Alka M. Kanaya, MD; David V. Glidden, PhD; and Henry F. Chambers, MD

Transcription

1 Identifying Pulmonary Tuberculosis in Patients With Negative Sputum Smear Results* Alka M. Kanaya, MD; David V. Glidden, PhD; and Henry F. Chambers, MD Background: Clinicians need to decide whether to begin empiric therapy for patients who are suspected of having tuberculosis (TB) but have negative sputum smear results. Culture results may take weeks, and delaying treatment may allow further transmission of disease. Study objective: To identify the clinical, demographic, and radiographic characteristics that identify smear-negative patients who have TB, and to create a TB prediction rule. Design: Retrospective chart review. Setting: University-affiliated public hospital in San Francisco, CA, between 1993 and Patients: Forty-seven patients with TB and 141 control patients who were hospitalized with a suspicion of pulmonary TB; all had negative sputum smear results. Measurements and results: Demographic, clinical, and radiographic variables were determined by chart review. In multivariate analysis, a positive tuberculin skin test result (odds ratio [OR], 4.8; 95% confidence interval [CI], 2.0 to 11.9) was independently associated with an increased risk of a positive TB culture finding. A radiographic pattern not typical of pulmonary tuberculosis (OR, 0.3; 95% CI, 0.1 to 0.7) and expectoration with cough (OR, 0.3; 95% CI, 0.1 to 0.6) were predictive of a decreased risk. An interaction between HIV seropositivity and mediastinal lymphadenopathy on the chest radiograph was also associated with a positive TB culture result (OR, 7.2; 95% CI, 1.4 to 36.0). The TB prediction score (TPS) was created with widely ranging likelihood ratios that could affect the posterior probability of TB by 30-fold. Conclusion: The TPS put into context with the overall prevalence of TB in a given area may help clinicians decide if a patient with negative sputum smear results should start empiric antituberculous therapy or wait for culture results. These results need prospective validation. (CHEST 2001; 120: ) Key words: prediction rule; smear negative; tuberculosis Abbreviations: AFB acid-fast bacilli; CI confidence interval; LR likelihood ratio; OR odds ratio; TB tuberculosis; TPS tuberculosis prediction score; TST tuberculin skin test Since the resurgence of pulmonary tuberculosis (TB) in the United States in the late 1980s, much attention has been focused on early case identification and treatment. 1 Several studies 2 9 have identified the clinical characteristics of persons with the most infectious form of TB, those harboring the largest number of organisms, with acid-fast bacilli *From the Divisions of General Internal Medicine (Dr. Kanaya) and Infectious Diseases (Dr. Chambers), Department of Medicine, and Department of Epidemiology and Biostatistics (Dr. Glidden), University of California, San Francisco, San Francisco, CA. Work performed at San Francisco General Hospital, University of California, San Francisco, CA. Dr. Kanaya is supported by a grant from the Department of Health and Human Services: Faculty Development in General Internal Medicine grant No. 1D08PE Correspondence to: Alka M. Kanaya, MD, 1701 Davisadero St, Suite 554, San Francisco, CA ; alkak@ itsa.ucsf.edu (AFB) found by microscopic examination of stained sputum (AFB smear positive). However, patients with active TB who have negative sputum smear results are also capable of transmitting the infection. 10,11 The relative transmission rate of smearnegative TB patients compared to smear-positive TB For editorial comment see page 330 patients has been calculated at 22% using a molecular epidemiologic technique. 12 Although persons with smear-negative TB are less infectious than the smear-positive patients, their overall contribution to disease transmission is considerable because half of all patients with TB can present with negative sputum smear findings. 13 Given the smaller mycobacterial burden present with smear-negative disease, these patients may have CHEST / 120 / 2/ AUGUST,

