Glucocorticoid use, other associated factors, and the risk of tuberculosis

Authors


Abstract

Objective

To evaluate the association of glucocorticoids and other purported risk factors with the development of tuberculosis.

Methods

We conducted a case-control study of tuberculosis cases identified during 1990–2001 using the General Practice Research Database in the United Kingdom. Cases were patients with a first time diagnosis of tuberculosis accompanied by at least 6 months of treatment with at least 3 different tuberculosis medications. Up to 4 controls were matched to each case on age, sex, the practice attended by the case, index date, and amount of prior computerized records.

Results

The study encompassed 497 new cases of tuberculosis and 1,966 controls derived from 16,629,041 person-years at risk (n = 2,757,084 persons). The adjusted odds ratio (OR) of tuberculosis for current use of a glucocorticoid compared with no use was 4.9 (95% confidence interval [95% CI] 2.9–8.3). The adjusted ORs for use of <15 mg and ≥15 mg of prednisone or its equivalent daily dose were 2.8 (95% CI 1.0–7.9) and 7.7 (95% CI 2.8–21.4), respectively. Adjusted ORs of tuberculosis were 2.8 for patients with a body mass index (BMI) <20 compared with normal BMI; 1.6 for current smokers compared with nonsmokers; and 3.8, 3.2, 2.0, and 1.4 for those with history of diabetes, emphysema, bronchitis, and asthma, respectively, compared with those without such history (all P values <0.05).

Conclusion

These results indicate that patients treated with glucocorticoids have an increased risk of developing tuberculosis, independent of other risk factors. Low adiposity, diabetes, current smoking, and obstructive pulmonary disorders are also important independent risk factors for tuberculosis.

INTRODUCTION

Tuberculosis is the leading cause of death due to an identifiable infectious pathogen worldwide (1). It is estimated that there were 8.7 million new cases of tuberculosis in 2000 (2). Well-accepted risk factors for tuberculosis include human immunodeficiency virus (HIV) infection, low socioeconomic status, birth or travel in developing countries, silicosis, diabetes, gastrectomy, jejunoileal bypass, chronic renal failure, certain malignancies, and organ transplants; these factors have been incorporated in the risk stratification in tuberculosis guidelines (3). There are also other purported risk factors such as smoking, low adiposity, underlying medical disorders such as pulmonary disease or rheumatic disorders, and use of low-dose immunosuppressive agents (e.g., antirheumatic agents) or glucocorticoids (3), but substantive epidemiologic data about these risk factors for tuberculosis are scarce. Comprehensive understanding of these risk factors on this important infectious disease is essential.

Recently, a substantially increased risk of tuberculosis associated with infliximab use (a tumor necrosis factor α [TNFα] inhibitor) was reported (4), especially in populations at high risk for tuberculosis (5, 6). A similar reactivation of tuberculosis may occur in persons receiving glucocorticoids, especially in those receiving substantial doses of glucocorticoids for prolonged periods (3). The joint statement of the American Thoracic Society and the Centers for Disease Control and Prevention (CDC) acknowledges that >15 mg/day of prednisone (or its equivalent) administered for ≥1 month is a risk factor for tuberculosis (3), primarily because this dosage has been shown to suppress tuberculin reactivity (7, 8). However, specific thresholds of dose and duration that could increase the risk for tuberculosis remain unknown (3, 9). Because glucocorticoids are commonly used immunosuppressive agents, a quantitative assessment of their impact on the risk of tuberculosis is important. To address these issues, we evaluated the relationship between various purported factors and risk of incident tuberculosis with a focus on glucocorticoid effects using the UK-based General Practice Research Database (GPRD), a computerized medical record system (10).

PATIENTS AND METHODS

This study was based on information derived from the GPRD. Since 1987, more than 3 million residents in the United Kingdom have been enrolled with selected general practitioners who use office computers provided by In Practice Systems (London, England) and have agreed to provide data to the GPRD for research purposes. The general practitioners received 12 months of instruction on the standardized recording of medical information, and they agreed to supply anonymized information to academic researchers on an ongoing basis. The information recorded includes patient characteristics, drugs dispensed, clinical diagnoses, notation of referrals to consultants, hospitalizations, certain historical information, and other findings (e.g., smoking status, blood pressure, height, and weight). Referral letters from consultants and hospitalizations are kept in a manual file. The general practitioners generate prescriptions directly from the computer, and these are automatically transcribed into the patient's computer record. The details of each prescription, including dose, instructions, and quantity, are automatically recorded on computer and can be used to determine dose and duration of drug exposure. The “Read” code is used to enter medical diagnoses, and a coded drug dictionary called multilex is used for the recording of prescriptions. Large validation studies have determined that information on all patient referrals and hospitalizations present in the manual medical records in the general practitioners' offices was recorded on the computer >90% of the time and that the data are of high quality in research use (10, 11). We conducted a case-control study of patients with tuberculosis in the GPRD.

