0
Clinical Investigations: TUBERCULOSIS |

Predicting Active Pulmonary Tuberculosis Using an Artificial Neural Network* FREE TO VIEW

Ali A. El-Solh, MD; Chiu-Bin Hsiao, MD; Susan Goodnough, RN; Joseph Serghani, MD; Brydon J. B. Grant, MD, FCCP
Author and Funding Information

*From the Department of Medicine (Drs. El-Solh and Grant), Division of Pulmonary and Critical Care Medicine, the Division of Infectious Disease (Dr. Hsiao and Ms. Goodnough), and the Department of Radiology (Dr. Serghani), Erie County Medical Center, and the Veterans Affairs Medical Center, State University of New York at Buffalo, School of Medicine and Biomedical Sciences, Buffalo, NY.

Correspondence to: Ali El-Solh, MD, Division of Pulmonary and Critical Care Medicine, Erie County Medical Center, 462 Grider St, Buffalo, NY 14215; e-mail: solh@buffalo.edu



Chest. 1999;116(4):968-973. doi:10.1378/chest.116.4.968
Text Size: A A A
Published online

Background: Nosocomial outbreaks of tuberculosis (TB) have been attributed to unrecognized pulmonary TB. Accurate assessment in identifying index cases of active TB is essential in preventing transmission of the disease.

Objectives: To develop an artificial neural network using clinical and radiographic information to predict active pulmonary TB at the time of presentation at a health-care facility that is superior to physicians’ opinion.

Design: Nonconcurrent prospective study.

Setting: University-affiliated hospital.

Participants: A derivation group of 563 isolation episodes and a validation group of 119 isolation episodes.

Interventions: A general regression neural network (GRNN) was used to develop the predictive model.

Measurements: Predictive accuracy of the neural network compared with clinicians’ assessment.

Results: Predictive accuracy was assessed by the c-index, which is equivalent to the area under the receiver operating characteristic curve. The GRNN significantly outperformed the physicians’ prediction, with calculated c-indices (± SEM) of 0.947 ± 0.028 and 0.61 ± 0.045, respectively (p < 0.001). When the GRNN was applied to the validation group, the corresponding c-indices were 0.923 ± 0.056 and 0.716 ± 0.095, respectively.

Conclusion: An artificial neural network can identify patients with active pulmonary TB more accurately than physicians’ clinical assessment.

Figures in this Article

The most important aspect of a tuberculosis (TB) infection control program is to identify patients who may have contagious active TB, to isolate them while they are contagious, and to treat them effectively. The process of recognizing those persons with active TB is, however, fraught with difficulty. As a result, numerous outbreaks of Mycobacterium tuberculosis have been reported in several types of facilities. At least 21 episodes of nosocomial transmission of M tuberculosis have been documented in the United States medical literature.1Among the factors that have been associated with missed or delayed diagnosis are failure to consider the diagnosis, nonclassical or atypical radiographic presentation, delayed recognition of drug resistance, lapses in TB isolation practices, and lack of adequate respiratory protection.25

Prediction models to identify patients with active TB have been lacking. The reason for this lies in the complexity of the clinical and radiographic presentation, the relatively small patient samples, and the use of modeling techniques that are poorly suited for the task. Recently, El-Solh et al6 introduced a classification tree to assist physicians in their decision regarding whether respiratory isolation for suspicion of active pulmonary TB is needed. The model achieved a high degree of sensitivity at the expense of low specificity.

Previous investigators have used an artificial intelligence paradigm, referred to as a neural network, to provide a prediction outcome for complex clinical problems. Neural networks are computation systems that process information in parallel, using large numbers of simple units, and that excel in tasks involving pattern recognition. These intrinsic properties of the neural networks have been translated into a higher performance accuracy in outcome prediction compared with expert opinion or conventional statistical methods.78 Therefore, we hypothesized that the ability to identify patients correctly with active pulmonary TB could be improved by using computer analyses involving neural networks. To test this hypothesis, we have applied an artificial neural network (available at: http://bgrant.med.buffalo.edu/activetb/) to the analysis of data from patients who are considered to be at high risk for active pulmonary TB and compared the network output to physicians’ prediction.

