Minor, Odile Le VD, MS*; Germani, Yves PhD†; Chartier, Loic MS*; Lan, Nguyen Huu MD‡; Lan, Nguyen T P MD‡; Duc, Nguyen H MD§; Laureillard, Didier MD‖; Fontanet, Arnaud MD, PhD*; Sar, Borann MD, PhD#; Saman, Manil MD‡‡; Chan, Sarin MD¶; L'Her, Pierre MD**; Mayaud, Charles MD††; Vray, Muriel MS*
Asia is faced with the fastest growing HIV-AIDS epidemic worldwide.1 Cambodia has the highest HIV prevalence in the Asia-Pacific region. Additionally, Southeast Asia carries a high burden of tuberculosis, Cambodia being the most affected country in the world.2,3
Definitive diagnosis of pulmonary tuberculosis (TB) is based on culture for Mycobacterium tuberculosis. However, cultures are not routinely available in developing countries. Microscopic examination of sputum smears for acid-fast bacilli (AFB) remains the cornerstone of the diagnosis as it is rapid, simple, and cheap.4,5 Unfortunately, AFB microscopy lacks sensitivity, especially in AIDS patients.6-8
Pneumocystis jiroveci pneumonia (PCP) is also a diagnosis of increasing importance in HIV-positive patients,9 accounting for significant mortality. However, the diagnosis of PCP is not easy in resources-constrained settings where fibreoptic bronchoscopy is not available.10
Therefore, despite the availability of efficient treatments, appropriate management of patients with TB or PCP remains a challenge in routine practice. Clinical tools available at admission could minimize avoidable drug toxicities and facilitate the timely choice of appropriate therapy as early initiation is a crucial component of successful treatment.
In this study, we aimed to identify clinical predictors of TB or PCP in HIV-infected patients with AFB smear-negative pneumonia in Cambodia and Vietnam to furthermore identify patients who are likely to have this pneumonia.
MATERIALS AND METHODS
The data were obtained in the Asian part of the ANRS-1260 cohort study conducted in the Preah Bat Norodom Sihanouk Hospital in Phnom Penh, Cambodia, and in the Pham Ngoc Thach Hospital in Ho Chi Min City, Vietnam, between September 2002 and October 2004. ANRS-1260 study aimed to determine the main causes of AFB sputum smear-negative pneumonia in Asian and African HIV-infected patients.11 Patients were included if they (1) were aged 18 years or older, (2) presented at least 1 clinical sign of lung infection, (3) presented new radiological diffuse or localized lung opacities and/or mediastinal adenopathies, and (4) were HIV infected.
Our analysis focused on patients who had at least 1 AFB smear-negative sputum.
The study was approved by the Cambodian and Vietnamese National Ethics Committees for Health Research.
Data collection and laboratory methods have been reported in detail elsewhere.11
After giving informed consent to participate, patients were questioned on general characteristics and clinical history including previous TB and PCP prophylaxis with trimethoprim-sulphamethoxazole (TMP-SMX) taken over the last month. Other clinical information included body mass index (BMI), the presence of headache, diarrhea, dyspnea, hemoptysis, and the stage of HIV infection [World Health Organization (WHO) classification]. Resting arterial oxygen saturation (Sao2) on ambient air was measured by a pulse oxymeter. Chest x-ray analysis was assessed by the local physician and/or radiologist and later reviewed by 2 specialists in pulmonary medicine blinded to microbiological and clinical data. Abnormalities, usually opacities, were summarized as localized and/or diffuse shadowing. Radiological findings of particular interest were the presence of pleural effusion, mediastinal adenopathy, and cavitation. In case of discordance, the specialists were asked to review the chest x-rays and to reach a consensus classification. Laboratory tests included CD4 cell counts, hemoglobinemia, and serologic testing for HIV.
These data were recorded on a report form without knowledge of the diagnosis.
Patients were asked to submit 1 sputum sample by spontaneous expectoration on the day of inclusion, and 2 other sputum samples thereafter. Samples were examined for AFB by microscopy after a Ziehl-Neelsen staining. In AFB smear-positive patients, antituberculous treatment was started according to the National Tuberculosis Program guidelines.