2 different clinical and radiographic findings than those with smear-positive disease. We are aware of only one previous study 14 from West Africa that identified predictors of smear-negative patients with TB. They found that an absence of cavitation, lack of cough, presence of HIV seropositivity, CD4 cell count 200/ L, and age 40 years predicted patients with smear-negative TB. 14 Likewise, multiple reports from sub-saharan countries have found an increased prevalence of smear-negative TB in their HIV-infected patients. However, this association between HIV and AFB smear-negative disease has not been observed in the United States. 19,20 We investigated the predictors of TB in smear-negative patients in an urban population in the United States, since these predictors may be quite different from those reported from Africa. Early identification of persons who have TB, whether smear positive or smear negative, is desirable both to enable appropriate isolation procedures and to provide a basis for early institution of therapy. Conversely, correct prediction of persons who are unlikely to have TB is important as well to limit the expense and potential toxicity of empiric therapy. We aimed to identify the clinical, demographic, and radiographic predictors for smear-negative TB to aid clinicians in predicting the likelihood of TB in persons with negative sputum smear results. We developed a clinical prediction rule that may help physicians identify those patients with negative smear results who are likely to have TB and should be started on a regimen of empiric antituberculous therapy. Study Setting and Patients Materials and Methods The study was conducted at San Francisco General Hospital, a university-affiliated public hospital. Patients included in the study were adults admitted to the hospital between 1993 and 1998 with the suspicion of TB, who had at least two negative sputum smear samples and corresponding sputum culture results available. Sputum samples were obtained by spontaneous morning expectoration, saline solution induction, tracheal aspiration, or bronchoscopy with BAL. All sputum smears were concentrated and examined by trained microbiology technicians. Each sputum smear was cultured by both BACTEC 12B broth (Becton- Dickenson; Cockeysville, MD) and Middlebrook 7H11 selective agar and maintained for at least 6 weeks to detect the presence of growing organisms. All patients had a medical chart with a hospital admission note, microbiology results, and chest radiograph interpretation by a board-certified radiologist. Physician, nurse, social worker, and medical student notes during the hospitalization were reviewed for complete data on each patient. Patients with a history of TB (pulmonary or extrapulmonary) or those currently receiving antituberculous therapies were excluded. Control patients could not have a diagnosis of TB made in the year following their index hospitalization. Study Design and Data Collection We used a case-control design to evaluate potential predictors of TB in smear-negative patients. We sampled all consecutive patients with smear-negative culture-positive TB from the San Francisco General Hospital Clinical Laboratory records. Three control patients per study patient were sampled at random from patients who were admitted to the hospital for suspicion of TB during the same week and had negative smear and culture results. Approximately five potential control patients existed for each study patient (case). We recorded standard risk factors for TB infection and disease as defined by the American Thoracic Society and Centers for Disease Control and Prevention, 21 including sex, age, ethnicity, country of birth, current or prior homelessness, and current or prior history of incarceration. Clinical information recorded included the presence of cough, expectoration, hemoptysis, temperature 38.5 C, night sweats, weight loss, symptom chronicity, HIV seropositivity, current or prior tuberculin skin test (TST) result, known exposure to TB, prior isoniazid prophylaxis, alcoholism, tobacco use, and other comorbidities associated with TB, such as diabetes, end-stage renal disease, hematologic cancer, or chronic steroid use. The results of the chest radiograph were categorized as normal, upper/apical lobe disease (either infiltrate or cavity), other pattern of infiltration not typically associated with reactivation TB (lobar or diffuse pattern), pleural effusion, mediastinal lymphadenopathy, or miliary pattern. We recorded the number of sputum smears analyzed for each patient as well. The chart reviewer was blinded to the sputum-culture status of the patient. Culture results and initial smear interpretation of each sputum sample were verified. Statistical Analysis and Missing Data Management Univariate comparisons between study patients and control patients were performed using Fisher s Exact Test for categorical variables and the Student s t test or the Wilcoxon rank-sum test for continuous variables where appropriate. All tests of significance were two sided; p 0.05 was considered statistically significant. