The typical pattern includes radiographs with upper lobe infiltrates
and cavitation (arrowhead). The atypical pattern includes radiographs with
effusions, lower and mid lung zone infiltrates, and adenopathy.
Geng E, Kreiswirth B, Burzynski J, Schluger NW. Clinical and Radiographic Correlates of Primary and Reactivation TuberculosisA Molecular Epidemiology Study. JAMA. 2005;293(22):2740-2745. doi:10.1001/jama.293.22.2740
Author Affiliations: College of Physicians
and Surgeons (Drs Geng and Schluger) and Mailman School of Public Health (Dr
Schluger), Columbia University, New York, NY; Public Health Research Institute,
Newark, NJ (Dr Kreiswirth); and New York City Department of Health Tuberculosis
Control Program, New York (Dr Burzynski).
Context The traditional teaching that pulmonary tuberculosis characterized by
lymphadenopathy, effusions, and lower or mid lung zone infiltrates on chest
radiography represents “primary” disease from recently acquired
infection, whereas upper lobe infiltrates and cavities represent secondary
or reactivation disease acquired in the more distant past, is not based on
well-established clinical evidence. Furthermore, it is not known whether the
atypical radiograph common in human immunodeficiency virus (HIV)–associated
tuberculosis is due to a preponderance of primary progressive disease or altered
Objective To analyze the relationship between recently acquired and remotely acquired
pulmonary tuberculosis, clinical and demographic variables, and radiographic
features by using molecular fingerprinting and conventional epidemiology.
Design, Setting, and Population A retrospective, hospital-based series of 456 patients treated at a
New York City medical center between 1990 and 1999. Eligible patients had
to have had at least 1 positive respiratory culture for Mycobacterium tuberculosis and available radiographic data.
Main Outcome Measures Radiographic appearance as measured by the presence or absence of 6
features: upper lobe infiltrate, cavitary lesion, adenopathy, effusions, lower
or mid lung zone infiltrate, and miliary pattern. Radiographs were considered typical if they had an upper lobe infiltrate or cavity whether or not
other features were present. Atypical radiographs
were those that had adenopathy, effusion, or mid lower lung zone infiltrates
or had none of the above features.
Results Human immunodeficiency virus infection was most commonly associated
with an atypical radiographic appearance on chest radiograph with an odds
ratio of 0.20 (95% confidence interval, 0.13-0.31). Although a clustered fingerprint,
representing recently acquired disease, was associated with typical radiograph
in univariate analysis (odds ratio, 0.68; 95% confidence interval, 0.47-0.99),
the association was lost when adjusted for HIV status.
Conclusions Time from acquisition of infection to development of clinical disease
does not reliably predict the radiographic appearance of tuberculosis. Human
immunodeficiency virus status, a probable surrogate for the integrity of the
host immune response, is the only independent predictor of radiographic appearance.
The altered radiographic appearance of pulmonary tuberculosis in HIV is due
to altered immunity rather than recent acquisition of infection and progression
to active disease.
Traditionally, active tuberculosis (TB) disease has been classified
as either primary or secondary. Many researchers consider primary and secondary
TB to reflect the time between the initial infection with Mycobacterium tuberculosis and the onset of clinical disease. In the
literature, the exact interval that distinguishes primary from secondary TB
ranges from 1 to 5 years.1
Primary and secondary TB are also thought to have characteristic radiographic
and clinical features: primary TB is said to be characterized by lower-lobe
disease, adenopathy, and pleural effusions, and termed atypical, whereas secondary, or reactivation, TB is associated with
upper lobe disease and cavitation, termed typical.2- 5 These
clinical observations, however, were based on studies conducted before the
availability of molecular fingerprinting techniques and relied on often incomplete
and circumstantial data. The Figure shows
an example of the typical and atypical patterns in 2 of our study patients.
Pulmonary TB in the human immunodeficiency virus (HIV)/AIDS population
is often characterized by adenopathy, mid or lower lung zone disease, effusions,
and a paucity of cavitary lesions.6- 8 Because
persons with HIV/AIDS are prone to TB, some have attributed this atypical
radiographic appearance to the susceptibility of HIV-infected patients to
rapid progression from initial infection to active TB disease.9 Others
have raised the question of whether the atypical radiograph results from altered
immunity and may therefore also represent reactivation of long-standing latent
infection in the setting of an abnormal host immune response.10
Molecular epidemiology studies of TB allow comparison of clinical and
radiographic features of TB cases, which are likely to be related in time
and space (ie, clustered isolates) with those that are not (unique isolates).11,12 In the present study, we use this
approach to test whether recently transmitted cases have radiographic features
distinct from distantly acquired infection and secondly, whether the atypical
features of the radiograph in HIV-associated TB are due to a preponderance
of recent infection or are manifestations of altered immunity in the reactivation
of latent infection.
