Risk of incident active tuberculosis disease in patients treated with non-steroidal anti-inflammatory drugs: a population-based study

We carried out a population-based study using the National Health Insurance Research Database (NHIRD) of Taiwan, done under the approval of the institutional review board of National Taiwan University Hospital. The NHIRD database contains de-identified secondary data, and met the requirements of the “Personal Information Protection Act” in Taiwan. Thus, the data were analyzed anonymously and the need for informed consent was waived. NHIRD records the complete claim history of 1 million randomly selected participants enrolled in Taiwan’s National Health Insurance (NHI), which is a single-payer universal public health insurance [15]. These 1 million participants are believed to be representative of the entire Taiwanese population.

Several studies showed that the NHIRD database is appropriate for use in pharmacoepidemiologic research [12, 16, 17]. The NHIRD database contains participant demographics and detail claim history, which includes medications prescribed, medications quantity, route of administration, individual diagnoses, operations, and outpatient and inpatient electronic claim records.

We used a study cohort that consists of a longitudinally followed Taiwanese’ population from January 1999 to December 2011. A new-user cohort design was adopted by excluding existing users of NSAID and prevalent cases of TB in the year 1999 (pre-enrollment period) [18]. Patients entered the cohort on the first day of the year 2000 and were followed up until the first occurrence of any of these events: diagnosis of active TB, termination of health insurance coverage, death, or end of the study period. People who were less than 18 years old on January 1, 2000, were not eligible for entering the cohort until they became 18 years old in later years. This cohort design thus allows a clean selection of case and controls, in which all patients investigated are new users of NSAIDs and have new cases of active TB.

Patients with new onset active TB disease were identified using the following criteria: at least one outpatient visit or one hospital admission with ICD-9-CM codes of TB (010-018, including all subcategories), plus the prescription of more than two anti-tuberculosis medications for more than 28 days. We excluded patients with a subsequent diagnosis of nontuberculosis mycobacterial infection or lung cancer. This is because the diagnosis of nontuberculosis lung cancer is clinically difficult, and lung cancer has been associated with an increase in the risk of TB. Our case definition had been used in previous studies and had been validated in a linked survey database [19, 20]. One hundred controls were selected to match cases on index date, 5-year age group, and gender using risk set sampling scheme. Index date referred to the first date of TB diagnosis. The one-year period preceding index date was used for assessment of NSAID exposure status.

NSAIDs exposures were defined as having a prescription record of NSAID ≧ 7 days. By restricting to NSAID users having a drug prescription record ≧ 7 days, we found that most of the NSAID users were associated with codes for arthritis and other rheumatic conditions. Few patients were associated with codes for fever. As NSAID may be used to treat the early symptoms of active TB, we defined use of NSAID as a prescription that ended between 31 and 90 days prior to TB diagnosis to avoid the possible reverse causation between use of NSAID and active TB. Traditional NSAIDs are defined as drugs with any of the following compounds: ibuprofen, naproxen, ketoprofen, dexibuprofen, piroxicam, tolfenamic acid, diclofenac, etodolac, nabumetone, and meloxicam. However, aspirin is not included as a drug in the traditional NSAIDs. This is because aspirin has a unique mechanism of action and are usually used on cardiovascular patients. Specific COX-2 inhibitors (COXIBs) are defined as drugs with any of the following compound: celeCoxib, rofeCoxib, valdeCoxib, pareCoxib and etoriCoxib.

In order to be as comprehensive as possible in adjusting for factors that might affect the outcome, we looked at published risk factors for TB and identified 71 covariates (Table 1) [1, 19, 20]. All covariates were assessed before NSAID exposure, except for the one-year period used for exposure assessment. Using ICD9-CM codes, the following covariates were assessed: age, sex, burden of comorbid conditions, indicators for frailty, risk factors for TB disease and use of specific medications. Risk factors for TB included alcoholism-related disease (291, 303, 305.0, 357, 425, 535.3, 571.0-571.3, 577.1), diabetes mellitus (DM) (250), chronic renal failure or hemodialysis (585, V56.0), solid organ transplantation (V42), COPD (491,492, 496), silicosis (502), presence of cancer (140-239), malnutrition-related disorders such as cachexia, anorexia, abnormal loss of weight (799.4, 263, 783.2, 783.3), and postgastric surgery (564.2). Use of specific medications included aspirin (N02BA), systemic immunosuppressive agents and biologics (L04A), systemic corticosteroids, and disease modifying anti-rheumatic drugs (M01B, M01C). Each individual’s burden of comorbidity was quantified by a combined weighted comorbidity index. This index is an improved Charlson Index that combines the Elixhauser system [21]. The score contains common comorbidities such as myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease, ulcer disease, liver disease, diabetes, hemiplegia, renal disease, neoplasms and AIDS.