In other patients, endotracheal aspiration and/or bronchoalveolar lavage (BAL) were performed with Sao2 monitoring. Respiratory specimens were subjected to standard microbiological procedures for the identification of bacteria, fungi, and mycobacteria.11
“Diagnosis of confirmed TB” required the growth of M. tuberculosis from sputum or respiratory secretions obtained bronchoscopically on Lowenstein-Jensen media. TB was “excluded” when the culture remained negative at day 42.
“Diagnosis of confirmed PCP” required the presence of at least 5 typical cysts of P. jiroveci in the BAL sample using an immunofluorescence antibody staining. A double reading was performed in case of uncertain results. PCP was “excluded” in patients with a negative-stained BAL sample.
Data were double entered and analyzed using STATA software version 8 (Stata Corporation, College Station, TX).
Logistic Regression Analysis
Quantitative variables were expressed as median ± interquartiles range (IQR) (Q1-Q3). Quantitative variables were dichotomized around the median (BMI, age) or the clinical threshold value (CD4 cell counts, hemoglobinemia, Sao2).
Univariate analyses were performed, using the χ2 test or the Fisher's exact test where appropriate, to assess the association between categorical variables and pneumonia (PCP or TB). The variables potentially associated with a pneumonia (P ≤ 0.20) were included in the multivariate analysis. Stepwise logistic (backward) regression was undertaken; final models included variables independently associated with each disease at the threshold of P < 0.05. Other models were also developed using only data directly available to the physicians at the time of admission (clinical and radiological data).
The goodness-of-fit of the models was studied using 2 statistical analyses: the Hosmer-Lemeshow statistic12 and the Harrell c-index.13
To provide a simple-to-use estimate of the risk of disease for PCP or TB, we created a scoring system that utilized predictors identified in the multivariate analysis. The regression coefficients of the final model were rounded to the nearest integer.14 We verified that these rounded coefficients provided predictive ability similar to that provided by the original coefficients.15 Likelihood ratios (LRs) were calculated for each potential score by dividing the proportion of patients with the disease (confirmed diagnosis) by the proportion of patients without the disease (excluded diagnosis). Accepted levels for robust tests are as follows: LR ≤ 0.1 or LR ≥ 10; LR ≥ 5 or LR ≤ 0.2 give strong diagnostic evidence.16
We performed an independent validation in a different population because variations in case mix and local policies could affect the selection of variables used to predict the disease.17 We chose to develop our clinical decision rule on the largest Cambodian sample (training sample), which also corresponded to the first collected data. Then, the predictive ability and the calibration of the model were evaluated using the Vietnamese sample (validation sample).
Two hundred ninety-one AFB smear-negative patients meeting the inclusion criterions (193 in Cambodia and 98 in Vietnam) were consecutively included in the study. Among these patients, 16% submitted more than 1 sputum for AFB examination (3% in Cambodia and 43% in Vietnam).
Demographic Feature of the AFB Smear-Negative Population (Cambodian Sample)
The majority of the 193 AFB smear-negative Cambodian patients were men (65%). The median age was 34 years (IQR 30-39); the median BMI was 16.9 (IQR 14.7-18.6); and the median CD4 cell count was 11 cells/mm3 (IQR 4-27). Almost all of the patients were in stage III or IV of the WHO classification (99%). Most of them had knowledge of their HIV seropositivity before inclusion (62%), but few of them had received antiretroviral therapy (10%) and/or TMP-SMX prophylaxis (39%).
A definitive diagnosis of pneumonia was reported for 143 (74%) patients.11 The most frequent etiology was PCP (84/193; 43.5%), followed by bacterial pneumonia (36/193; 18.7%), and TB (27/193; 14.0%). The most common bacterial pathogens were Staphylococcus aureus and Pseudomonas sp.
Among patients with PCP, 3 were coinfected with bacteria (2 with S. aureus and 1 with Pseudomonas aeruginosa), 3 with cryptococcosis, and 5 with TB. Among patients with TB, 2 were coinfected with bacteria (1 with S. aureus and 1 with Streptococcus pneumoniae) and 1 with cryptococcosis.
