HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:
Guest Access | Sign In via User Name/Password

This Article Abstract Full Text (PDF) Free Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Article Archive Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (9) Citing Articles via Google Scholar Google Scholar Articles by Arozullah, A. M. Articles by Yarnold, P. R. Search for Related Content PubMed PubMed Citation Articles by Arozullah, A. M. Articles by Yarnold, P. R.

A Rapid Staging System for Predicting Mortality From HIV-Associated Community-Acquired Pneumonia^*

^* From the VA Chicago Healthcare System (Dr. Arozullah), Westside Division, Department of Medicine, University of Illinois College of Medicine, Chicago; Hines VA Hospital (Dr. Parada), Hines, and Department of Medicine, Loyola University Medical School, Maywood; VA Chicago Healthcare System (Dr. Bennett), Lakeside Division, Chicago; Department of Medicine (Ms. Phan), Northwestern University Medical School, Chicago; Department of Preventive Medicine (Drs. Deloria-Knoll and Chmiel), Northwestern University Medical School, Chicago; and Department of Psychology (Dr. Yarnold), University of Illinois at Chicago, Chicago, IL.

Correspondence to: Ahsan M. Arozullah, MD, MPH, VA Chicago Healthcare System, Westside Division (151WS), 820 S Damen Ave, Chicago, IL 60612; e-mail: arozulla{at}uic.edu

	Abstract

TOP Abstract Introduction Materials and Methods Results Discussion References

 
Study objective: Community-acquired pneumonia (CAP) accounts for an increasing proportion of the pulmonary infections in individuals with HIV infection. During the mid-1990s, hospital mortality rates for HIV-associated CAP ranged from 0 to 28%. While hospital differences in case mix may account for mortality rate variation, few methods to evaluate illness severity for HIV-associated CAP have been reported previously. The study objective was to develop a staging system for categorizing mortality risk of patients with HIV-associated CAP using information available prior to hospital admission.

Design/setting/patients: Retrospective medical records review of 1,415 patients hospitalized with HIV-associated CAP from 1995 to 1997 at 86 hospitals in seven metropolitan areas.

Measurements: In-patient mortality rate.

Results: Hierarchically optimal classification tree analysis was used to develop a preadmission staging system for predicting inpatient mortality. The overall inpatient mortality rate was 9.1%. The significant predictors of mortality included the presence of neurologic symptoms, respiratory rate 25 breaths/min, and creatinine > 1.2 mg/dL. The model identified a five-category staging system, with the mortality rate increasing by stage: 2.3% for stage 1, 5.8% for stage 2, 12.9% for stage 3, 22.0% for stage 4, and 40.5% for stage 5. The classification accuracy of the model was 85.2%.

Conclusions: Our staging system categorizes inpatient mortality risk for patients with HIV-associated CAP using three routinely available variables. The staging system may be useful for guiding clinical decisions about the intensity of patient care and for case-mix adjustment in future studies addressing variation in hospital mortality rates.

Key Words: community-acquired pneumonia • HIV • hospital mortality

	Introduction

TOP Abstract Introduction Materials and Methods Results Discussion References

 
Patients with HIV infection are at increased risk for pulmonary infections including Pneumocystis carinii pneumonia (PCP), tuberculosis, and bacterial pneumonia.^{1 2} For the past several years, the incidence of bacterial pneumonia exceeded the incidence of PCP among patients with HIV infection.³ Since the introduction of protease inhibitors, there has been a 75% decline in the incidence of AIDS-defining opportunistic infections. While community-acquired pneumonia (CAP) has been responsible for a majority of the pulmonary-related morbidity and mortality in patients with HIV,^{4 5 6} the majority of HIV-associated pneumonia quality-of-care studies have focused on PCP.^{7 8 9 10}