Study Setting

The study was conducted at the Erie County Medical Center, a 479-bed tertiary-care teaching facility affiliated with the State University of New York at Buffalo. During the study period from August 1992 to June 1997, the hospital was the major referral center for TB in Erie County and provided all inpatient medical care for inmates from county and state correctional facilities in the area. Because of the shortcomings in the diagnosis of TB and resulting delay in considering the diagnosis, an automatic isolation policy was instituted by the Infection Control Service in August of 1992 for all patients from whom an acid-fast smear and culture were requested. Isolation was discontinued only after documentation of three negative results of acid-fast bacilli smears that were obtained on 3 separate days or a negative result of an acid-fast bacilli smear derived from BAL.

Study Population

Between August 1992 and June 1997, 704 patients were isolated for suspicion of active pulmonary TB. Twenty-two patients were excluded from the study: 17 were discharged before three respiratory specimens were collected, and 5 refused diagnostic bronchoscopy. Five hundred sixty-three consecutive patients were used to design the neural network and were referred to as the derivation set. The remaining 119 patients formed the validation set.

The decision to isolate patients for suspicion of active TB was made by emergency department physicians, medical residents’ or infectious disease fellows after consultation with the attending physician based on symptoms, history of TB exposure, HIV status, positive results of tuberculin skin tests, and radiographic findings. Information regarding demographics (age, gender, date, and duration of isolation), social status (risk factors for HIV purified protein derivative status), and clinical symptoms (fever, night sweats, chest pain, and productive cough for > 2 weeks) was collected from each patient at the time of presentation at the health-care facility. Weight loss was defined as a fall of > 10% of ideal body weight within the previous 6 months. The physicians’ prediction regarding whether the patient had active pulmonary TB was also recorded. Data concerning the results of acid-fast bacilli smears and cultures were recorded once the data were available. For those patients who are known to be HIV seropositive, the cluster of differentiation 4 (CD4) counts were entered into the database only if they were obtained within the previous 3 months of patient isolation. HIV-seronegative patients were presumed to have CD4 counts > 200 cells/μL.

Radiographic Analysis

Chest roentgenograms were divided into two zones: the upper zones delineated by the area above the right and the left fifth ribs posteriorly, and the lower zones below the right and left fifth ribs posteriorly. Upper zone disease was defined as absent only if there were no radiographic abnormalities involving the area above the fifth rib posteriorly. The pattern and distribution of the parenchymal infiltrates (interstitial, nodular, or miliary) or cavities were recorded. The presence and location of adenopathy and pleural effusion were also noted. Interpretation of the chest radiographs was performed by a pulmonologist and a radiologist who were blinded to the microbiology results of sputum stains or cultures.

Bacteriology

The auramine-rhodamine9 fluorescent stain was used to detect acid-fast organisms on respiratory specimens. Radiometric broth medium (BACTEC; Becton Dickinson Diagnostic Instruments Systems; Sparks, MD) was used for inoculation of acid-fast bacilli cultures. M tuberculosis isolates were confirmed with nucleic acid probes (Gene-Probe; San Diego, CA).

Development of the Artificial Neural Network

A general regression neural network (GRNN) was used in the development of the predictive model.10 The advantage of the GRNN lies in the fact that whereas conventional nonlinear regression techniques involve a priori specification of the structure of the regression equations to yield a best fit for the data presented, the GRNN circumvents these restrictions by adjusting the surface dimension in which the regression surface resides without constraining it to a specific form. Generalization is optimized by modifying the smoothing factor, δ, which determines how tightly the network matches its predictions to the data in the training patterns.

The structure of the GRNN used in this model consists of three layers: an input layer, a hidden layer, and an output layer. Input parameters were chosen based on data collected in a previous study.6 The input patterns are formed by 21 distinct parameters (Table 1 ). These parameters are divided into three groups: demographic variables, constitutional symptoms, and radiographic findings. Intervening layers of processors, called hidden units, detect higher-order features in the input layer, analyze the signal, and relay the output to other neurons to make a correct response. The number of neurons in the hidden layer is determined by the number of patterns in the training set as GRNNs require one neuron per pattern processed. The output of the GRNN provides an estimate of the likelihood of active pulmonary TB.

A 10-fold cross-validation approach was used for evaluation.11 The entire data set of the derivation group was divided with a random number generator into 10 subsets. Nine of the 10 subsets were pooled and used for training. The data from the 10th subset were used as an evaluation set during training. The entire process was repeated nine additional times by rotating the subset that was used as the evaluation set during training. The mean square error was computed for each of the 10 neural networks on the entire derivation data set. The mean square errors were averaged, and the artificial neural network that had a mean square error closest to the average was selected.

To normalize the inputs, all independent variables were scaled to a value over a range between 0 and 1. Missing values were substituted with the class mean.