Among the 193 AFB smear-negative patients, we kept only 172 (89%) patients fully investigated for TB (confirmed/excluded TB) in the model for predicting TB; patients who did not undergo the Lowenstein culture (n = 11) and those for whom the culture was contaminated (n = 10) were excluded. Likewise, only 160 (87%) patients fully investigated for PCP (confirmed/excluded PCP) were included in the model for predicting PCP. Thirty-three patients were excluded as they did not undergo BAL due to refusal or critical clinical condition (Fig. 1).
Univariate comparisons between confirmed TB (n = 27) and non-TB (n = 145) patients indicated that mediastinal adenopathy, localized opacity, and a lack of diffuse shadowing were the most strongly predictive factors of TB (Table 1). The results of the univariate analysis between confirmed PCP (n = 84) and non-PCP (n = 76) showed that dyspnea, Sao2 <90%, the absence of TMP-SMX prophylaxis, normal hemoglobinemia value, diffuse shadowing, and a lack of localized opacity were strongly associated with PCP (Table 2).
Logistic Regression Analysis
In the multivariate analysis, headache, localized opacity, and mediastinal adenopathy were independently associated with an increased risk of TB (Table 1). The predictive ability and the calibration of the model were good (c-index = 0.85; Hosmer-Lemeshow statistic = 0.7, degree of freedom = 3, P = 0.88).
For PCP, independent predictors included CD4 cell counts ≤50 cells/mm3, diffuse shadowing, Sao2 <90%, and the absence of TMP-SMX prophylaxis (Table 2). The predictive ability of the model was good (c-index = 0.89), and the Hosmer-Lemeshow test showed an excellent calibration (0.97, df = 5, P = 0.97). When data were analyzed with the exclusion of biological results (ie, data available immediately on admission), diffuse shadowing, Sao2 <90%, and the absence of TMP-SMX prophylaxis remained independently associated with the risk of PCP (Table 3). The discrimination and the calibration of this simplified model were good (c-index = 0.87; Hosmer-Lemeshow statistic = 1.17, df = 4, P = 0.88).
Because of the small number of patients with TB (n = 27), we did not attempt to develop a prediction rule. We focused solely on patients with PCP. Using the clinical and radiographic variables that were independently associated with the risk of PCP, we created a score to differentiate PCP from other diseases in AFB smear-negative patients. The β-coefficients derived from the predictors were rounded to the nearest integer to make the scoring system simpler to use in routine clinical practice. Loss of fit due to the rounded coefficients was negligible (c-index = 0.87; Hosmer-Lemeshow statistic = 0.47, df = 4, P = 0.98).
The score is simply computed by a linear combination of the coefficients: PCP score = 2 × diffuse shadowing + 1 × Sao2 <90% and ≥80% or + 2 × Sao2 <80% − 2 × TMP-SMX prophylaxis.
Each factor was assigned the value of 1 if present or 0 if absent. PCP score ranged from −2 to +4.
The risk of PCP can thus be calculated in our population by the appropriate logistic transformation:
Equation (Uncited)Image Tools
Table 3 displays, for each possible score, the frequency of PCP (observed risk of PCP) and the predicted risk of PCP in Cambodian patients.
These predicted risks of PCP were correlated to the prevalence of PCP in the population. Therefore, it is useful to calculate LRs, which measure the performance of the score, independent of the prevalence, to the extent that the prediction rules remain valid in different populations. The LRs ranged from 0.04 for the negative scores to 10.4 for the highest score (Table 3).
Validation (Vietnamese Sample)
Most of the 98 AFB smear-negative Vietnamese patients were men (76%) with a median age of 29 years (IQR 24-35) and a median BMI of 16.9 (IQR 15.6-18.2). The median CD4+ cell counts were 22 cells/mm3 (IQR 10-42), and 72% of the patients were in stage III or IV of the WHO classification for HIV infection. Few of these patients had knowledge of their HIV seropositivity on admission (16% vs 62% in Cambodia). None of them received antiretroviral therapy (vs 10% in Cambodia), and the administration of TMP-SMX prophylaxis was extremely limited (7% vs 39% in Cambodia).