Any assessment that evaluates hospital performance for HIV-associated pneumonia must take into account differences in patient factors associated with mortality. The prior quality-of-care studies that evaluated HIV-associated PCP or even those that evaluated non–HIV-associated CAP have incorporated disease-specific and condition-specific staging systems. Fine and colleagues^{11 12} developed a pneumonia severity index for non–HIV-associated CAP that was based on 19 clinical, laboratory, and radiographic variables. We previously developed similar disease-specific staging systems for HIV-associated PCP that was diagnosed in the first and second decades of the AIDS epidemic.^{8 9 10} These PCP staging systems were based on clinical factors such as age, history of AIDS, and weight loss, as well as laboratory factors such as alveolar-arterial oxygen gradient and serum hemoglobin and albumin levels. The purpose of this study was to develop and validate a disease-specific and condition-specific severity-of-illness staging system for HIV-associated CAP. We also compared our staging system with previous models developed for non–HIV-associated CAP and HIV-associated PCP.

	Materials and Methods

TOP Abstract Introduction Materials and Methods Results Discussion References

 
Sampling of Patients and Hospitals
Medical records were abstracted for patients with HIV infection with confirmed or probable CAP who received medical care at a study institution between January 1, 1995, and December 31, 1997. Study institutions included 86 public and private hospitals in seven metropolitan areas, including New York City, Los Angeles, Miami, Chicago, Seattle, Nashville/Memphis, and Phoenix/Tucson. During 1996–1997, New York City had the largest number of reported AIDS patients in the United States, with Los Angeles having the second, Miami having the fifth, and Chicago having the eighth.¹³ The institutional review board of each participating hospital approved the study protocol.

We used a sampling methodology, similar to previous studies, that employed two hierarchical levels: hospitals within metropolitan areas and patients within hospitals.^{9 10 14} The number of randomly selected patient charts reviewed at each hospital relative to the charts reviewed in each metropolitan area was roughly proportional to the square root of the individual hospital caseload divided by the metropolitan area caseload. The number of charts reviewed within each hospital was stratified by patient discharge status to approximate the actual hospital mortality rate.

Registered nurses experienced in caring for patients with HIV and utilization review methods were trained to perform retrospective chart reviews of patients identified by medical information system analysts at study hospitals. All discharges including International Classification of Diseases, ninth revision (ICD-9) codes for bacterial pneumonia (481–486) and HIV-related disease (042–044) were screened. Patients were excluded if they had any one of the following criteria: age < 18 years, transfer from another acute care hospital inpatient service, cytologically proven PCP in the previous 30 days, culture-proven pulmonary tuberculosis or CAP within the past 30 days, or hospitalization for any other reason within the past 30 days. Patients with the following cancers were also excluded: adenocarcinoma with unknown primary, bladder, brain, colon, esophagus, gastric or stomach, larynx, leukemia, liver, lung, oat cell, or small cell, oropharynx, ovary, pancreas, sarcoma (other than Kaposi). Patients with a history of these cancers were excluded in order to minimize the misclassification of abnormal findings on chest radiography.

Data abstracted from medical charts included the following: patient sociodemographics; insurance status; HIV-related and non–HIV-related coexisting illnesses; cigarette, alcohol, and drug use history; preadmission use of antiretroviral and prophylactic medications; T-cell count; initial vital signs, arterial blood gas, and laboratory data; treatment medications received; principal and secondary diagnostic and procedure codes; length of stay; discharge status. Neurologic symptoms were considered present if a physician noted that the patient was either comatose, confused, and/or had symptoms of neurologic change on day 1 or 2 of hospital admission. These three categories were combined based on chart abstractor observations that the majority of symptoms of neurologic change noted were mental status changes. Two physicians trained in quality assurance for the study maintained data quality by reviewing completed medical chart abstraction forms. The physician overreaders categorized < 1% of entries as possibly inaccurate.