Performance Evaluation:

The predictive model derived from the artificial neural network was tested on an entirely different set of patients (validation cohort) who were not included in the derivation set. The validation cohort comprised all patients who were isolated between January 1996 and June 1997.

Predictive Properties of the Artificial Neural Network:

A receiver operating characteristic curve12was generated for the artificial neural network. The receiver operating characteristic curve represents a graphic display of the true-positives (sensitivity) plotted against the false-positives (1-specificity) for various thresholds that are used to define active pulmonary TB. The c-index was used to estimate diagnostic accuracy by a method described in detail elsewhere.13 The c-index is equivalent to the area under the receiver operating characteristic curve. In brief, it is calculated by determining the probability of diagnosing active TB correctly in every possible pair of patients: one who has active TB, the other who does not. A bootstrap method was used to calculate directly this measure of accuracy by generating 1,000 data sets from our database by random sampling with replacement. Comparisons between the c-indices were assessed based on the confidence intervals (CIs). Commercially available software was used for designing the artificial neural networks (Neuroshell 2; Ward Systems; Frederick, MD) and for CI analyses (CIA; British Medical Journal; London, UK). Statistical significance was accepted at the 5% level.

The characteristics of the population under study are shown in Table 2 . A total of 10 neural networks were trained. The models were designed to produce output values ranging from 0 (no active pulmonary TB) to 1 (active pulmonary TB). The average mean squared error for all 10 neural networks was 0.009. The neural network with the closest mean square error to the average was used for further analysis. The chosen network achieved a sensitivity of 100% (95% CI, 91 to 100%) and a specificity of 72% (95% CI, 65 to 77%). The physicians correctly diagnosed active pulmonary TB in 22 of 47 patients for a sensitivity of 47% (95% CI, 32 to 62%) and a specificity of 75% (95% CI, 71 to 79%). The corresponding c-indices (± SEM) for the artificial neural network and the physicians were 0.947 ± 0.028 and 0.61 ± 0.045, respectively (p < 0.001).

The performance of the neural network was tested prospectively on a new set of 119 patients isolated for suspicion of active TB. The network identified all 11 patients with active pulmonary TB for a sensitivity of 100% (95% CI, 72 to 100%) and a specificity of 69% (95% CI, 61 to 78%). In comparison, the physicians correctly diagnosed active pulmonary TB in 7 of 11 patients, yielding a sensitivity of 64% (95% CI, 31 to 89%) and a specificity of 79% (95% CI, 72 to 87%). Table 3 depicts a comparison of the diagnostic performance of the neural networks and that of the physicians. The diagnostic accuracy of the model, when applied to the validation set as reflected by the c-index, was 0.923 ± 0.056 compared with 0.716 ± 0.09 for the physicians’ prediction (p < 0.05; Fig 1 ).

This study is, to our knowledge, the first to use a neural network for the diagnosis of active pulmonary TB. The recommendation issued by the Centers for Disease Control and Prevention to control the spread of TB calls for direct isolation of any patient suspected of having or known to have infectious TB.14 Standard criteria for early identification of patients with infectious TB have not been well established. The task was rendered difficult by the HIV epidemic in the late 1980s, which has created a new profile for patients with active TB that has none of the typical features recognized in classic cases of active pulmonary TB.4

Predictive models have not fared much better, and their lack of sensitivity is evidence of the complexity of the problem. A review of the literature revealed only a handful of studies that have attempted to tackle this problem. In a study assessing the usefulness of routine admission chest radiography for the detection of pulmonary TB, the authors concluded that chest roentgenograms are still useful in suggesting the diagnosis, particularly in geographic areas with high prevalence for TB.15Nonetheless, failure to suspect TB occurred in 64 of 177 cases of culture-proven TB. Seventeen cases had atypical presentation, and in 29 patients, TB was not diagnosed because of the failure to consider TB despite the presence of upper lobe disease or miliary pattern. In a similar study, Cohen et al16evaluated the clinical symptoms and radiographic configuration in 101 patients who were isolated for suspicion of active TB. The absence of a typical chest radiograph along with the presence of cough, sputum production, and weight loss for < 2 weeks were strong negative predictors of active TB. The authors acknowledged, however, that the population under study was relatively small in number and did not include HIV patients with normal radiographic presentation, as has been described in 6 to 18% of HIV-infected patients with pulmonary TB.1718 Recently, El-Solh et al6 developed a classification and regression tree to predict active pulmonary TB at the time of admission to a health-care facility. The predictor variables were upper zone disease on chest roentgenogram, fever, weight loss, and CD4 count. The tree was validated in a separate cohort of patients yielding a sensitivity and a specificity of 100% and 48.1%, respectively. The high precision achieved in that population was supposedly less than perfect when tested in a different setting.19