Full investigations for TB and PCP were, respectively, performed in 83 and 69 AFB smear-negative patients (Fig. 1); TB was confirmed in 11 (13%) patients and PCP in 38 (55%) patients (1 coinfection with Escherichia coli).
The prevalence of PCP was similar in the Cambodia and Vietnam (53% and 55%, respectively) samples; therefore, the risk predicted by the score remained valid. The risk of PCP observed in this validation sample is given in Table 3. The LRs ranged from 0.08 to 6.7. When tested in this independent validation sample, the PCP-predicting score maintained correct predictive ability (c-index = 0.72) but showed poor calibration (Hosmer-Lemeshow statistic = 6.7, df = 2, P = 0.04). Figure 2 shows the percentage of PCP observed in the training and validation data sets plotted against the percentage of PCP predicted by the score. Unfortunately, the probability of PCP was not assessable for some values of the score (−2, −1, 1). The line produced for the validation sample is close to the diagonal, meaning that the predicted probabilities are good estimations, except for the score value = 3.
This study evaluated 291 Asian HIV-infected patients affected by AFB smear-negative pneumonia. These patients were young adults at an advanced stage of HIV infection with unequal use of PCP prophylaxis and antiretroviral treatment. Among Cambodian and Vietnamese patients in whom reference tests were performed, the prevalence of TB was, respectively, 16% and 13%, and the prevalence of PCP was 53% and 55%. A comparison with previous Southeast Asian studies (with prevalence of PCP ranging from 5% to 30%) is very difficult because these studies did not focus on AFB smear-negative patients and/or usually the diagnosis was made with clinical standards without laboratory confirmation.18-25
Predictors of AFB Smear-Negative TB
In our analysis, radiographic features predictive of AFB smear-negative TB were the presence of mediastinal adenopathy and localized opacity. The association of mediastinal adenopathy and TB was already well known in severely immunocompromised HIV-infected patients.26-28 As in other studies, cavitations were infrequent in our patients with AFB smear-negative TB.14,29,30 Interestingly, cavitations were mostly present in patients with bacterial pneumonia. Associated pleural effusion was observed in only 4 patients (1 TB), consistent with the previous findings that tuberculous pleuritis is more common in HIV-infected patients with ≥200 CD4 cells/mm3.27,28
Headache was the sole independent clinical predictor of AFB smear-negative TB. According to a Vietnamese study, headache might be the manifestation of tuberculous meningoencephalitis or metabolic disorder such as hyponatremia.31
Previous studies identified hemoptysis, weight loss, fever, and persistent cough (lasting longer than 2 weeks) without expectoration as clinical predictors of AFB smear-negative TB in African populations with high prevalence of HIV infection.14,29,30,32 However, in our study, hemoptysis was less common in TB than in other diseases (not significant). Likewise, the presence of cough was not a useful indicator of TB because it lacked sufficient specificity (present in 99% of the patients). Unfortunately, data on duration of cough, which could have increased the specificity of the factor, were unreliable. Indeed, the interval of time between the onset of pneumonia and admission was indefinite because of confusion between the symptoms of pneumonia and those of AIDS. This is especially true because the majority of patients in the study were at late stage of HIV disease. Temperature at admission was also unreliable because of the frequent and uncontrolled use of antipyretics.
The c-index and Homer-Lemeshow test showed good fit of the model; however, it is well known that goodness-of-fit tests are less powerful for moderate sample sizes.33 Because of the low power of the statistical comparison (27 TB), we cannot exclude other factors as potential predictors, and developing a score seemed not acceptable.
Predictors of PCP and Scoring System
Diffuse shadowing on radiography was the strongest predictor of PCP [odds ratio (OR) = 7.0, P < 10−3]. Low Sao2 (≤90% and <80%) was statistically associated with an increased risk of PCP (OR = 3.3 and OR = 9.1, respectively, P = 0.005). Eighty percent of patients with Sao2 ≤90% and 91% with Sao2 <80% had PCP. This result, consistent with previous reports, highlights the usefulness of the noninvasive and safe Sao2 measurement.34-37 The equipment required for Sao2 monitoring is inexpensive, portable, does not require a specialized operator, and provides results immediately available.36 Other studies have found high respiratory rate as useful predictor of PCP.24,38 Further studies should evaluate if this feature could provide an alternative measure when a pulse oxymeter is unavailable. Dyspnea was more common in patients with PCP but was not an independent predictor (P = 0.06). Indeed, dyspnea may result from hypoxemia as in PCP but also from anemia as in TB. The only laboratory data independently associated with the presence of PCP were low CD4 cell counts. However, the high level of immunosuppression in the whole population (82% of the patients had CD4 cell count <50 cells/mm3) lessens the relevance of this factor.