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. Multiple logistic regression analysis was conducted using software (Stata 6.0; Stata Corporation; College Station, TX) to identify independent characteristics of positive culture results for Mycobacterium tuberculosis. We identified potential predictor variables for smear-negative TB using univariate analysis, in which p 0.10 determined entry into the multivariate models. We checked for interactions between HIV and chest radiograph findings and TST results. We used a backward selection technique to choose our final model. Missing data were managed using three different analytical methods. We ran our logistic regression (1) by excluding subjects who were missing variables necessary to enter the model, (2) by eliminating the variable with the largest amount of missing data, and (3) by creating an indicator variable to represent the missing data. 22,23 We found that all three techniques yielded similar multivariate results. The results obtained using indicator variables are presented. Using the -coefficients derived from the independent predictors in our logistic regression model, we created a scoring system to clinically apply these predictors. To simplify the predictive model we rounded the -coefficient to the nearest integer. We 350 Clinical Investigations

3 calculated the area under a receiver operating characteristic curve for the prediction rule. We calculated likelihood ratios (LRs) for each potential score. LRs were determined by dividing the proportion of study patients with the score by the proportion of control patients with the score. Results The study included 47 patients with smear-negative TB (study patients) and 141 control patients. Demographic characteristics, clinical findings, and radiographic results of study patients and control patients are shown in Table 1. Most study patients and control patients were men with a median age of approximately 40 years. Study patients were more likely to be Latino, born in Central America, and have histories of incarceration and a positive TST result. Control patients were more likely to be white, born in the United States, and have expectoration with cough and a fever. We found no difference in the rates of HIV seropositivity between the study patients and the control patients. There was no clear relationship with symptom duration for either group, but control patients had a trend toward longer symptom duration. A total of 85 patients (45%) did not have a TST placed during their hospitalization Table 1 Characteristics of the Study Population* Characteristics Study Patients (n 47) Control Patients (n 141) OR (95% CI) p Value Demographic Median age (SD) 38.4 (13.0) 42.0 (13.2) 0.8 ( ) 0.08 Male sex 40 (85) 123 (87) 1.2 ( ) 0.71 Race/ethnicity: White 13 (28) 64 (45) 1.0 African American 13 (28) 44 (31) 3.0 ( ) 0.02 Latino 14 (30) 23 (16) 1.5 ( ) 0.39 Asian/Pacific Islander 7 (15) 9 (6) 3.8 ( ) 0.02 Country of Birth United States 30 (64) 117 (83) 1.0 Central America 12 (26) 14 (10) 3.3 ( ) Asia/Pacific Islands 5 (11) 8 (5) 2.4 ( ) 0.14 Incarceration 14 (30) 19 (14) 2.7 ( ) 0.01 Homelessness 14 (30) 49 (35) 0.8 ( ) 0.53 Clinical Any cough 27 (58) 125 (89) 0.2 ( ) Expectoration 17 (36) 98 (70) 0.3 ( ) Hemoptysis 6 (13) 10 (7) 1.9 ( ) 0.23 Temperature 38.5 C 23 (49) 96 (68) 0.4 ( ) 0.02 Night sweats 22 (47) 63 (45) 1.1 ( ) 0.80 Weight loss 10 lb 20 (44) 55 (41) 1.1 ( ) 0.69 Symptom chronicity, wk 1 9 (21) 8 (6) (10) 55 (40) 0.1 ( ) (33) 33 (24) 0.4 ( ) (7) 18 (13) 0.2 ( ) (29) 24 (17) 0.4 ( ) 0.18 HIV positive 19 (46) 74 (54) 0.6 ( ) 0.15 TST result positive 29 (81) 33 (50) 4.3 ( ) Exposure to TB 5 (11) 9 (6) 1.7 ( ) 0.34 Prior isoniazid use 8 (17) 14 (10) 1.7 ( ) 0.19 Alcoholism 20 (43) 56 (40) 1.1 ( ) 0.73 Tobacco use 30 (64) 96 (68) 0.8 ( ) 0.59 Other comorbidities 1 (2) 6 (4) 0.5 (0 3.2) 0.50 Radiographic findings Abnormal 45 (96) 131 (93) 1.7 ( ) 0.49 Apical/upper lobe disease 20 (43) 35 (25) 2.2 ( ) 0.02 Atypical infiltrate for TB 13 (28) 94 (67) 0.2 ( ) Mediastinal lymphadenopathy 10 (21) 10 (7) 3.5 ( ) Pleural effusion 7 (15) 15 (11) 1.5 ( ) 0.43 Miliary infiltrate 3 (6) 8 (6) 1.1 ( ) 0.87 *Data are presented as No. (%) unless otherwise indicated. Based on 36 study patients and 67 control patients for whom data were available. Patients with diabetes, end-stage renal disease, hematologic cancer, or long-term steroid use. CHEST / 120 / 2/ AUGUST,