Study participants included all adult patients with culture-proven pulmonary
TB at Columbia University Medical Center between 1990 and 1999 with pulmonary
involvement, defined by at least 1 positive culture from sputum or from pleural
effusion, and who also had available radiographic data. Demographic data were
abstracted from the New York City Tuberculosis Control Program’s registry.
This study was approved by the Institutional Review Board of the College of
Physicians and Surgeons, Columbia University.
Restriction fragment length polymorphism (RFLP) analysis (DNA fingerprinting
with the IS6110 insertion sequence) was performed at the Public Health Research
Institute, Newark, NJ, according to internationally standardized methods.13 In summary, M tuberculosis DNA
was extracted, digested with PvuII, subjected to
electrophoresis, and hybridized by Southern blotting with a fragment of the
insertion element IS6110. Identical strains recovered from 2 or more patients
comprised a cluster of cases. Strains found in only 1 person were considered
unique. M tuberculosis strains with fingerprints
of 5 or fewer bands were subjected to secondary DNA analysis with spoligotyping
(spacer oligonucleotide typing).14- 16 These
low-band RFLP patterns were assigned as clustered only if their spoligotypes
were identical and specific to the IS6110 fingerprint. If they were not, they
were classified as unique cases. Clustered cases were assumed to represent
recent transmission, whereas a unique case was considered to result from the
reactivation of latent disease.17
Data from the chest radiographs were obtained from review of reports
dictated by attending radiologists at the time of admission and recorded in
the electronic patient records. If multiple radiographs were available, the
first radiograph performed on an admission for which a positive sputum culture
was isolated was chosen for analysis. Radiographic findings were recorded
as categorical variables and included the presence or absence of 6 features:
upper lobe infiltrate cavitary lesion, hilar or paratracheal lymphadenopathy,
middle or lower lobe infiltrates, pleural effusion, and miliary pattern.
The radiographs were considered typical if either an upper lobe infiltrate
or a cavitary lesion in the upper lung zones was present. The presence of
lymphadenopathy, a lower or middle lobe infiltrate, or effusion in conjunction
did not change the characterization as typical. On the other hand, atypical
radiographs were those with lymphadenopathy, lower or middle lobe infiltrates,
or effusions without the presence of either a cavity or upper lobe infiltrate.
Radiographs with a cavitary lesion in the middle or lower lung zones were
considered atypical. Available radiographs were reviewed independently by
2 of the authors (J.B. and N.W.S.) and the κ statistic was calculated
between each reader and the abstracted radiographic data from the chart.
We did not have access to information on treatment with antiretroviral
drugs. We conducted a separate and parallel analysis using only data before
1997, ie, before the advent of widespread treatment with highly active antiretroviral
treatment to see if its use coincided with changes in the observed relationships
between clinical, molecular, and demographic predictors and radiographic appearance.
Associations between radiographic and clinical variables were tested
with χ2 statistical tests. Unknown values were excluded from
univariate analysis. Continuous variables were made into categorical variables
for univariate and multivariable analysis. Multivariable analysis was conducted
with both logistic regression modeling and a generalized estimating equation
model to adjust standard error estimates for potential correlation among clustered
observations.18 Human immunodeficiency virus
status was recorded as a 3-leveled categorical predictor with positive, negative,
and unknown. Age was divided between those 60 years or younger and those older
than 60 years. The age of 60 years was chosen as a cutoff because previous
analysis involving this group of patients showed a precipitous drop in proportion
of clustered patients older than 60 years.19 A
significant drop in the risk of clustering occurred among patients in whom
TB was diagnosed after 1993; therefore, the year of diagnosis was analyzed
as a categorical variable (1990-1993 or 1994-1999). All analyses were conducted
using SAS version 9.1 (SAS Institute Inc, Cary, NC). P<.05
was considered statistically significant.
Race and ethnicity classification was based on patient self-report,
as recorded in the New York City Department of Health Tuberculosis Control
Program’s case registry. This variable was included in our analyses
because of prior suggestions that racial groups differ in susceptibility to
TB and could have therefore influenced findings.
There were 546 culture-proven adult cases of TB at Columbia University
Medical Center between 1990 and 1999 with corresponding demographic data available
at the Department of Health, Tuberculosis Control Program. Of these, 484 (89%)
had pulmonary involvement and among these 456 (94%) had radiographic data
available in the computerized charts.
Columbia University Medical Center serves Washington Heights and is
adjacent to Harlem. Washington Heights has served as a destination for foreign-born
persons from the Caribbean for a long time and the average time in the United
States among our foreign-born patients was 14.1 years with a median of 14
years. In Harlem most patients are African American and US born. A total of
35.8% of our study population was foreign-born, and 54.3% with known HIV status
were HIV infected. Most were men (69.5%) and 83.6% were younger than 60 years.