Statistical Analysis
Univariate associations between measured attributes and in-hospital mortality status were evaluated using univariate optimal discriminant analysis. A multivariate nonlinear model was obtained via hierarchically optimal classification tree analysis (CTA) using an algorithm that maximized mean sensitivity.^{10 15 16} Mean sensitivity is the average of the sensitivity for patients classified as dead and for those classified as alive. Parametric analyses were inappropriate due to violations of assumed distributional features, and traditional nonparametric analyses were inappropriate due to the many tied data values. Jackknife validity analysis was conducted to assess the potential generalizability of the findings. For characteristics with stable effect strength in jackknife validity analysis, the optimal cut points identified would be expected to cross-generalize at the jackknife estimate if they were used to classify independent random patient samples. In contrast, if statistically significant characteristics were unstable in jackknife validity analysis, these characteristics would be expected to be predictive of mortality, but using model cut points different than those identified. For the CTA model, a sequentially rejective Sidak Bonferroni-type multiple comparisons procedure was used to ensure an experiment-wise type I error rate of p 0.05.¹⁵

	Results

TOP Abstract Introduction Materials and Methods Results Discussion References

 
There were 1,415 patients with HIV-associated CAP who met our eligibility criteria with an in-hospital mortality rate of 9.1%. Hospital-specific mortality rates for hospitals with > 10 eligible patients varied from 0 to 28%. Patient demographic and clinical characteristics are given in Tables 1 , 2 . The mean age was 39.8 years, with 78% of the patients being male and 47% being African American. Over half of the patients had a prior AIDS-related diagnosis; however, only 43% of patients reported receiving any antiretroviral therapy prior to hospital admission, and only 15% reported receiving protease inhibitors. Nearly one third (32%) of the patients had a history of non-PCP pneumonia, while 25% had a history of PCP. Nearly 21% of the patients had neurologic symptoms, including 1.6% who were comatose and 8.0% who were confused.

CTA Model
A nonlinear multivariate model for predicting inpatient mortality was created using hierarchically optimal CTA (Fig 1 ). The presence of neurologic symptoms was selected as the initial (root) attribute in the CTA model because it had the highest effect strength for predicting mortality that was stable in jackknife validity analysis. The other significant variables used in the CTA model were respiratory rate and creatinine level. While creatinine level was unstable in jackknife validity analysis for the entire sample, it provided stable mortality estimates when used to classify patients in the subset without neurologic symptoms and 25 breaths/min.

View larger version (24K): [in this window] [in a new window] [Download PPT slide]   Figure 1.. Schematic illustration of the CTA model for predicting mortality. Circles represent nodes, (decision points), arrows indicate branches (decision paths), and rectangles represent prediction end points. (As indicated, the model classifies patients having a given attribute profile as either dead or alive.) Numbers underneath nodes give the generalized type I error for the node. Numbers/words adjacent to arrows indicate the value of the cut point or category for each node.

Of the 1,415 patients, the CTA model classified 1,403 patients (99.2%) with nonmissing data, of which 1,196 patients (85.2%) were classified correctly (Table 5 ). The CTA model yielded relatively strong effect strength for sensitivity of 45.5, indicating that the model provided 45.5% of the theoretical classification improvement possible to achieve beyond chance. For effect strength, 0 is equal to the classification accuracy expected by chance, and 100 is equal to perfect (errorless) classification accuracy. All model performance indexes and end point predictive values were closely approximated in jackknife and bootstrap validity analysis, indicating model performance stability for a 50% reduced sample size (Table 6 ).

We also compared CTA model performance with pneumonia severity index performance in our sample of patients with HIV-associated CAP.^{11 12} Since the pneumonia severity index was developed for the general population, it is reasonable to expect that a different set of threshold values might be applicable to the present population of patients with HIV.¹⁷ Accordingly, we used optimal discriminant analysis to identify the optimal pneumonia severity index score cut point for accurately classifying patient mortality status in our sample. The cut point value that satisfied this criterion was 80.5, meaning that patients with pneumonia severity index scores 80.5 were predicted to be alive (accurate for 832 of 857 patients; 97.1%), while patients with pneumonia severity index scores > 80.5 were predicted to be dead (accurate for 107 of 548 patients; 19.5%). The pneumonia severity index score had effect strength for sensitivity of 46.4, or 2% greater effect strength than our CTA model. However, in contrast to our CTA model, the pneumonia severity index was not stable in jackknife validity analysis as its effect strength for sensitivity fell to 40.0, or 88% of the CTA model. The pneumonia severity index also requires the use of 19 variables, while our CTA model provides similar predictive ability and stable estimates while using only three variables.

