule should be validated in independent cohorts.