A study carried out in the United Kingdom has found that exercise-induced oxygen desaturation, interstitial infiltrates, absence of PCP prophylaxis, and symptoms triad of dyspnea, cough, and fever as useful predictors of PCP in HIV-infected patients.35 In the Unites States, independent predictors of PCP in HIV-positive patients presented with respiratory disease were exertional dyspnea, interstitial infiltrate, and presence of oral thrush.39 However, in our patients hospitalized at a more advanced stage of the disease, hypoxemia was present at rest, opacities were diffuse in 90% of the cases, and it was impossible to disentangle alveolar and interstitial patterns.
We developed a PCP score that includes low Sao2, absence of TMP-SMX prophylaxis, and diffuse shadowing. This score, based on rounded coefficients, provides a simple and rapid assessment of the probability of PCP in AFB smear-negative patients with advanced HIV disease.
PCP score ranges from −2 to +4. Using this score with a cutoff of ≥+3 to identify patients with PCP, we would have correctly classified 55 of 84 PCP (sensitivity = 65%) and 68 of 76 excluded PCP (specificity = 89%) in the Cambodian sample. Using ≥+2 as a cutoff would have increased the sensibility to 89% but lowered the specificity to 74%. In the Vietnamese sample, sensitivity and specificity would have been 43% and 77%, respectively, for a cutoff ≥+3 and 97% and 46% for a cutoff ≥+2. The LRs ranged from 0.04 to 10.4. Considering the Cambodian sample, the score seemed to be particularly useful for 31% of the patients with extreme values (+4 and negative scores), for whom estimated probabilities are robust (respectively, 92% and 4%), and helpful for 46% (+3 and 0). The scoring system was of limited use for 23% of the patients with intermediate scores (+1 and +2). We would like to emphasize that the score allows for the identification of patients needing further investigation for PCP. Indeed, among patients with intermediate scores, it may be desirable to perform induced sputum or BAL when available.
Owing to possible biases in 1-center population sampling,40 we performed an independent validation study in a different center located in another country. The probability of PCP was not assessable for some values of the score (−2, −1, and +1) not met by the patients of the validation sample. The score maintained a good predictive ability, but Hosmer-Lemeshow statistic showed a lack of fit. The absence of important predictors in our model could explain this weakened performance in a population with different case mix. However, no other clinical predictors of PCP could be identified in the Vietnamese population. The sample size may also be too small to perform a correct validation. Comparison of observed to predicted probabilities indicated that the model did not perform satisfactorily for the score +3. This discrepancy might be the consequence of the small size of the validation sample; for example, with 14 patients presenting a score +3, minor variations in the number of PCP could strongly impact the observed risk of PCP.
There are certain limitations to this study. First, 62 patients who did not undergo BAL were not included in the PCP statistical analyses, which could lead to selection biases. Statistical comparisons pointed out that the mortality rate was significantly higher in this group but did not show any difference between the population included in the analyses and the population without BAL. Second, although WHO recommends the examination for AFB on repeated sputum, only 1 sputum was collected for most of the patients. Consequently, this score should be calculated in patients directly after obtaining a first negative sputum. This fits well with observed clinical practice, as the score can be computed directly at admission, allowing for early identification of PCP. Third, this hospital-based study involved highly immunocompromised patients; it is unclear whether these findings may be applied to general practices.
This study confirms that PCP is of major importance in HIV-positive patients with “AFB smear-negative pulmonary disease” in Southeast Asia. The PCP-predicting score, based on few clinical and radiographic predictors available on admission to hospital in low-income countries, will optimize the identification of patients with PCP, allowing timely therapy. However, other validation studies would be required in similar and different settings to confirm these initial findings.