	Discussion

TOP Abstract Introduction Materials and Methods Results Discussion References

 
Opportunistic infections, such as PCP, have been responsible for a majority of the pulmonary-related morbidity and mortality among patients with HIV infection.^{1 2} However, bacterial pneumonia was recognized as a common cause for hospitalization among HIV-infected gay/bisexual men early in the HIV epidemic.¹⁸ In 1995, Hirschtick et al³ reported that among their prospective cohort of 1,130 patients with HIV infection, the incidence of bacterial pneumonia was greater than the incidence of PCP. They estimated that the mortality attributed to bacterial pneumonia was 6%, and that patients with bacterial pneumonia had a mortality rate ratio of 3.9 compared to patients without bacterial pneumonia.³ An objective evaluation of the quality of care for HIV-associated CAP is needed in light of the increasing incidence of CAP among patients with HIV infection. We studied 1,415 patients from seven major geographic areas in the United States who were enrolled in the Multi-City Study of Quality of Care for HIV-related CAP. The overall in-hospital mortality rate for these patients was 9.1%. Using information easily available at hospital admission, we created a five-stage system using three variables for predicting in-hospital mortality in patients with HIV-associated CAP.

Previous studies of HIV-associated bacterial pneumonia reported several risk factors associated with mortality. One study of 350 episodes of bacterial pneumonia in 285 patients with HIV infection found that CD4⁺ lymphocyte count 100 cells/µL, neutropenia, PO₂ 70 mm Hg, and Karnofsky score 50 were associated with increased mortality.¹⁹ A study of 355 patients with HIV infection with bacterial CAP found that shock, CD4⁺ lymphocyte count 100 cells/µL, and chest radiograph findings of pleural effusion, cavities, or multilobar infiltrates were risk factors for mortality.²⁰ Similarly, in univariate analyses, we found that lower systolic or diastolic BP and CD4⁺ lymphocyte count 50 cells/µL were associated with higher mortality (p < 0.001). In contrast to previous studies, our results indicated that hypoxemia 60 mm Hg or the presence of a pleural effusion was not associated with higher mortality. Similar to prior studies in HIV-associated PCP,¹⁰ there was an increased risk of mortality among patients with prior use of M avium complex prophylaxis (16.0% vs 7.7%, p < 0.001). The increased mortality risk may be related to the underlying immunocompromised state of patients receiving M avium complex prophylaxis or macrolide prophylaxis may be selecting for drug-resistant bacterial organisms leading to increased mortality.

The variables used in our staging system reflect the severity of pulmonary illness (respiratory rate) and two comorbid medical conditions (renal and neurologic status). These factors are similar to those included in our HIV-associated PCP severity-of-illness system, and are routinely recorded in hospital admission notes.¹⁰ The PCP model included alveolar-arterial oxygen gradient that reflects the severity of pulmonary illness; however, we found that the rate of room air arterial blood gas results recorded for patients with HIV-associated CAP was much lower than with HIV-associated PCP, rendering the alveolar-arterial oxygen gradient less useful as a potential predictor of mortality in HIV-associated CAP. With respect to severity of pulmonary illness, the CAP model uses respiratory rate that was readily available from the hospital admission physical examination or nursing notes for all of the study patients.