Case and control selection

We identified all patients in the GPRD who had a first time diagnosis of tuberculosis followed by antituberculosis therapy during the period 1990–2001. All cases and controls were required to have at least 1 year of recorded data in their computer record prior to the index date (i.e., date of diagnosis among cases and the corresponding date among their matched controls). The computer record of each potential case was reviewed by hand to determine if the patient was a true case. Participants were classified as cases if they received at least 3 different antituberculosis medications and if the treatment lasted at least 6 months. Both cases and controls with HIV or malignancy were excluded.

Up to 4 controls, selected from noncases at risk for developing incident tuberculosis in the study base population, were matched to each case on age (within 1 year), sex, the practice attended by the case (a proxy for geography), index date, and amount of prior computerized medical history. The index date of each case was used as the index date for determining exposure and covariate information for each matched control.

We sent a questionnaire to a sample of general practitioners to confirm case status and to obtain information on time the patient spent in tuberculosis-endemic areas. A questionnaire was also sent to a sample of general practitioners to ascertain information on controls' time spent in an area endemic for tuberculosis. We received questionnaires for 14 cases and 34 controls. All cases were confirmed except for 1 (8%), leading us to accept all tuberculosis cases based on the computer record. The cases and controls for whom additional information was sent via questionnaire were all participants from discordant sets in the case-control matching, and therefore were the only sets that contributed to analyses for glucocorticoid exposure. Only 1 case (8%) and its matched 4 controls were from, or had traveled to, a country endemic for tuberculosis.

Assessment of glucocorticoid exposure

Assessment of glucocorticoid exposure was based on the GPRD drug dispensing records. Patients were considered currently exposed to a glucocorticoid if they had received a prescription for any oral glucocorticoid and the supply lasted until within 120 days prior to the index date. Recent exposure was defined as glucocorticoid use that ended 121–180 days prior to the index date. All other study drug use >180 days prior to the index date was considered past use.

Assessment of covariates

Assessment of current exposure to antirheumatic agents or immunosuppressants was based on the GPRD drug dispensing records and included azathioprine, gold, cyclophosphamide, mycophenolate mofetil, sulfasalazine, methotrexate, cyclosporine, penicillamine, etanercept, and infliximab, using the same exposure definition that was used for glucocorticoids. The presence of diabetes mellitus was defined by the presence of the diagnosis with antidiabetic treatment any time prior to the index date. We also evaluated presence of the following diagnoses prior to the index date: pulmonary disorders (emphysema, bronchitis, asthma, and other pulmonary diseases), rheumatic disorders (rheumatoid arthritis, lupus, polymyalgia rheumatica, vasculitis, and other rheumatic diseases), inflammatory bowel diseases (Crohn's disease and ulcerative colitis), dermatitis, silicosis, renal failure, gastrectomy, and jejunoileal bypass surgery. The general practitioners regularly record information on height, weight, and smoking. Body mass index (BMI) was calculated by dividing the weight in kilograms by the square of the height in meters.

Statistical analysis

We used conditional logistic regression analysis to evaluate associations between glucocorticoid use and the risk of tuberculosis (SAS Institute, Cary, NC). We evaluated the impact of current, recent, and past exposure (as defined above) to glucocorticoid compared with no exposure. We further examined the impact of duration and dose of glucocorticoid use including total number of prescriptions, total duration of use, maximum daily dose, most recent daily dose, and cumulative dose among those currently exposed to a glucocorticoid. Dose equivalents of prednisone were calculated using the following potency: 1 mg of prednisone = 0.15 mg of betamethasone = 0.15 mg of dexamethasone = 0.8 mg of triamcinolone = 0.8 mg of methylprednisolone = 1 mg of prednisolone = 4 mg of hydrocortisone = 5 mg of cortisone. Multivariate models were adjusted for BMI (<20, 20–25, >25, and unknown), smoking (current, past, nonsmoker, and unknown), disease-modifying antirheumatic drug use, and history of diabetes and pulmonary disease. Our estimates did not materially change when we further adjusted for other covariates. We explored potential subgroup effects by restricting our analyses to patients without prior pulmonary disorders, those with normal to high BMI, or those who were not current smokers. For all odds ratios (ORs), we calculated 95% confidence intervals (95% CIs). All P values are 2-sided.

