Prophylaxis for Latent Tuberculosis Infection Prior to Anti–Tumor Necrosis Factor Therapy in Low-Risk Elderly Patients With Rheumatoid Arthritis: A Decision Analysis

Authors

  • Glen S. Hazlewood,

    Corresponding author
    1. University of Toronto, Toronto, Ontario, Canada
    Search for more papers by this author
  • David Naimark,

    1. University of Toronto, Toronto, Ontario, Canada
    Search for more papers by this author
    • Dr. Naimark has received honoraria (less than $10,000) from Hoechst.

  • Michael Gardam,

    1. University of Toronto, Toronto, Ontario, Canada
    Search for more papers by this author
  • Vivian Bykerk,

    1. Hospital for Special Surgery, New York, New York
    Search for more papers by this author
  • Claire Bombardier

    1. University of Toronto, Toronto, Ontario, Canada
    Search for more papers by this author
    • Dr. Bombardier holds a Canada Research Chair in Knowledge Transfer for Musculoskeletal Care (2002–2016) and a Pfizer Research Chair in Rheumatology.

    • Dr. Bombardier has received honoraria (less than $10,000 each) from and/or has served on the Advisory Board for Abbott Canada, AstraZeneca, BioGen, BMS, Pfizer, Wyeth, Merck, Schering, Janssen, and Takeda, and has received honoraria (more than $10,000) from Abbott International.


Abstract

Objective

To determine if low-risk elderly patients with rheumatoid arthritis (RA) who screen positive for latent tuberculosis (TB) infection prior to anti–tumor necrosis factor (anti-TNF) therapy should be given isoniazid (INH).

Methods

A Markov model was developed. The base case was a patient age 65 years with RA starting anti-TNF therapy with a positive tuberculin skin test (TST) finding of 5–9 mm, who was born in a country with low TB prevalence and had no other TB risk factors. The decision was 9 months of INH or not. The primary outcome was quality-adjusted life expectancy. Multiple sensitivity analyses were performed.

Results

No prophylaxis was favored, with a gain of 1.1 quality-adjusted life days, but the decision was sensitive to several variables. Prophylaxis was favored for patients ages <61 years, if the relative risk (RR) of TB reactivation with RA alone was >2.5, if the RR with anti-TNF therapy was >5.8, or if the utility associated with INH therapy was >0.98. Prophylaxis was also preferred for patients with a TST result >10 mm and for patients from higher risk countries. If 6 months of INH or 4 months of rifampin were used, prophylaxis was preferred, providing that therapy reduced the risk of TB reactivation by >47% and >27%, respectively.

Conclusion

Withholding prophylaxis prior to anti-TNF therapy may be reasonable for low-risk elderly RA patients with a TST finding of 5–9 mm, although the decision is sensitive to patient preferences. For patients age <61 years from a higher risk country, or with a TST finding >10 mm, prophylaxis is preferred.

INTRODUCTION

Anti–tumor necrosis factor (anti-TNF) inhibitors were the first class of biologic medications introduced for the treatment of rheumatoid arthritis (RA) and remain the usual first biologic agent for patients with an inadequate response to traditional disease-modifying antirheumatic drugs (DMARDs) ([1]). Shortly after their introduction, an increased incidence of active tuberculosis (TB) was reported, which has now been confirmed by several cohorts ([2, 3]). Due to this increased risk, universal screening for latent TB infection (LTBI) prior to anti-TNF therapy was adopted worldwide, with prophylactic therapy given to those who screen positive. In Spain, screening programs have had a significant impact on lowering the incidence of TB reactivation with anti-TNF therapy ([4]). Universal screening, however, may not be appropriate in countries with a lower TB prevalence, as patients who test positive may still have a low absolute risk of TB reactivation and treatment with isoniazid (INH) carries a risk of hepatitis. This is especially true for elderly patients, who have a higher risk of hepatitis from INH ([5]). The British Thoracic Society has recently questioned universal prophylaxis for patients who screen positive, recommending prophylactic therapy only when the annual risk of TB exceeds the risk of hepatitis ([6]). The decision has not, however, been subjected to a formal decision analysis.

The aim of this study was to determine if low-risk elderly patients with RA who screen positive for LTBI prior to anti-TNF therapy should be given prophylactic therapy with 9 months of INH. For the primary analysis, we were interested in patients with a low epidemiologic risk, for whom we hypothesized that no prophylaxis may be favored. We performed secondary analyses across a range of alternate scenarios, including cohorts from different countries with different screening test results and alternative treatment options (6 months INH or 4 months rifampin [RIF]).

Box 1. Significance & Innovations

  • Currently, patients with rheumatoid arthritis (RA) who screen positive for latent tuberculosis (TB) infection (LTBI) prior to anti–tumor necrosis factor (anti-TNF) therapy are offered prophylactic therapy.
  • This study demonstrates that for elderly patients at low risk for TB reactivation, no prophylaxis is favored over 9 months of isoniazid therapy, but the decision is a “close call” and sensitive to patients' preferences and the estimated relative risk of TB reactivation associated with RA and anti-TNF therapy.
  • The therapeutic decision to offer prophylaxis for LTBI should be tailored to the screening test result (size of tuberculin skin test or interferon-γ–release assay result), the patient's age, epidemiologic risk factors, and preferences.