Severity of comorbid medical illness was assessed by serum albumin in the PCP model and by neurologic symptoms and creatinine level in the CAP model. Comments on neurologic symptoms were included in the majority of hospital admission notes for patients with HIV infection with respiratory complaints. As opposed to albumin level, patients routinely receive testing for creatinine level during hospital admission. The upper limit of the normal range for creatinine level in most laboratories is 1.4 mg/dL. However, among the subset of patients without neurologic symptoms and respiratory rate of 25 breaths/min, we found that those with creatinine > 1.2 mg/dL had a mortality rate of 22% vs 5.8% for those with creatinine 1.2 mg/dL. The mortality rate of 10.3% for this subset of patients was very similar to the overall mortality rate of 9.1%. In prior studies of HIV-associated PCP, we found that wasting was a risk factor for increased mortality.¹⁰ One possible explanation for the lower threshold value for creatinine is that patients with HIV infection have lower muscle mass resulting in lower baseline creatinine levels. Therefore, a creatinine level of 1.2 mg/dL may represent a significant worsening of renal function in patients with lower-than-normal baseline values. Renal disease and BUN are both risk factors for increased mortality among patients with non–HIV-associated CAP as well.^{11 12}

In addition to using routinely measured clinical findings and laboratory values, our staging system is both disease specific and HIV specific. The effect strength for sensitivity for our CTA model was similar to the pneumonia severity index that was developed for non–HIV-associated CAP.^{11 12} However, the effect strength for our CTA model was more stable in validity analysis. We have previously reported similar findings for HIV-associated PCP, in which disease-specific models performed better than general HIV severity-of-illness models.^{9 10} Our staging system uses only 3 variables, compared to 19 variables needed to calculate the pneumonia severity index. Although other variables were significantly associated with increased mortality, their addition to the CTA model did not improve predictive accuracy. The presence of liver disease and congestive heart failure were significant risk factors for mortality in the pneumonia severity index, but in our sample of patients with HIV infection, renal disease and neurologic symptoms were the only two coexisting illnesses associated with increased inpatient mortality. Our disease-specific staging system should be useful in future studies focused on patients with HIV-associated CAP by providing an efficient and easy method to measure and compare severity of illness at hospital admission.

There are several limitations to our study. We studied patients from 86 hospitals in seven metropolitan areas, and our staging system performed well in jackknife and bootstrap validity testing; however, further validation of our staging system is needed on different populations, such as outpatients with HIV-associated CAP and patients from rural areas. Second, validation of our findings in patients treated in the current era of highly active antiretroviral therapy is needed because our study was based on retrospective review of medical records from the early highly active antiretroviral therapy era (1995–1997) with a low rate of protease inhibitor use. Third, since CAP is caused by a variety of infectious organisms, it is possible that organism-specific severity-of-illness staging systems might perform better than our system. However, the etiologic cause of pneumonia is discovered in a minority of patients with HIV-associated CAP, making organism-specific systems less applicable to most patients.²¹ Fourth, our definition of CAP was based on ICD-9 codes for bacterial pneumonia and HIV-related disease, so that differences in hospital coding practices may have resulted in hospital level ascertainment bias. Fifth, dependence on ICD-9 codes may have also resulted in misclassification of pneumonia cases that were actually caused by PCP or other pathogens.²² We attempted to minimize misclassification by excluding patients with cytologically proven PCP or culture-proven pulmonary tuberculosis in the previous 30 days and patients with hospitalization for any reason in the previous 30 days.

Additional limitations to our study are related to the definitions of "neurologic symptoms" and "symptoms of neurologic change." First, the ascertainment of the presence of neurologic symptoms was dependent on whether a physician noted symptoms on day 1 or day 2 of the hospital admission. Second, there is a possibility that conditions such as mild peripheral neuropathy or HIV-associated opportunistic infections involving the CNS were included as "symptoms of neurologic change" because chart abstractors were not required to record symptom etiology. Third, the neurologic symptoms that were recorded may have been acute or chronic in nature. However, chart abstractors reported that most physician notes reporting neurologic symptoms referred to acute changes in mental status as opposed to chronic neurologic conditions. Also, if a patient had severe neurologic symptoms that required hospitalization within the past 30 days, they would have been excluded from our study population. Therefore, a majority of the neurologic symptoms included in our definition were probably acute in nature and related to mental status changes.