4 and had no documentation of a prior known result. The proportion of patients with missing TST varied between the study patients and control patients (23% vs 52%, respectively). Of note, 19 patients (40%) with smear-negative TB were not started on any antituberculous therapy during their hospitalization or at the time of hospital discharge. There was no difference between the groups in the number of sputum smears analyzed; a majority of patients had two to three sputum smears (79% of study patients vs 74% of control patients) during their hospitalization. In the multivariate analysis, a positive TST result (OR, 4.8; 95% CI, 2.0 to 11.9) was independently associated with an increased risk of a positive TB culture result, whereas a radiographic pattern not typical of reactivation pulmonary TB (OR, 0.3; 95% CI, 0.1 to 0.7) and expectoration with cough (OR, 0.3; 95% CI, 0.1 to 0.6) were predictive of a decreased risk. We detected an interaction between HIV-positive status and the finding of mediastinal lymphadenopathy on a chest radiograph that was associated with an increased risk of a positive TB culture result as well (OR, 7.2; 95% CI, 1.4 to 36.0; Table 2). Using the four variables that were associated with the risk of TB in multivariate analysis, we created a TB prediction score (TPS) to help distinguish patients who were culture positive from patients likely to be culture negative. Based on the magnitude of the -coefficient, a positive TST result was given a point score of 1, the negative predictors (expectoration and an infiltrate not typical of TB) received 1 point each, and HIV positivity and mediastinal lymphadenopathy together received 2 points. Table 3 displays the range and frequency of each score for the study patients and control subjects. Since LRs offer more valuable information than sensitivity and specificity when a test has more than two possible results, we calculated separate LRs for each possible score. Since no cases had a score of Table 2 Multivariate Predictors of TB in Smear- Negative Patients Predictors Multivariate OR (95% CI) p Value Points Clinical Expectoration 0.3 ( ) TST result positive 4.8 ( ) Radiographic Infiltrate not typical of TB 0.3 ( ) interaction HIV-positive status and mediastinal lymphadenopathy on chest radiograph 7.2 ( ) Table 3 TPS Using Multivariate Predictors* Score of Points Study Patients 2or 3, we combined all of the positive scores in order to ascertain a meaningful LR. The LRs ranged widely, from 0.2 for a score of 2, to 7.1 for positive scores. A receiver operating characteristic curve, which plots the false-positive rate against the truepositive rate for each possible cutoff for a diagnostic test, had an area under the curve of 0.89 for our prediction model. A perfect diagnostic test with 100% sensitivity and specificity would have an area under the curve of 1.0. To demonstrate the utility of the TPS for clinical decision making, we estimated the probability of TB in a hypothetical patient with negative sputum smear results in three different areas with varying prevalence rates of TB (Table 4). The TPS could affect the posterior likelihood of TB by approximately 30-fold in each of the scenarios described. If the threshold for empiric treatment is 5%, in cities with high TB prevalence, a score of 0 would predict the risk of TB to exceed the threshold for empiric treatment. In a city with low prevalence of TB, even a positive score would not significantly affect the likelihood that the patient has TB, and withholding treatment pending final culture results for these patients may be prudent. Discussion Control Patients Total * 1 point positive TST result; 1 point expectoration or infiltrate observed on chest radiograph not typical of TB; 2 points both HIV-positive status and mediastinal lymphadenopathy observed on chest radiograph. Despite the initial clinical suspicion of TB, when a patient s sputum smear results are negative for AFB, the diagnosis of TB may by missed. For those patients with a high clinical suspicion, clinicians must face the dilemma of empirically treating or waiting for up to 8 weeks for the final culture results. Though newer rapid diagnostic tests are available in most large medical centers, they are expensive, have poor sensitivity and specificity for smear-negative sputum samples, and are not yet considered standard of practice Clinicians may use criteria for smear- LR 352 Clinical Investigations