RESULTS

The study encompassed 497 new cases of tuberculosis and 1,966 controls identified from a base population of 2,757,084 persons covering 16,629,041 person-years of experience. The incidence rate based on these figures was 3.0 per 100,000 person-years. Patient characteristics are summarized in Table 1. Cases were more likely to be underweight, current smokers, and have a history of diabetes, emphysema, bronchitis, or asthma. The number of patients taking antirheumatic drugs was low, reflecting the prevalence in the general population (Table 1). Only 12 patients (5 cases and 7 controls) were currently exposed to these medications: 6 to sulfasalazine, 5 to azathioprine, 1 to penicillamine (with sulfasalazine), and 1 to cyclophosphamide. Sulfasalazine and azathioprine are the 2 most widely prescribed antirheumatic drugs in the GPRD, so the distribution of use of these drugs in this study population is expected. Of the 2,463 study participants, 17 had rheumatoid arthritis, 1 had lupus, 12 had polymyalgia rheumatica, and 7 had a diagnosis of arteritis. There was only 1 patient with a prior diagnosis of renal failure, but none had silicosis, gastrectomy, or jejunoileal bypass surgery.

Table 1. Characteristics of cases and controls*
VariablesCases (n = 497)Controls (n = 1,966)P
  • *

    Values are the number (percentage) unless otherwise indicated.

  • Includes asthma, bronchitis, emphysema, and other pulmonary diseases.

  • Includes azathioprine, gold, cyclophosphamide, mycophenolate mofetil, sulfasalazine, methotrexate, cyclosporine, and penicillamine.

Sex   
 Male259 (52)1,026 (52)0.98
 Female238 (48)940 (48) 
Age, years   
 0–1949 (10)197 (10) 
 20–39114 (23) 460 (23) 
 40–59136 (27)532 (27) 
 60–79163 (33)650 (33) 
 ≥8035 (7)127 (7)0.99
Smoking status   
 Nonsmoker169 (34)782 (40) 
 Current smoker142 (29)377 (19) 
 Past smoker36 (7)173 (9) 
 Unknown smoker150 (30)634 (32) 
Body mass index (kg/m2)  < 0.001
 <2081 (16)118 (6) 
 20–25125 (25)485 (25) 
 >2570 (14)550 (28) 
 Unknown221 (44)813 (41) 
Diabetes36 (7)56 (3)< 0.001
Pulmonary disease144 (29)305 (16)< 0.001
 Emphysema20 (4)13 (1) 
 Bronchitis77 (15)154 (8) 
 Asthma90 (18)185 (9) 
Rheumatic disease45 (9)172 (9)0.83
Dermatitis19 (4)87 (4)0.55
Inflammatory bowel disease2 (<1)9 (<1)0.87
Antirheumatic or immunosuppressive drug use   
 Current5 (1)7 (<1)0.17
 Recent or past1 (<1)6 (<1) 

The adjusted OR for current use of glucocorticoids compared with nonexposure was 4.9 (95% CI 2.9–8.3) (Table 2). The risk for tuberculosis remained elevated in recent users of glucocorticoids (OR 4.3). When we repeated our analysis, restricting it to those with no history of pulmonary disease, the adjusted OR for current glucocorticoid use compared with nonexposure was 4.2 (95% CI 1.5–11.5). The adjusted OR for current glucocorticoid users with a history of pulmonary disease was 3.8 (95% CI 1.2–12.0). There was no significant interaction between a history of pulmonary disease and recent or current use of glucocorticoids (P = 0.40). The adjusted OR did not change materially in the analyses after eliminating patients with other medical conditions 1 group at a time (i.e., rheumatic disease, dermatitis, inflammatory bowel disease, and diabetes). There were too few patients within each category to allow reliable analyses of independent associations in each subgroup.

Table 2. Relationship between glucocorticoid (GC) and risk of tuberculosis*
ExposureCases (n = 497)Controls (n = 1,966)Crude OR (95% CI)Adjusted OR (95% CI)
  • *

    Values are the number unless otherwise indicated. OR = odds ratio; 95% CI = 95% confidence interval.