About half the TB isolates from these patients (51.8%) were part of molecular
epidemiologic clusters. The mean age of the population was 41 years with a
median of 38 years. Characteristics of the patient population are shown in Table 1. Because RFLP data were available for
546 patients, clustering data were assigned based on the original number of
patients. There were 51 separate clusters with mean size of 5.2 and a median
Overall 276 (60.5%) had typical radiographs while 180 (39.5%) had atypical
radiographs. Cavitary lesions were present in 28.7%, upper lobe infiltrates
in 58.3%, and lymphadenopathy in 22.6% of the patients. The radiographic characteristics
are shown in Table 2. Association between
radiographic predictors was assessed using a binomial test whereby each paired
observation is assigned a number 1 for concordance and 0 for discordance and
the sum is compared with a normal distribution. Fifty-four physical radiographs
were read by 2 authors (J.B. and N.W.S.). The κ statistic between radiology
reading and N.W.S. was 0.799; between radiology reading and J.B., 0.761; and
between N.W.S. and J.B., 0.683.
Clustering was significantly associated in an inverse manner to the
presence of an upper lobe infiltrate with an odds ratio (OR) of 0.68 and a
95% confidence interval (CI) of 0.47-0.99, as well as with a typical radiograph
(OR, 0.58; 95% CI, 0.39-0.84). Clustering was not significantly related to
any other radiographic feature. There were 26 patients (5.7%) whose radiographs
had only effusions, and the presence of only an effusion was not significantly
associated with a clustered strain (OR, 0.56; 95% CI, 0.25-1.27). Clustered
isolates were not significantly associated with the presence of a cavitary
lesion on the radiograph (OR, 0.85; 95% CI, 0.56-1.27).
When the relationship between cluster status was stratified by HIV status,
several strong and significant associations became evident. In both groups
of patients with clustered and unique strains, HIV infection was associated
with fewer cavitary lesions, fewer upper lobe infiltrates, more lymphadenopathy,
and fewer typical patterned radiographs. No significant interaction between
HIV and cluster status was found: the Breslow-Day tests for heterogeneity
between HIV status and radiographic outcome across strata of cluster status
was not statistically significant in any case. These results are shown in Table 3.
Univariate analysis of characteristics associated with typical radiographs
found the following associations: any resistance (OR, 2.11; 95% CI, 1.00-4.47),
clustered RFLP (OR, 0.68; 95% CI, 0.47-0.99), HIV status (OR, 0.21; 95% CI,
0.13-0.34). Age older than 60 years (OR, 1.64; 95% CI, 0.97-2.78), and non–US
birth (OR, 1.42; 95% CI, 0.96-2.11) were nearly significantly associated with
typical radiographs. Analysis of the relationship between demographic and
social predictors across strata of HIV showed no significant interaction as
tested by the Breslow-Day test for heterogeneity (Table 4).
Generalized estimating equations were used to model the relationship
between significant predictors in univariate analysis with radiographic appearance
in multivariable analysis to provide a more conservative approach to estimates
of the SE in the variable cluster status based on the number of clusters (n=51)
rather than the number of clustered observations. The model found that HIV
was the most significant predictor of radiographic appearance with an OR of
0.20 (95% CI, 0.13-0.31) for association between HIV infection and typical
radiograph. Clustering, age, and foreign birth were also entered into the
model but were not significantly associated with radiographic appearance.
Any resistance remained significant (OR, 3.02; 95% CI, 1.34-6.78). These results
are shown in Table 5.
We repeated the analysis using only the data gathered before 1997, when
the use of highly active antiretroviral treatment came into widespread use,
to see if its use coincided with significant changes in the data. There were
no changes in the outcome, and again HIV status and any resistance were the
only significant predictors of radiographic appearance.
We demonstrate in a large series of epidemiologically and clinically
well-defined patients with TB that the most significant independent predictor
of radiographic appearance is HIV status. Cluster status, which allows us
to distinguish recently acquired from remotely acquired TB, is not a significant
predictor of radiographic appearance.