5 Geographic Area Table 4 Posttest Probability of TB Given the Prevalence of Smear-Negative TB in Three Different Settings Using the TPS* Prevalence of Smear- Negative TB, % Hypothetical Score LR Posttest Probability of TB, % High prevalence Moderate-to-low prevalence Very low prevalence *The risk of TB with a given score (posterior probability) was derived by multiplying the patient s risk of TB based on the estimated prevalence of TB (prior probability) with the LR, and converting the result (posterior odds) to posterior probability. positive disease (ie, fever, productive cough, weight loss) to predict risk in the patient with smearnegative disease. But given the smaller mycobacterial burden present with smear-negative disease, these patients may have different clinical and radiographic findings. In our study, we attempted to ascertain the predictors of this narrowly defined population with smear-negative pulmonary TB. We derived a prediction rule based on our findings, but future studies need to validate this rule in different populations of patients. Our clinical findings demonstrate that the presence of expectoration in a patient with negative sputum smear results is not associated with TB, a clear distinction from the smear-positive patient. Another more common pulmonary process, such as bacterial pneumonia or bronchitis, is more likely than TB possibly because the major cough receptors in the airways may not be stimulated by the relatively small amount of bacteria in smear-negative disease. This finding supports the results 14 from Senegal, where patients with smear-negative disease had an absence of cough. Similar to smear-positive TB, we found that a positive TST result was highly predictive of TB in smear-negative patients as well. In support of our findings, a recent case-control study 29 that examined predictors of culture-positive TB (regardless of smear status) found a positive purified protein derivative tuberculin test result to be their strongest predictor of TB (OR, 13.2; 95% CI, 4.4 to 40.7). A likely explanation is that the immunologic response to the TST is determined by a much lower burden of TB organisms than what exists even in smear-negative TB. Our radiographic findings are interesting from a biological perspective, and are different from those classically found in smear-positive TB. The typical radiographic pattern of reactivation TB in smearpositive patients, apical or upper lobe infiltrates or cavities, was not likely in patients with smear-negative disease. Having an infiltrate that is not typical for reactivation TB, such as a lobar consolidation or a diffuse pattern, was a negative predictor in our study. This radiographic finding would make another pulmonary disease process such as a bacterial pneumonia, pulmonary edema, or interstitial disease more likely, and TB less likely. We found a significant interaction between the radiographic finding of mediastinal lymphadenopathy and HIV seropositivity in patients who were culture positive for TB. The association of mediastinal lymphadenopathy in HIV-infected patients with TB is well-known. 30,31 Our findings confirm this association and show a striking association of this interaction with smear-negative TB. The TPS was derived using the -coefficients of the four multivariate predictors in our study to help clinicians apply our findings. Future studies will need to validate our TPS using a prospective cohort of inpatients. Using the prediction score with a simple cutoff of 0 to identify persons with TB, we would have missed 12 of 47 study patients (25%) in this study. Using 1 as the cutoff point for identifying patients with TB would increase our sensitivity to 94%, but lower the specificity. In the case of any infectious disease with public health implications, it is important to have a test or rule that has a high sensitivity and negative predictive value, so that patients with true disease are treated and those with a low possibility of disease can be discharged from the hospital safely without treatment. However, using an arbitrary cut point to identify patients with TB dichotomizes the TPS and wastes valuable information from each possible result. LRs have the advantage of giving more precise information when a test has more than two outcomes. For the TPS, the range of outcomes extended from 2 to 3, with LRs ranging from 0.2 to 7.1. CHEST / 120 / 2/ AUGUST,