  • Adjusted for body mass index, smoking, diabetes, pulmonary diseases, and use of antirheumatic or immunosuppressive agents.

  • Referent.

Nonexposed3951,7981.0 (–)1.0 (–)
Current GC51386.8 (4.3–10.7)4.9 (2.9–8.3)
Recent GC10104.8 (2.0–11.6)4.3 (1.6–11.1)
Past GC411201.7 (1.1–2.5)1.4 (0.9–2.1)

We evaluated the number of prescriptions filled for glucocorticoids. The ORs were elevated at all levels of use among current users of glucocorticoids. The OR for users of only 1 prescription was 3.2 (95% CI 1.4–7.4), which doubled among current users of ≥2 prescriptions (Table 3). We also evaluated the time from first use of a glucocorticoid, among current users, to the index date. The results of this analysis were similar to those of the number of prescriptions analysis. When we assessed the highest daily dosage received among current users, the adjusted ORs were 2.3 (95% CI 0.7–7.5) and 7.0 (95% CI 2.9–16.8) for users of prednisone equivalents compared with nonusers at a daily dosage of <7.5 mg (i.e., physiologic dosage) and ≥7.5 mg (i.e., supraphysiologic dosage), respectively. We also evaluated the impact of daily dosage using a cutoff of 15 mg of prednisone equivalents per day (the cutoff dose adopted by the American Thoracic Society in tuberculosis risk stratification [3]). The adjusted ORs were 2.8 (95% CI 1.0–7.9) for use of <15 mg and 7.7 (95% CI 2.8–21.4) for ≥15 mg. When we examined the most recent daily dose received, the results were similar (Table 4). The adjusted OR for current users with a cumulative dosage <1,000 mg was 4.1 (95% CI 1.8–9.3). For those with cumulative dosages of 1,000–2,999 mg and ≥3,000 mg, the adjusted ORs were 8.3 (95% CI 2.1–33.5) and 3.9 (95% CI 1.5–9.7), respectively. The adjusted ORs for cumulative use in all users of glucocorticoids were 2.0, 5.1, and 4.8 for users of <1,000 mg, 1,000–2,999 mg, and ≥3,000 mg, respectively, suggesting no clear effect of cumulative dose >1,000 mg.

Table 3. Relationship between the number of glucocorticoid (GC) prescriptions and risk of tuberculosis*
ExposureCases (n = 497)Controls (n = 1,966)Crude OR (95% CI)Adjusted OR (95% CI)
  • *

    Values are the number unless otherwise indicated. OR = odds ratio; 95% CI = 95% confidence interval; Rx = number of prescriptions.

  • Adjusted for body mass index, smoking, diabetes, pulmonary diseases, and use of antirheumatic or immunosuppressive agents.

  • Referent.

Nonexposed3951,7981.0 (–)1.0 (–)
1 Rx GC13134.9 (2.5–9.9)3.2 (1.4–7.4)
2–9 Rxs GC18128.4 (3.7–19.1)7.0 (2.8–17.5)
10–19 Rxs GC9510.3 (3.1–34.3)8.7 (2.2–33.9)
≥20 Rxs GC1186.3 (2.5–15.7)4.1 (1.6–11.4)
Recent GC10104.6 (1.9–11.3)4.2 (1.6–10.8)
Past GC411201.7 (1.2–2.5)1.4 (0.9–2.2)
Table 4. Relationship between glucocorticoid dose and risk of tuberculosis*
ExposureCases (n = 497)Controls (n = 1,966)Crude OR (95% CI)Adjusted OR (95% CI)
  • *

    Values are the number unless otherwise indicated. OR = odds ratio; 95% CI = 95% confidence interval.

  • Represents the dose equivalent of prednisone (see text for details).

  • Adjusted for body mass index, smoking, diabetes, pulmonary diseases, and use of antirheumatic or immunosuppressive agents.

  • §

    Referent.