We estimated the rate ratio by using a time matched case control-sampling scheme [22]. Incidence rate ratios of active TB (plus 95% confidence intervals [CIs]) were estimated by conditional logistic regression analysis adjusted for all covariates. Disease risk score (DRS) was used to adjust for confounders, and compared the different exposures in a case-control study design [23]. A study-specific DRS was constructed to balance disease risks between different drug exposure groups. On Additional file 1, we reported the c- statistic (0.81) of the DRS model, component variables, and the respective weights of the component variables. The DRS was defined as the probability of developing active TB among the participants unexposed to NSAID or Coxibs based on the individual’s baseline covariates. To estimate DRS, we carried out multivariate logistic regression analysis, where active TB was treated as the dependent variable, and all empirical clinical predictors were treated as independent variables. Hence, the DRS characterize the relationship between potential risk factors and active TB. To avoid the potential unrealistic linear assumption of continuous variables in the regression model, we carried out a fractional polynomial DRS (fpDRS) adjustment.

We also carried out a dose response analyses and subgroup analyses in high-risk patients, to further assess the robustness of our results. Predefined subgroups included sex and age of 70 years. Attributable risk percentage and population attributable risk percentage were also calculated (Additional file 2). All analyses were carried out with SAS 9.3 for Windows (SAS Institute Inc, Cary, NC) and the data are reported in accordance with STROBE guidance.

Table 1 shows the baseline characteristics of the cohort. The source population comprises of 1,000,000 participants. After exclusion of prevalent users of NSAIDs and prevalent cases of TB in the 1 year pre-enrollment period, 123,419 person years were exposed to traditional NSAIDs and 16,392 person years were exposed to Coxibs. Coxibs users were found to increase steadily from 2001 to 2004. However, there is a sharp decline in Coxib users after 2004, when rofecoxib revealed safety concerns and was pulled off the market. The use of Coxib was found to increase steadily again from 2007 to 2011. There were significant differences in the age, and Charlson index comorbidity score of the different participants. In general, Coxibs users were the oldest (75.5 ± 9.8 years old) and had the highest comorbidity score (2.54 ± 2.43), while the unexposed control was the youngest (55.9 ± 19.9 years old) and had the lowest comorbidity score (0.84 ± 1.54). Traditional NSAIDs users had an average age and comorbidity score in between that of Coxibs users and unexposed control. The number of comorbidities, known risk factors and medications also tend to agree with the pattern observed in the comorbidity score. Coxibs users generally had a higher number of comorbidities, known risk factors and medications usage, then followed by traditional NSAIDs users and finally unexposed users.

We identified 340,396 number of participants unexposed to traditional NSAIDs and Coxibs in our cohort, of which 3,206 developed active TB. The crude TB incidence rate for the unexposed control is 0.93% (3,206/343,602 person-years). (Table 1) In our cohort, 1,407 traditional NSAIDs users were found to be associated with active TB, resulting in a crude incidence rate of 1.13% (1,407/124,826 person-years). We also identified 206 Coxibs users associated with active TB, resulting in a crude incidence rate of 1.24% (206/16,598 person-years).

Table 2 shows the rate ratios associating the use of NSAIDs within 31-90 days, on the risk of new active TB onset. The first method, which is matching only on agegroup, gender, and calendar year, can be considered an unadjusted way of obtaining rate ratio (RR). The age-sex-year matched RR is 1.39 (95%CI, 1.24 – 1.57) for traditional NSAIDs and 1.40 (95%CI, 1.03 – 1.92) for Coxibs. Adjustment for individual confounders attenuated the RR to 1.31 (95%CI, 1.15 – 1.49) for use of traditional NSAIDs and 1.18 (95%CI, 0.85 – 1.63) for use of Coxibs. To balance the disease risk in multiple drug exposure groups independent of the changing indications for NSAIDs, a disease risk score (DRS) was created. The effect estimates after DRS adjustment or fractional polynomial (fp) DRS adjustment was similar to the adjustment by individual confounders. In all the different type of adjustment, use of traditional NSAIDs was associated with a significantly increased risk for TB, but the use of Coxibs was not associated with a significantly increased risk for TB.

Table 3 investigates whether the use of traditional NSAIDs with a different duration can affect the risk of active TB. The cumulative incidence of active TB for the non NSAID user (5.74%), is lower than either the 7-14 days of NSAID user (8.95%) or >14 days of NSAID user (9.00%). After fpDRS adjustment of multiple covariates, use of 7-14 days of NSAID ([ARR], 1.46; 95%CI, 1.34 – 1.59) or >14 days’ use of traditional NSIADs ([ARR], 1.25; 95%CI, 1.04 – 1.49) was still associated with higher risk than the reference. However, no duration response effect was observed.