6 Using LRs, a clinician has the advantage of calculating a specific patient s probability of having TB based on the estimated prevalence of TB in his or her specific setting and the LR result of a given score. We demonstrated the application of LRs in areas with varying prevalence of TB. Areas of higher TB prevalence should have a more conservative threshold for treating TB, perhaps with a TPS of 0. In contrast, areas with very low prevalence of TB can afford to wait until final culture results are available, since it is likely that a person with negative smear results has another more common disease. It is difficult to know how low the posterior probability of TB needs to be to withhold empiric treatment for TB. Given the limited resources and the potential for drug toxicity, we believe it is reasonable to choose a threshold for empiric treatment at 5%. However, this decision should be made together by the physician and patient on a case-by-case basis. Although we were limited to chart review to assess all of the clinical and demographic variables, most of these were routinely assessed in patients in whom TB was suspected. However, we did encounter missing data on an important predictor of smearnegative TB, the TST. We ran our multivariate analysis using the three different techniques outlined in the Materials and Methods section and found similar results. The proportion of patients who had documented TST results was significantly different among the study patients and control patients (77% vs 48%, respectively), suggesting a differential suspicion of TB between the two groups of patients. Perhaps the control patients were less likely to have a TST placed because the alternative, more common, diagnosis such as a bacterial pneumonia was responding appropriately to therapy. The study patients may not have responded as well to these therapies, causing the practitioners to pursue the diagnosis of TB more aggressively. However, then we would have expected more bronchoscopic evaluations in the study patients, which was not seen. Due to limitations in our retrospective study design, we cannot eliminate the possibility of bias playing a role in which patients had a TST applied. Nevertheless, a positive TST result was an important predictor of TB in smear-negative patients, and was a statistically significant finding in our study despite the missing data; we encourage the practice of placing a TST on every patient who is suspected of having TB. Since we were limited to chart review, we were unable to collect information on other factors that clinicians may use in deciding whether to begin a patient on empiric antituberculous therapy. These unmeasured variables may be more predictive of culture-positive TB. For example, it is difficult to quantify the conviction of an alternate diagnosis or the response to nontuberculous therapies from medical notes alone, but these factors may be strongly associated with TB. We evaluated clinical, demographic, and radiographic characteristics that would be easily available early in the diagnostic workup of TB, but may have missed other more subtle determinants that clinicians use that could better predict TB. The findings of our study can guide clinical practice in settings similar to that of San Francisco General Hospital, a large public hospital in a city that has moderate-to-high levels of TB and HIV prevalence compared to other US cities. It is unclear whether these findings may be generalizable to private hospitals, academic settings with different high-risk groups of patients, and in developing countries. Future studies will need to validate our findings in cohorts of patients in different clinical settings. Our study provides potentially important information to help clinicians refine decision making in the case of patients with negative sputum smear results. Using our simple scoring method, a practitioner could apply easily available clinical, demographic, and radiographic characteristics to assess the patient s risk of pulmonary TB. If the sum of the points of the TPS is 0, the patient has a low probability of having TB, even in areas of higher TB prevalence, and may not need to be started on empiric antituberculous therapy pending final culture results. However, patients with positive scores with our prediction rule in cities that have moderate levels of TB prevalence or greater may benefit by early institution of therapy. ACKNOWLEDGMENT: The authors thank Philip C. Hopewell, MD; Deborah Grady, MD; Warren Browner, MD; Michael Shlipak, MD; Peter M. Small, MD; and Eliseo J. Pérez-Stable, MD, for their assistance in reviewing and revising the article. References 1 Fairchild AL, Oppenheimer GM. Public health nihilism vs pragmatism: history, politics, and the control of tuberculosis. Am J Public Health 1998; 88: Bock NN, McGowan JE Jr, Ahn J, et al. Clinical predictors of tuberculosis as a guide for a respiratory isolation policy. Am J Respir Crit Care Med 1996; 154: Tattevin P, Casalino E, Fleury L, et al. The validity of medical history, classic symptoms, and chest radiographs in predicting pulmonary tuberculosis: derivation of a pulmonary tuberculosis prediction model. Chest 1999; 115: Cohen R, Muzaffar S, Capellan J, et al. The validity of classic symptoms and chest radiographic configuration in predicting pulmonary tuberculosis. Chest 1996; 109: El-Solh A, Mylotte J, Sherif S, et al. Validity of a decision tree for predicting active pulmonary tuberculosis [published erratum appears in Am J Respir Crit Care Med 1997 Dec; 156:2028]. Am J Respir Crit Care Med 1997; 155: Moran GJ, Fuchs MA, Jarvis WR, et al. Tuberculosis infec- 354 Clinical Investigations