Highest daily dose    
 Nonexposed§3951,7981.01.0
 <7.5 mg774.8 (1.6–13.8)2.3 (0.7–7.5)
 ≥7.5 mg20109.7 (4.4–21.6)7.0 (2.9–16.8)
 Unknown dose24213.6 (0.9–13.4)3.4 (0.7–14.6)
 Recent or past use511301.9 (1.3–2.7)1.5 (1.0–2.2)
Most recent daily dose    
 Nonexposed§3951,7981.01.0
 <7.5 mg875.5 (1.9–16.1)3.1 (0.9–10.3)
 ≥7.5 mg19109.1 (4.1–20.1)6.0 (2.5–14.5)
 Unknown dose42213.0 (0.8–10.7)3.0 (0.7–12.3)
 Recent or past use511301.9 (1.3–2.7)1.5 (1.0–2.2)
Cumulative dose    
 Nonexposed§3951,7981.01.0
 <1,000 mg17126.5 (3.0–13.9)4.1 (1.8–9.3)
 1,000–2,999 mg859.4 (2.7–32.0)8.3 (2.1–33.5)
 ≥3,000 mg14125.9 (2.6–13.2)3.9 (1.5–9.7)
 Recent or past use511301.9 (1.3–2.7)1.5 (1.0–2.3)

The adjusted OR for current smokers as compared with nonsmokers was 1.6 (95% CI 1.4–2.4), whereas the OR for past smokers was not significant (Table 5). Compared with patients with a BMI 20–25 kg/m2, the adjusted ORs of tuberculosis were 2.8 (95% CI 1.9–4.1) for those with a BMI <20 kg/m2 and 0.5 (95% CI 0.4–0.7) for those with a BMI >25 kg/m2 (Table 5). To assess whether the association between low BMI and tuberculosis was an artifact of weight loss due to tuberculosis or whether low BMI is a risk factor for tuberculosis, we evaluated the time lapse from the last recorded weight before the index date to the index date. The mean time from last weight to index date was 2.6 years (median 2.1 years) among patients with a known BMI, and there was little difference between cases and controls with respect to the time interval between last recorded weight and index date. Fewer than 10% of all participants had <6 months between weight recording and index date, and excluding these patients from our analyses resulted in no material difference.

Table 5. Relationship between other risk factors and risk of tuberculosis*
Risk factorCases (n = 497)Controls (n = 1,966)Crude OR (95% CI)Adjusted OR (95% CI)
  • *

    Values are the number (percentage) unless otherwise indicated. OR = odds ratio; 95% CI = 95% confidence interval.

  • Adjusted for glucocorticoid use, smoking, body mass index, diabetes, pulmonary diseases, and use of antirheumatic or immunosuppressive agents.

  • Referent.

  • §

    Includes asthma, bronchitis, emphysema, and other pulmonary diseases.

  • Includes azathioprine, gold, cyclophosphamide, mycophenolate mofetil, sulfasalazine, methotrexate, cyclosporine, and penicillamine.

Smoking status    
 Nonsmoker169 (34)782 (40)1.01.0
 Current smoker142 (29)377 (19)1.8 (1.4–2.4)1.6 (1.4–2.4)
 Past smoker36 (7)173 (9)0.9 (0.6–1.4)0.8 (0.5–1.2)
 Unknown smoker150 (30)634 (32)1.0 (0.7–1.4)0.8 (0.5–1.1)
Body mass index (kg/m2)    
 <2081 (16)118 (6)2.9 (2.1–4.2)2.8 (1.9–4.1)
 20–25125 (25)485 (25)1.01.0
 >2570 (14)550 (28)0.5 (0.3–1.5)0.5 (0.4–0.7)
 Unknown221 (44)813 (41)1.1 (0.9–1.5)1.4 (1.0–2.0)
Diabetes36 (7)56 (3)2.7 (1.7–4.1)3.8 (2.3–6.1)
Pulmonary disease§144 (29)305 (16)2.4 (1.9–3.1)1.8 (1.4–2.4)
 Emphysema20 (4)13 (1)6.8 (3.2–14.2)3.2 (1.3–7.6)
 Bronchitis77 (15)154 (8)2.5 (1.8–3.5)2.0 (1.4–2.9)
 Asthma90 (18)185 (9)2.1 (1.7–2.9)1.4 (1.0–2.0)
Antirheumatic or immunosuppressive drug use77 (15)154 (8)2.5 (1.8–3.5)2.0 (1.4–2.9)
 Current5 (1)7 (<1)2.7 (0.9–9.0)1.6 (0.4–5.8)
 Recent or past1 (<1)6 (<1)0.7 (0.1–5.5)0.6 (0.1–5.0)

Prior pulmonary diagnoses of emphysema, bronchitis, and asthma were all associated with an increased risk of tuberculosis, independent of other risk factors including glucocorticoid use (Table 5). To evaluate whether the association between bronchitis and tuberculosis was an artifact of early symptoms of tuberculosis rather than a risk factor, we looked at the time interval from the diagnosis of bronchitis to the index date in cases compared with controls among all 238 patients with a diagnosis of bronchitis. The mean time from bronchitis diagnosis to the index date was 4.4 years for cases (median 3.9 years) and 4.8 years for controls (median 3.9 years), revealing no material difference between cases and controls.