We wish to thank Pamela Snupf for rereading the manuscript.
1. Ruxrungtham K, Brown T, Phanuphak P. HIV/AIDS in Asia. Lancet. 2004;364:69-82.
2. World Health Organization. Global Tuberculosis Control. WHO Report 1998. Geneva, Switzerland, WHO/CDS/TB/2001.287.
3. Dye C, Scheele S, Dolin P, et al. Consensus statement. Global burden of tuberculosis: estimated incidence, prevalence, and mortality by country. WHO Global Surveillance and Monitoring Project. JAMA. 1999;282:677-686.
4. World Health Organization. Laboratory Services in Tuberculosis Control. WHO Report 1998. Geneva, Switzerland. WHO\TB\98.258.
5. Rieder HL, Chonde TM, Myking H, et al. The Public Health Service National Tuberculosis Reference Laboratory and the National Laboratory Network. Minimum Requirements, Role and Operation in a Low-Income Country. Paris, France; IUATLD 1998.
6. Elliott AM, Namaambo K, Allen BW, et al. Negative sputum smear results in HIV-positive patients with pulmonary tuberculosis in Lusaka, Zambia. Tuber Lung Dis. 1993;74:191-194.
7. Colebunders RL, Braun MM, Nzila N, et al. Evaluation of the World Health Organization clinical case definition of AIDS among tuberculosis patients in Kinshasa, Zaire. J Infect Dis. 1989;160:902-903.
8. Brindle RJ, Nunn PP, Githui W, et al. Quantitative bacillary response to treatment in HIV-associated pulmonary tuberculosis. Am Rev Respir Dis. 1993;147:958-961.
9. Morris A, Lundgren JD, Masur H, et al. Current epidemiology of Pneumocystis pneumonia. Emerg Infect Dis. 2004;10:1713-1720.
10. Coleman DL, Dodek PM, Luce JM, et al. Diagnostic utility of fiberoptic bronchoscopy in patients with Pneumocystis carinii pneumonia and the acquired immune deficiency syndrome. Am Rev Respir Dis. 1983;128:795-799.
11. Vray M, Germani Y, Chan S, et al. Clinical features and etiology of pneumonia in AFB sputum smear-negative HIV-infected patients hospitalised in Asia and Africa. AIDS. 2008;22:1323-1332.
12. Hosmer DW, Lemeshow S. Applied Logistic Regression. New York, NY: Wiley; 1989.
13. Harrell FE Jr., Lee KL, Califf RM, et al. Regression modelling strategies for improved prognostic prediction. Stat Med. 1984;3:143-152.
14. Kanaya AM, Glidden DV, Chambers HF. Identifying pulmonary tuberculosis in patients with negative sputum smear results. Chest. 2001;120:349-355.
15. Coste J, Wasserman D, Venot A. Predicting mortality in adult burned patients: methodological aspects of the construction and validation of a composite ratio scale. J Clin Epidemiol. 1996;49:1125-1131.
16. Jaeschke R, Guyatt GH, Sackett DL. Users' guides to the medical literature. III. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients? The Evidence-Based Medicine Working Group. JAMA. 1994;271:703-707.
17. McGinn TG, Guyatt GH, Wyer PC, et al. Users' guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group. JAMA. 2000;284:79-84.
18. Kimerling ME, Schuchter J, Chanthol E, et al. Prevalence of pulmonary tuberculosis among HIV-infected persons in a home care program in Phnom Penh, Cambodia. Int J Tuberc Lung Dis. 2002;6:988-994.
19. Senya C, Mehta A, Harwell JI, et al. Spectrum of opportunistic infections in hospitalized HIV-infected patients in Phnom Penh, Cambodia. Int J STD AIDS. 2003;14:411-416.
20. Pichith K, Chanroeun H, Bunna P, et al. Clinical aspects of AIDS at the Calmette hospital in Phnom Penh, Kingdom of Cambodia A report on 356 patients hospitalized in the Medicine “B” Department of the Calmette Hospital. Sante. 2001;11:17-23.