7 tion-control practices in United States emergency departments. Ann Emerg Med 1995; 26: Scott B, Schmid M, Nettleman MD. Early identification and isolation of inpatients at high risk for tuberculosis. Arch Intern Med 1994; 154: Murphy GD, Rydman RR, Batesky D, Hill S. Clinical parameters that predict culture-positive pulmonary tuberculosis in the emergency department [abstract]. Ann Emerg Med 1995; 25:137 9 Kanaya AM, Modine GW, Chambers HF. Clinical and radiographic predictors of smear positive pulmonary tuberculosis [abstract]. J Gen Intern Med 1998; 13:16 10 McPhedran FM, Opie EL. The spread of tuberculosis in families. Am J Hygiene 1935; 22: Grzybowski S, Barnett GD, Styblo K. Contacts of cases of active pulmonary tuberculosis. Bull Int Union Tuberc 1975; 50: Behr MA, Warren SA, Salamon H, et al. Transmission of Mycobacterium tuberculosis from patients smear-negative for acid-fast bacilli [published erratum appears in Lancet 1999; 353:1714]. Lancet 1999; 353: Dutt AK, Stead WW. Smear-negative pulmonary tuberculosis [published erratum appears in Semin Respir Infect 1994; 9:261]. Semin Respir Infect 1994; 9: Samb B, Sow PS, Kony S, et al. Risk factors for negative sputum acid-fast bacilli smears in pulmonary tuberculosis: results from Dakar, Senegal, a city with low HIV seroprevalence. Int J Tuberc Lung Dis 1999; 3: Nunn P, Mungai M, Nyamwaya J, et al. The effect of human immunodeficiency virus type-1 on the infectiousness of tuberculosis. Tuber Lung Dis 1994; 75: Colebunders RL, Ryder RW, Nzilambi N, et al. HIV infection in patients with tuberculosis in Kinshasa, Zaire. Am Rev Respir Dis 1989; 139: 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: Karstaedt AS, Jones N, Khoosal M, et al. The bacteriology of pulmonary tuberculosis in a population with high human immunodeficiency virus seroprevalence. Int J Tuberc Lung Dis 1998; 2: Theuer CP, Hopewell PC, Elias D, et al. Human immunodeficiency virus infection in tuberculosis patients. J Infect Dis 1990; 162: Smith RL, Yew K, Berkowitz KA, et al. Factors affecting the yield of acid-fast sputum smears in patients with HIV and tuberculosis. Chest 1994; 106: Bass JB Jr, Farer LS, Hopewell PC, et al. Treatment of tuberculosis and tuberculosis infection in adults and children. American Thoracic Society and the Centers for Disease Control and Prevention. Am J Respir Crit Care Med 1994; 149: Greenland S, Finkle WD. A critical look at methods for handling missing covariates in epidemiologic regression analyses. Am J Epidemiol 1995; 142: Katz M. Multivariable analysis: a practical guide for clinicians. 2nd ed. Cambridge, UK: Cambridge University Press, 1999; Barnes PF. Rapid diagnostic tests for tuberculosis: progress but no gold standard [editorial] [published erratum appears in Am J Respir Crit Care Med 1997; 156(3 pt 1):1022]. Am J Respir Crit Care Med 1997; 155: Rapid diagnostic tests for tuberculosis: what is the appropriate use? American Thoracic Society Workshop. Am J Respir Crit Care Med 1997; 155: Cohen RA, Muzaffar S, Schwartz D, et al. Diagnosis of pulmonary tuberculosis using PCR assays on sputum collected within 24 hours of hospital admission. Am J Respir Crit Care Med 1998; 157: Brugiere O, Vokurka M, Lecossier D, et al. Diagnosis of smear-negative pulmonary tuberculosis using sequence capture polymerase chain reaction. Am J Respir Crit Care Med 1997; 155: Kwiatkowska S, Marczak J, Zieba M, et al. Clinical utility of a commercial ligase chain reaction kit for the diagnosis of smear-negative pulmonary tuberculosis. Int J Tuberc Lung Dis 1999; 3: Wisnivesky JP, Kaplan J, Henschke C, et al. Evaluation of clinical parameters to predict Mycobacterium tuberculosis in inpatients. Arch Intern Med 2000; 160: 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: Modilevsky T, Sattler FR, Barnes PF. Mycobacterial disease in patients with human immunodeficiency virus infection. Arch Intern Med 1989; 149: CHEST / 120 / 2/ AUGUST,