A recent study also found no difference in radiographic presentations
between primary and reactivation disease.20 Our
study, larger and with more statistical power, confirms and strengthens those
findings. The current study involved 456 patients and was powered at 95% to
detect a difference of 15% between study populations, which makes us reasonably
confident that we would have been able to detect such an association if one
Our findings are in conflict with older literature showing reactivation
of infection acquired long ago to be manifest as upper lobe infiltrates and
cavitary lesions and recently acquired disease to produce lymphadenopathy,
effusions, and lower and middle lung zone infiltrates. These studies, however,
were conducted and published before molecular techniques were available and
were limited by unreliable definitions of recently acquired disease that included
both valid measures, such as documented purified protein derivative conversion
to more suspect historical and even tautological radiographic findings. Choyke
et al2 in 1983 reported the radiographic features
of primary pulmonary TB in a case series, but only 64% had documented purified
protein derivative conversions, whereas the other patients were classified
based on radiographic features, such as the presence of adenopathy or pleural
effusions or clinical criteria. Woodring et al4 found
less cavitation and fewer upper lobe infiltrates among patients with primary
TB, but inclusion criteria did not require culture-proven diagnoses, and nearly
half of the primary cases were children.
The highly significant and strong association between HIV status and
atypical radiographs argue that immune status is the major determinant of
atypical radiographs in HIV patients and imply strongly that time had elapsed
between acquisition of infection and development of active disease is not
a strong association. Outbreak investigations have shown that HIV patients
are susceptible to infection and clinical disease shortly thereafter,21 and a consequent theory has been that the atypical
radiograph of HIV-associated TB is due to preponderance of primary disease.
Our findings argue against this theory. In fact, our results show that HIV
infection predicts atypical radiographs in both primary disease and reactivation,
that is, among clustered and nonclustered cases.
In our study and in a previously reported study, pleural TB was associated
with clustering 35% to 40% of the time.22 We
found, however, that as was true with radiographic presentations overall,
pleural effusions were more likely to reflect underlying HIV infection than
molecular epidemiologic linkages. The absence of an association in our study
between the presence of an effusion and clustering, in contrast to earlier
studies, may perhaps be explained by the fact that 53.4% of patients with
known HIV status were HIV-positive.
The association of drug-resistant isolates with typical radiographic
findings was interesting because it is not immediately apparent why drug susceptibility
or resistance should affect radiographic features of TB. We speculate that
this finding may be explained by considering that the manifestations of active
TB result from a balance between the host immune response and mycobacterial
virulence. Both laboratory and clinical studies have suggested that drug-resistant
organisms are less virulent.23,24 It
seems plausible that either weakened host immune responses or more virulent
pathogens could result in atypical presentations and, conversely, that less
virulent pathogens may predispose to more typical presentations. This is consistent
with our finding that HIV-infected individuals have atypical radiographic
findings and could explain why drug-resistant but less virulent strains appear
There are limitations to our study. Each cluster necessarily includes
a case of reactivation TB, we were unable to determine which member of a cluster
that would be and hence which radiograph would be misclassified.25 This
bias, however, would tend to bias the results toward the null so significant
findings are likely to remain valid. Furthermore, our study population is
derived from a single academic medical center, and hence may be susceptible
to selection bias. Our patients included a high percentage of HIV patients.
Although this may distinguish our patients from those in other studies, we
sought to test hypotheses about the pathogenesis of TB rather than the epidemiology,
and hence the results may be generalizable to different populations. That
nearly half of our patients with known HIV status were HIV-infected maximizes
the power of the comparison between HIV-infected and HIV-uninfected groups.
In the past 10 years, fingerprinting of M tuberculosis has led to numerous important insights into the epidemiology and program
management of TB that have significantly changed the way we understand this
disease.26 This study applies this technique
to advancement of our understanding of fundamental features of M tuberculosis infection and pathogenesis and reiterates the utility
of fingerprinting in TB research.
In summary our findings argue that the terms primary and reactivation
TB are misleading when used to make inferences linking radiographic findings
to epidemiologic characteristics of patients. Radiographic findings have implications
regarding host immune status of patients, but whether a patient’s disease
is due to recently transmitted or remotely acquired infection cannot be determined
Corresponding Author: Neil W. Schluger,
MD, Division of Pulmonary, Allergy, and Critical Care Medicine, Columbia University
Medical Center, 630 W 168th St PH-8 East, New York, NY 10032 (firstname.lastname@example.org).
Author Contributions: Dr Schluger had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Geng, Burzynski,
Acquisition of data: Geng, Kreiswirth, Burzynski.
Analysis and interpretation of data: Geng,
Drafting of the manuscript: Geng, Burzynski,
Critical revision of the manuscript for important
intellectual content: Geng, Kreiswirth, Schluger.
Statistical analysis: Geng, Schluger.
Obtained funding: Schluger.
Administrative, technical, or material support:
Geng, Burzynski, Schluger.
Study supervision: Kreiswirth, Burzynski, Schluger.
Financial Disclosures: None reported.
Funding/Support: Supported in part by grant
K24 HL004074 from the National Heart, Lung, and Blood Institute of the National
Institutes of Health.
Role of the Sponsor: The NIH played no role
in the design and conduct of the study; collection, management, analysis,
or interpretation of the data; or preparation, review, or approval of the