21. Tansuphasawadikul S, Amornkul PN, Tanchanpong C, et al. Clinical presentation of hospitalized adult patients with HIV infection and AIDS in Bangkok, Thailand. J Acquir Immune Defic Syndr. 1999;21:326-332.
22. Oh MD, Park SW, Kim HB, et al. Spectrum of opportunistic infections and malignancies in patients with human immunodeficiency virus infection in South Korea. Clin Infect Dis. 1999;29:1524-1528.
23. Wannamethee SG, Sirivichayakul S, Phillips AN, et al. Clinical and immunological features of human immunodeficiency virus infection in patients from Bangkok, Thailand. Int J Epidemiol. 1998;27:289-295.
24. Mootsikapun P, Chetchotisakd P, Intarapoka B. Pulmonary infections in HIV infected patients. J Med Assoc Thai. 1996;79:477-485.
25. Louie JK, Chi NH, Thao le TT, et al. Opportunistic infections in hospitalized HIV-infected adults in Ho Chi Minh City, Vietnam: a cross-sectional study. Int J STD AIDS. 2004;15:758-761.
26. Jasmer RM, Edinburgh KJ, Thompson A, et al. Clinical and radiographic predictors of the etiology of pulmonary nodules in HIV-infected patients. Chest. 2000;117:1023-1030.
27. Perlman DC, el-Sadr WM, Nelson ET, et al. Variation of chest radiographic patterns in pulmonary tuberculosis by degree of human immunodeficiency virus-related immunosuppression. The Terry Beirn Community Programs for Clinical Research on AIDS (CPCRA). The AIDS Clinical Trials Group (ACTG). Clin Infect Dis. 1997;25:242-246.
28. Pozniak AL, MacLeod GA, Ndlovu D, et al. Clinical and chest radiographic features of tuberculosis associated with human immunodeficiency virus in Zimbabwe. Am J Respir Crit Care Med. 1995;152(5 Pt 1):1558-1561.
29. Samb B, Henzel D, Daley CL, et al. Methods for diagnosing tuberculosis among in-patients in eastern Africa whose sputum smears are negative. Int J Tuberc Lung Dis. 1997;1:25-30.
30. Tessema TA, Bjune G, Assefa G, et al. An evaluation of the diagnostic value of clinical and radiological manifestations in patients attending the Addis Ababa tuberculosis centre. Scand J Infect Dis. 2001;33:355-361.
31. Ngo AT, Duc NH, Lan NH, et al. Mechanisms and causes of death in 143 Vietnamese VIH-infected patients hospitalized for tuberculosis. Rev Pneumol Clin. 2007;63:139-146.
32. Bah B, Massari V, Sow O, et al. Useful clues to the presence of smear-negative pulmonary tuberculosis in a West African city. Int J Tuberc Lung Dis. 2002;6:592-598.
33. Hosmer DW, Hosmer T, Le Cessie S, et al. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med. 1997;16:965-980.
34. Smith DE, McLuckie A, Wyatt J, et al. Severe exercise hypoxaemia with normal or near normal X-rays: a feature of Pneumocystis carinii infection. Lancet. 1988;2:1049-1051.
35. Smith DE, Forbes A, Davies S, et al. Diagnosis of Pneumocystis carinii pneumonia in HIV antibody positive patients by simple outpatient assessments. Thorax. 1992;47:1005-1009.
36. Chouaid C, Maillard D, Housset B, et al. Cost effectiveness of noninvasive oxygen saturation measurement during exercise for the diagnosis of Pneumocystis carinii pneumonia. Am Rev Respir Dis. 1993;147(6 Pt 1):1360-1363.
37. Aderaye G, Bruchfeld J, Aseffa G, et al. Pneumocystis jiroveci pneumonia and other pulmonary infections in TB smear-negative HIV positive patients with atypical chest X-ray in Ethiopia. Scand J Infect Dis. 2007;39:1045-1053.
38. Malin AS, Gwanzura LK, Klein S, et al. Pneumocystis carinii pneumonia in Zimbabwe. Lancet. 1995;346:1258-1261.
39. 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.
40. Altman DG, Royston P. What do we mean by validating a prognostic model? Stat Med. 2000;19:453-473.
© 2008 Lippincott Williams & Wilkins, Inc.