In conclusion, we have developed a staging system for HIV-associated CAP based on data from 1,415 patients with HIV infection who received care in 86 hospitals in seven metropolitan areas in the United States. The staging system categorizes patients into five stages associated with increasing risk of in-hospital death and had an overall classification accuracy of 85.2%. The clinical significance of our staging system is that by using three routinely obtained clinical and laboratory variables (presence of neurologic symptoms, respiratory rate, and creatinine level), one is able to estimate patient mortality risk at hospital admission. Clinicians may use the staging system to guide clinical decisions such as ICU admission, aggressive antibiotic treatment, and close monitoring of neurologic and renal status. Future studies and clinical guidelines examining outpatient vs inpatient management for HIV-associated CAP may utilize the staging system for classifying patients into high-risk and low-risk categories. Our disease-specific staging system can also be useful for stratifying patients in clinical trials evaluating antimicrobial agents for the treatment of HIV-associated CAP and for case-mix adjustment when evaluating variation in hospital mortality rates.

	Acknowledgements

 
We thank the nurse abstractors in each of the cities for data collection and the staffs of each of the hospitals for identification of medical records for review. We also thank the following collaborators: Shirin Ali, Robert Weinstein, Matthew Goetz, Rafael Campo, Jack Dehovitz, and Jeffrey Jacobson.

	Footnotes

 
Abbreviations: CAP = community-acquired pneumonia; CTA = classification tree analysis; ICD = International Classification of Diseases, ninth revision; PCP = Pneumocystis carinii pneumonia

Dr. Arozullah and Dr. Parada are supported by Research Career Development Awards from the Health Services Research and Development Service of the Department of Veterans Affairs.

This study was supported in part by a grant from the National Institute of Drug Abuse (5R01 DA10628-02) and the Health Services Research Division of the Department of Veterans Affairs.

Received for publication March 27, 2002. Accepted for publication October 8, 2002.

	References

TOP Abstract Introduction Materials and Methods Results Discussion References

 