The advantage of the neural networks lies in their ability to process nonlinear relationships. Because of the clinical complexity and pathologic heterogeneity of TB, correct identification of patients with active disease is unlikely to depend on the presence or absence of a single defining feature. Hence, it is not surprising that standard linear statistical methodologies are relatively inadequate solutions for this type of problem. In addition, previous studies have shown that clinicians are not aware of the complex interaction among variables that a neural network can exploit. Two separate studies have compared the accuracy of neural networks with that of clinicians to predict disease or outcome.7,20 In the first study, emergency department physicians and medical residents were asked to identify myocardial infarction in patients presenting at an emergency department based on clinical and ECG findings. Eight of 36 cases of myocardial infarction were missed by physicians, compared with only 1 case missed by using the neural network, yielding sensitivities of 77.7% (95% CI, 77 to 82.9%) and 97.2% (95% CI, 97.2 to 97.5%), respectively. In another scenario, the overall accuracy of physicians to predict outcome for colorectal cancer ranged from 75 (95% CI, 66 to 84%) to 79% (95% CI, 71 to 87%), compared with 90% (95% CI, 84 to 96%) for the neural network. The superior prediction capability of neural networks over physician assessment was observed also in this study, which implies that the complexity of biological systems may be beyond the analytic capabilities of physicians.

An essential component of the present study is the ability of the neural network to generalize to new population samples. This feature is, however, affected by many factors, such as the number of neurons in the hidden units, the type of connections in the network, and the extent to which the network has been trained. The results obtained from the validation set indicate that the network described herein may generalize well to new patient data.

Another advantage of the neural network is its ability to handle missing values. In logistic regression, missing values are usually omitted from further analysis.6,21 In our study, the highest percentage of missing data occurred in recalling the result of the purified protein derivative skin test (23%) and CD4 counts (6%). The neural networks incorporated these cases after substituting the missing value with the class mean.

There are several potential limitations to the study. Neural networks have the ability to approximate predictive output to any desirable degree of accuracy when provided with enough running time. This could result in overfitting, particularly when there is an attempt to increase the processing power of the network by adding a large number of hidden neurons. In this case, the network will end up learning not only the training set but also the noise in the data, which leads to poor generalization. It is encouraging that the accuracy of prediction observed in the validation set points to the fact that the network architecture is based on robust features rather than memorizing the idiosyncrasies embedded in the data set.

It is important to emphasize that these results may not be applicable to populations in locations where the epidemiology of TB differs substantially from the area where the study was conducted. Until the model is tested on a different population set, the study can be viewed only as the first attempt in the use of connectionist models in the diagnosis of pulmonary TB. In addition, only the diagnosis of active pulmonary TB was studied. Application of the model to extrapulmonary or extrathoracic TB is not recommended. A good case could be made for the extension of this technology for other aspects of TB should this technique prove to be accurate and reproducible, as the data imply.

Our study has several implications regarding the clinical application of artificial neural networks as a diagnostic tool for active TB. The use of the neural network could provide physicians and health-care workers with a simple and fast tool with which to assess the risk of active TB in any patient presenting at a health-care facility. The estimated probability would enable physicians to initiate isolation without delay, thus reducing the risk of TB exposure to health-care workers.

Abbreviations: CD4 = cluster of differentiation 4; CI = confidence interval; GRNN = general regression neural network; TB = tuberculosis

Table Graphic Jump Location
Table 1. Input Variables Used to Train the Artificial Neural Networks*
* 

PPD = purified protein derivative.

Table Graphic Jump Location
Table 2. Patients Characteristics*
* 

MTB = M tuberculosis; DM = diabetes mellitus. See Table 1 for other abbreviations. Values given as No. (%), unless otherwise indicated.

 

Values given as No. of patients/total No. of patients tested (%).

 

Values given as No. of patients/total No. of patients responding (%).

Table Graphic Jump Location
Table 3. Comparison of the Clinician and Artificial Neural Network Performance on the Validation Group (n = 119)*
* 

ANN = artificial neural network.