The crude OR for current use of antirheumatic or immunosuppressive agents compared with nonexposure was 2.7 (95% CI 0.9–9.0), but the OR was substantially attenuated after adjusting for glucocorticoid use (OR 1.2, 95% CI 0.3–4.1) or after further adjusting for other covariates in the final model (OR 1.6, 95% CI 0.4–5.8) (Table 5).

The majority of cases had pulmonary tuberculosis; therefore, the results of the analysis restricted to pulmonary tuberculosis were similar to those for all tuberculosis cases. Our results also remain similar in subgroup analyses restricted to patients with normal to high BMI or those who were nonsmokers or exsmokers.

DISCUSSION

In this case-control study based on a large general practice population, we found that patients who were currently exposed to a glucocorticoid had an approximately 5-fold increased risk for developing new tuberculosis. The magnitude of association was larger with a prednisone equivalent dosage than with a physiologic dosage (i.e., 7.5 mg daily), and was larger with >1 prescription for a glucocorticoid. Although a trend of increased risk (>2 fold) was observed even within the physiologic dose, the association did not reach statistical significance with relatively small numbers. However, the highest daily dose below the cutoff acknowledged by the American Thoracic Society (i.e., 15 mg/day of prednisone or its equivalent) was associated with an approximately 3-fold increased risk of tuberculosis. These associations were independent of other risk factors such as lower BMI, current smoking, diabetes, and pulmonary disorders and also persisted in subgroup analyses restricted to patients with normal to high BMI, those who were nonsmokers or exsmokers, and those without prior diabetes or pulmonary disorders. Although we cannot entirely rule out the potential contribution of underlying disease severity to these associations, our results suggest that glucocorticoid use is associated with a substantially increased risk of developing tuberculosis and that the risk increases with increasing daily dose.

There was no clear effect of duration or cumulative dose on the risk for tuberculosis. The risks increased from the lowest to the middle levels of use, but were lower again at the highest level. One possible explanation for this finding is that persons who are destined to get tuberculosis, for example, those with subclinical tuberculosis, progress to active disease relatively early in the course of treatment. This may explain a higher risk early in the treatment and a lower risk with extended therapy. Furthermore, patients who tolerate extended glucocorticoid therapy may also represent a healthier group than those with a shorter glucocorticoid duration. Nonetheless, there are several aspects of our results that meet the causal criteria proposed by Bradford Hill (12): the effect was present for current users of the drugs but not for past users, the effect was strong (nearly 5 fold), the risk increased with increased daily dose, and the biologic plausibility was clearly present.

The annual incidence in our study population was similar to that in the British white population (4.4 in 100,000 in 1998) (13) or the most recent US estimate by the CDC (5.1 per 100,000 persons in 2002) (14). The GPRD does not include incarcerated or institutionalized patients, and our study excluded those with malignancy and HIV because our focus was on glucocorticoid use and the less established associated risk factors. Therefore, our annual incidence rate should be an underestimate of the total general population. Furthermore, our results are more directly generalizable to populations with low prevalence of tuberculosis (e.g., the US or most European countries) than to endemic populations. These background rates of tuberculosis are important to consider because the risk associated with glucocorticoids likely depends on the background prevalence of latent tuberculosis, as is the case with immunosuppressive tuberculosis risk factors such as HIV infection or treatment with infliximab (4, 6, 15). It is conceivable that the substantial magnitude of association with glucocorticoid use observed in this low-prevalence population may be considerably magnified in higher prevalence populations.

There are many mechanisms by which glucocorticoids can increase the risk of tuberculosis. Systemic glucocorticoids have profound effects on the cellular immune response that tuberculosis requires for its control. Glucocorticoids inhibit the lymphokine effect and monocyte chemotaxis and also block Fc receptor binding and function (16–18). Glucocorticoids depress the number of peripheral blood monocytes as well as monocyte functions, including bactericidal activity and production of interleukin-1 and TNF-α (19). Glucocorticoids also inhibit T cell activation, leading to reduced proliferative responses and cytokine production, and they also induce a redistribution of lymphocytes (predominantly T cells) out of the circulation, leading to peripheral lymphocytopenia (20). These various effects of glucocorticoids on the cellular immune system may play a significant role in predisposing to tuberculosis infection.