1. Bartlett, JG (1998) Pneumonia in the patient with HIV infection. Infect Dis Clin North Am 12,807-820[ISI][Medline]
2. Murray, JF Pulmonary complications of HIV infection. Annu Rev Med 1996;47,117-126[CrossRef][ISI][Medline]
3. Hirschtick, RE, Glassroth, J, Jordan, MC, et al Bacterial pneumonia in persons infected with the human immunodeficiency virus. N Engl J Med 1995;333,845-851[Abstract/Free Full Text]
4. Palella, FJ, Delaney, KM, Moorman, AC, et al Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. N Engl J Med 1998;338,853-860[Abstract/Free Full Text]
5. Wallace, JM, Hansen, NI, Lavange, L, et al Respiratory disease trends in the pulmonary complications of HIV infection study cohort. Am J Respir Crit Care Med 1997;155,72-80[Abstract]
6. Afessa, B, Green, B Bacterial pneumonia in hospitalized patients with HIV infection: the pulmonary complications, ICU support, and prognostic factors of hospitalized patient with HIV (PIP) study. Chest 2000;117,1017-1022[Abstract/Free Full Text]
7. Mathews, WC, Ferdon, E, Bennett, CL, et al Evaluating institutional performance in AIDS-related Pneumocystis carinii pneumonia: a risk-adjustment approach. J Clin Epidemiol 1989;42,421-425[CrossRef][ISI][Medline]
8. Bennett, CL, Garfinkle, JB, Greenfield, S, et al The relation between hospital experience and in-hospital mortality for patients with AIDS related PCP. JAMA 1989;261,2975-2979[Abstract]
9. Bennett, CL, Weinstein, RA, Shapiro, MF, et al A rapid preadmission method for predicting inpatient course of disease for patients with HIV-related PCP. Am J Respir Crit Care Med 1994;150,1503-1507[Abstract]
10. Arozullah, AM, Yarnold, PR, Weinstein, RA, et al A new preadmission staging system for predicting inpatient mortality from HIV-associated Pneumocystis carinii pneumonia in the early highly active antiretroviral therapy (HAART) era. Am J Respir Crit Care Med 2000;161,1081-1086[Abstract/Free Full Text]
11. Fine, MJ, Hanusa, BH, Lave, JR, et al Comparison of a disease-specific and a generic severity of illness measure for patients with community-acquired pneumonia. J Gen Intern Med 1995;10,359-368[ISI][Medline]
12. Fine, MJ, Auble, TE, Yealy, DM, et al A prediction rule to identify low-risk patients with community-acquired pneumonia. N Engl J Med 1997;336,243-250[Abstract/Free Full Text]
13. Centers for Disease Control and Prevention. HIV AIDS Surveill Rep 1997;9,8-9
14. Oken, C, Archibald, N, Cvitanic, M, et al Multi-city study of quality of care for HIV-related PCP: successfully collecting highly sensitive information. Clin Perform Qual Health Care 1995;3,140-146[Medline]
15. Yarnold, PR Discriminating geriatric and non-geriatric patients using functional status information: an example of classification tree analysis via UniODA. Educ Psychol Meas 1996;56,656-667[Abstract]
16. Yarnold, PR, Soltysik, RC, Bennett, CL Predicting in-hospital mortality of patients with AIDS-related Pneumocystis carinii pneumonia: an example of hierarchically optimal classification tree analysis. Stat Med 1997;16,1451-1463[CrossRef][ISI][Medline]
17. Flanders, WD, Tucker, G, Krishnadasan, A, et al Validation of the pneumonia severity index: importance of study-specific recalibration. J Gen Intern Med 1999;14,333-340[CrossRef][ISI][Medline]
18. Cohn, DL Bacterial pneumonia in the HIV-infected patient. Infect Dis Clin North Am 1991;5,485-507[Medline]
19. Tumbarello, M, Tacconelli, E, de Gaetano, K, et al Bacterial pneumonia in HIV-infected patients: analysis of risk factors and prognostic indicators. J Acquir Immune Defic Syndr Hum Retrovirol 1998;18,39-45[ISI][Medline]
20. Cordero, E, Pachon, J, Rivero, A, et al Community-acquired bacterial pneumonia in human immunodeficiency virusinfected patients: validation of severity criteria. Am J Respir Crit Care Med 2000;162,2063-2068[Abstract/Free Full Text]
21. Magnenat, J, Nicod, LP, Auckenthaler, R, et al Mode of presentation and diagnosis of bacterial pneumonia in human immunodeficiency virus-infected patients. Am J Respir Crit Care Med 1991;144,917-922
22. Selwyn, PA, Pumerantz, AS, Durante, A, et al Clinical predictors of Pneumocystis carinii pneumonia, bacterial pneumonia, and tuberculosis in HIV-infected patients. AIDS 1998;12,885-893[CrossRef][ISI][Medline]

This article has been cited by other articles:

				C. R. Uphold, M. Deloria-Knoll, F. J. Palella Jr, J. P. Parada, J. S. Chmiel, L. Phan, and C. L. Bennett US Hospital Care for Patients With HIV Infection and Pneumonia: The Role of Public, Private, and Veterans Affairs Hospitals in the Early Highly Active Antiretroviral Therapy Era Chest, February 1, 2004; 125(2): 548 - 556. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Free Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Article Archive Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (9) Citing Articles via Google Scholar Google Scholar Articles by Arozullah, A. M. Articles by Yarnold, P. R. Search for Related Content PubMed PubMed Citation Articles by Arozullah, A. M. Articles by Yarnold, P. R.