Figure Jump LinkFigure 1. Comparison of the receiver operating characteristic curves for the artificial neural network (ANN) and clinicians’ performance as applied to the validation set.Grahic Jump Location
Dooley, SW, Tapper, ML (1997) Epidemiology of nosocomial tuberculosis. Wenzel, RP eds.Prevention and control of nosocomial infections,357-394 Williams & Wilkins. Baltimore, MD:
 
McGowan, JE, Jr Nosocomial tuberculosis: new progress in control and prevention.Clin Infect Dis1995;21,489-505. [PubMed] [CrossRef]
 
Mathur, P, Sacks, L, Auten, G, et al Delayed diagnosis of pulmonary tuberculosis in city hospitals.Arch Intern Med1994;154,306-310. [PubMed]
 
Kramer, F, Modilevsky, T, Waliany, AR, et al Delayed diagnosis of tuberculosis in patients with human immunodeficiency virus infection.Am J Med1990;89,451-456. [PubMed]
 
Flora, GS, Modilevsky, T, Antoniskis, D, et al Undiagnosed tuberculosis in patients with human immunodeficiency virus infection.Chest1990;98,1056-1059. [PubMed]
 
El-Solh, A, Mylotte, J, Sherif, S, et al Validity of a decision tree for predicting active pulmonary tuberculosis.Am J Respir Crit Care Med1997;155,1711-1716. [PubMed]
 
Baxt, WG Use of an artificial neural network for the diagnosis of myocardial infarction.Ann Intern Med1991;115,843-848. [PubMed]
 
Rutledge, R, Fakhry, S The use of neural networking and polynomial regression to predict the outcome in the critically ill: a comparison with APACHE II [abstract]. Crit Care Med. 1993;;21(suppl) ,.:S213
 
Kent, PT, Kubica, GP. Public health mycobacteriology: a guide for the level III laboratory. 1985; Centers for Disease Control. Atlanta, GA:.
 
Specht, D. A general regression neural network. IEEE Trans Neural Networks, 1991; 2:6, 568–576.
 
Weiss, SM, Kulikowski, CA How to estimate the true performance of a learning system.Computer systems that learn1991,17-49 Morgan Kaufmann. Palo Alto, CA:
 
Hanley, JA, McNeil, BJ The meaning and use of the area under the receiver operating characteristic (ROC) curve.Radiology1982;143,29-36. [PubMed]
 
El-Solh, AA, Grant, BJB A comparison of severity of illness scoring systems for critically ill obstetric patients.Chest1996;110,1299-1304. [PubMed]
 
Guidelines for preventing the transmission of.Mycobacterium tuberculosis in health care facilities; 1994. MMWR Morb Mortal Wkly Rep 1994; 43(RR-13):1–132.
 
Barnes, PF, Verdegem, TD, Vachon, LA, et al Chest roentgenogram in pulmonary tuberculosis: new data on an old test.Chest1988;94,316-320. [PubMed]
 
Cohen, R, Muzaffar, S, Capellan, J, et al The validity of classic symptoms and chest radiograhic configuration in predicting pulmonary tuberculosis.Chest1996;109,420-423. [PubMed]
 
Pitchenik, AE, Rubinson, HA The radiographic appearance of tuberculosis in patients with the acquired immunodeficiency syndrome (AIDS) and pre-AIDS.Am Rev Respir Dis1985;131,393-396. [PubMed]
 
Long, R, Maycher, B, Scalcini, M, et al The chest roentgenogram in pulmonary tuberculosis patients seropositive for human immunodeficiency virus type I.Chest1991;99,123-127. [PubMed]
 
Steenstra, R, Brandon, B, Gaeta, T External validation of a decision tree for predicting active pulmonary tuberculosis [abstract]. Am J Respir Crit Care Med. 1998;;157 ,.:A180
 
Bottaci, L, Drew, PJ, Hartley, JE, et al Artificial neural network applied to outcome prediction for colorectal cancer patients in separate institutions.Lancet1997;350,469-472. [PubMed]
 
Doyle, HR, Dvorchik, I, Mitchell, S, et al Predicting outcomes after liver transplantation.Ann Surg1994;219,408-415. [PubMed]
 

Figures

Figure Jump LinkFigure 1. Comparison of the receiver operating characteristic curves for the artificial neural network (ANN) and clinicians’ performance as applied to the validation set.Grahic Jump Location

Tables

Table Graphic Jump Location
Table 1. Input Variables Used to Train the Artificial Neural Networks*
* 

PPD = purified protein derivative.

Table Graphic Jump Location
Table 2. Patients Characteristics*
* 

MTB = M tuberculosis; DM = diabetes mellitus. See Table 1 for other abbreviations. Values given as No. (%), unless otherwise indicated.