We found that a low BMI was associated with a 3-fold increased risk of tuberculosis when compared with patients with a normal BMI, and an increased BMI was associated with a 50% decreased risk of tuberculosis. These results closely agree with previous findings based on 23,541 US Naval recruits with tuberculin reactions ≥10 mm (3, 21). That study demonstrated that recruits who were ≥15% underweight from the standard weight for their height had a risk of progression to disease that was 2 times that of persons who were within 5% of the standard weight for their height and more than 3 times that of persons who were overweight (3, 21). These data suggest a potential hazardous role of low adiposity and protective role of high adiposity against the risk of tuberculosis. However, we cannot rule out a potential reverse causation possibility (i.e., latent tuberculosis infection causing weight loss) given the indolent nature of the infection. Nonetheless, the knowledge that BMI is associated with the risk of tuberculosis (with a mean interval of 2.6 years between last recorded weight and index date) is valuable in risk stratification of developing tuberculosis.

A similar conclusion could be drawn for the substantial association between the prior diagnosis of obstructive pulmonary disorders (i.e., emphysema, bronchitis, and asthma) and the risk of tuberculosis, independent of glucocorticoid use. To our knowledge, this association has not been well appreciated in the literature, and therefore has not been included in stratifying tuberculosis risk in guidelines (3). A previous US hospital-based study found no tuberculosis cases during a 620 person-year followup of 136 corticosteroid-treated patients with asthma (7). If confirmed by future studies, these clinical conditions would be valuable to note in determining the pretest likelihood of tuberculosis in diagnosis and management.

Furthermore, we found that current smoking was associated with a 60% increased risk of tuberculosis independent of other factors. Although this effect is relatively low, because smoking is prevalent in this study population 17% of all cases are attributable to smoking compared with only 8% of cases attributable to glucocorticoid use in this population. These findings suggest that smoking can also be an important modifiable risk factor for tuberculosis in developed countries. Recently, a substantially larger association between smoking and tuberculosis was documented in a country with high tuberculosis prevalence (22). This large Indian study showed that individuals who had ever smoked are 3 times as likely as individuals who had never smoked to report a history of tuberculosis, and tuberculosis deaths among those who had ever smoked were 4 times higher than among nonsmokers (22). These data suggest that the impact of smoking on the potential conversion of latent infection into active disease can be substantially higher in the countries with high tuberculosis prevalence than in those with low prevalence. Iron loading in the bronchoalveolar macrophages, secondary to tobacco smoking, has been implicated in promoting the growth of Mycobacterium tuberculosis, thus potentially contributing to the disease manifestation and progress (23).

Several strengths and potential limitations of our study deserve comment. The data used in this study were from computerized medical records that were prospectively recorded; therefore, there is no concern about biased recall of exposure. All cases were required to have a diagnosis of tuberculosis with a full treatment regimen, ensuring that few noncases were included in the study. Although this approach is desirable for risk factor analyses, which was our main focus in this study, it may result in underestimation of tuberculosis incidence as we discussed above. Despite the large number of tuberculosis cases in this study, the prevalence of antirheumatic agent use was low (reflecting that of the general population), and therefore the independent effects for patients taking antirheumatic agents could not be reliably evaluated. There was also concern that there might be confounding by country of birth or travel to a tuberculosis-endemic country. Because <10% of patients were born in or had recently traveled to such a country according to our questionnaire survey, it is unlikely that this factor confounded our results.

In conclusion, our study results indicate that patients who are treated with glucocorticoids have a substantially increased risk of developing tuberculosis, independent of other risk factors. These data further support the importance of obtaining a thorough medical history regarding tuberculosis and tuberculin testing before initiation of glucocorticoid therapy. Because rheumatoid arthritis is associated with some of the risk factors for tuberculosis (e.g., smoking and pulmonary comorbidity), the precautionary measures for tuberculosis may be even more important. Low adiposity, diabetes, current smoking, and obstructive pulmonary disorders are all important independent risk factors for tuberculosis, and high adiposity is inversely associated with the risk of tuberculosis. These factors should be considered in accurately estimating the tuberculosis risk posed by other factors and may add to the risk assessment of tuberculosis.

Ancillary