 

Values given as No. of patients/total No. of patients tested (%).

 

Values given as No. of patients/total No. of patients responding (%).

Table Graphic Jump Location
Table 3. Comparison of the Clinician and Artificial Neural Network Performance on the Validation Group (n = 119)*
* 

ANN = artificial neural network.

References

Dooley, SW, Tapper, ML (1997) Epidemiology of nosocomial tuberculosis. Wenzel, RP eds.Prevention and control of nosocomial infections,357-394 Williams & Wilkins. Baltimore, MD:
 
McGowan, JE, Jr Nosocomial tuberculosis: new progress in control and prevention.Clin Infect Dis1995;21,489-505. [PubMed] [CrossRef]
 
Mathur, P, Sacks, L, Auten, G, et al Delayed diagnosis of pulmonary tuberculosis in city hospitals.Arch Intern Med1994;154,306-310. [PubMed]
 
Kramer, F, Modilevsky, T, Waliany, AR, et al Delayed diagnosis of tuberculosis in patients with human immunodeficiency virus infection.Am J Med1990;89,451-456. [PubMed]
 
Flora, GS, Modilevsky, T, Antoniskis, D, et al Undiagnosed tuberculosis in patients with human immunodeficiency virus infection.Chest1990;98,1056-1059. [PubMed]
 
El-Solh, A, Mylotte, J, Sherif, S, et al Validity of a decision tree for predicting active pulmonary tuberculosis.Am J Respir Crit Care Med1997;155,1711-1716. [PubMed]
 
Baxt, WG Use of an artificial neural network for the diagnosis of myocardial infarction.Ann Intern Med1991;115,843-848. [PubMed]
 
Rutledge, R, Fakhry, S The use of neural networking and polynomial regression to predict the outcome in the critically ill: a comparison with APACHE II [abstract]. Crit Care Med. 1993;;21(suppl) ,.:S213
 
Kent, PT, Kubica, GP. Public health mycobacteriology: a guide for the level III laboratory. 1985; Centers for Disease Control. Atlanta, GA:.
 
Specht, D. A general regression neural network. IEEE Trans Neural Networks, 1991; 2:6, 568–576.
 
Weiss, SM, Kulikowski, CA How to estimate the true performance of a learning system.Computer systems that learn1991,17-49 Morgan Kaufmann. Palo Alto, CA:
 
Hanley, JA, McNeil, BJ The meaning and use of the area under the receiver operating characteristic (ROC) curve.Radiology1982;143,29-36. [PubMed]
 
El-Solh, AA, Grant, BJB A comparison of severity of illness scoring systems for critically ill obstetric patients.Chest1996;110,1299-1304. [PubMed]
 
Guidelines for preventing the transmission of.Mycobacterium tuberculosis in health care facilities; 1994. MMWR Morb Mortal Wkly Rep 1994; 43(RR-13):1–132.
 
Barnes, PF, Verdegem, TD, Vachon, LA, et al Chest roentgenogram in pulmonary tuberculosis: new data on an old test.Chest1988;94,316-320. [PubMed]
 
Cohen, R, Muzaffar, S, Capellan, J, et al The validity of classic symptoms and chest radiograhic configuration in predicting pulmonary tuberculosis.Chest1996;109,420-423. [PubMed]
 
Pitchenik, AE, Rubinson, HA The radiographic appearance of tuberculosis in patients with the acquired immunodeficiency syndrome (AIDS) and pre-AIDS.Am Rev Respir Dis1985;131,393-396. [PubMed]
 
Long, R, Maycher, B, Scalcini, M, et al The chest roentgenogram in pulmonary tuberculosis patients seropositive for human immunodeficiency virus type I.Chest1991;99,123-127. [PubMed]
 
Steenstra, R, Brandon, B, Gaeta, T External validation of a decision tree for predicting active pulmonary tuberculosis [abstract]. Am J Respir Crit Care Med. 1998;;157 ,.:A180
 
Bottaci, L, Drew, PJ, Hartley, JE, et al Artificial neural network applied to outcome prediction for colorectal cancer patients in separate institutions.Lancet1997;350,469-472. [PubMed]
 
Doyle, HR, Dvorchik, I, Mitchell, S, et al Predicting outcomes after liver transplantation.Ann Surg1994;219,408-415. [PubMed]
 
NOTE:
Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Find Similar Articles
CHEST Journal Articles
PubMed Articles
Guidelines
  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543