MATERIALS AND METHODS

Description of model

A Markov decision analytic model was constructed. The model estimates the outcome of a cohort of patients by following patients as they transition between discrete health states during specified cycles. We used monthly cycles and a lifetime horizon and terminated the model when less than 0.001% of the cohort remained alive. All analyses were performed with TreeAge Pro 2012 software.

For the base-case (primary) analysis, we modeled a patient age 65 years, presenting with a positive tuberculin skin test (TST) result of 5–9 mm, who was born in a country with low TB prevalence, had never received BCG vaccine, and had no other risk factors for LTBI outside of RA and anti-TNF therapy. The base case was chosen specifically to model a patient for whom there would be the most uncertainty regarding the benefit of INH prophylaxis. A 5-mm cutoff was used to define a positive TST finding, as this is the usual recommended cut point for defining a positive TST result in RA patients prior to anti-TNF therapy ([7, 8]).

The decision branch was prophylaxis versus no prophylaxis (Figure 1). Prophylaxis was 9 months of INH 5 mg/kg/day, which is the current recommended care in North America ([9, 10]). Patients receiving prophylaxis had a risk of developing mild, severe, or fatal hepatitis. At each stage, patients had an age-specific risk of death from all other causes and a risk of developing active TB (mild, severe, or fatal). Patients who received prophylaxis had a decreased risk of TB reactivation dependent on the duration of prophylaxis received. The primary and secondary outcomes were quality-adjusted life expectancy (QALE) and life expectancy (LE), respectively. The expected lifetime cases of hepatitis and TB (mild, severe, or fatal) in a 1,000-patient cohort for each decision were also determined.

Figure 1.

Transition state diagram for Markov model. Each circle represents a discrete health state and the arrows point to the possible health states that patients can transition to at each 1-month cycle. Patients who received prophylaxis had a risk of developing mild, severe, or fatal hepatitis. At each stage of the model, patients had an age-specific risk of death from all other causes and a risk of developing mild, severe, or fatal active tuberculosis (TB) and could relapse. INH = isoniazid.

Model inputs

Annual risk of active TB

The annual risk of active TB was calculated by multiplying the patient's baseline risk by the relative risk (RR) associated with RA, the RR with anti-TNF therapy, and the risk reduction from prophylaxis if received. The baseline risk was determined by multiplying the annual probability of active TB for healthy patients with a positive TST ([11]) by the median positive predictive value (PPV) of a 5–9 mm TST across low-risk countries that have reported TB outcomes with anti-TNF therapy (Canada, France, Sweden, UK, and US) (see Supplementary Table 1, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). Calculation of the PPV of a TB screening test is challenging, as no gold standard exists for a diagnosis of LTBI. These challenges have been addressed and described previously, with an online calculator developed to assist clinicians in the calculation ([12, 13]). We relied on this calculator for the PPV, using a broad range for sensitivity analyses (Table 1). Briefly, the PPV is calculated from the pretest probability of having LTBI and the probability of a false-positive result through BCG vaccination (0% for the base-case patient) or previous nonmycobacterial TB (NTM) infection (estimated from the climate of the country of origin) ([13]). The pretest probability of LTBI is based on the annual risk of infection over the patient's lifetime, which is calculated from the annual incidence of smear-positive pulmonary TB in the patient's country of residence ([13]).

Table 1. Model inputs*
VariableValueRangeReference
  1. RR = relative risk; RA = rheumatoid arthritis; TB = tuberculosis; LTBI = latent TB infection; PPV = positive predictive value; TST = tuberculin skin test; anti-TNF = anti–tumor necrosis factor; INH = isoniazid; OR = odds ratio; RIF = rifampin.
  2. aFor detailed evidence tables, see Supplementary Tables 1–4 (available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract).
Patient characteristics   
Age, years6520–80Assumed
Female, %7560–90([47])
RR of death with RA1.51–2([30, 31])
Death rate from other causesAge-/gender-specific±20%([32])
Risk of developing active TB   
Annual probability of developing active TB in LTBI patients, %0.10.03–0.2([11])
PPV of TST, 5–9 mm0.560.3–0.8([12, 13])
RR of active TB with RAa21.2–3.4([14-20])
RR of active TB with anti-TNF therapy   
Constant risk over time (base case)a4.71.3–15.4([2, 4, 14, 19])
Risk elevated for 6 months, then reverts to baseline4.7–30Assumed
Risk declines after 12 months1–4.7Assumed
Outcomes with active TB   
Probability of severe (disseminated) TB in active TB patients, %a2913–50([2, 14, 17-19, 21, 48])
Probability of death in active TB patients, %a4.11–10([22])
Probability of primary treatment failure, %a235–50([22])
Benefits of prophylaxis   
Risk ratio for developing active TB with 9 months INH0.400.2–0.7([28])
Risk ratio for developing active TB with 3 months INH0.700.5–0.95([29])
Duration of benefit conferred by prophylaxisLifetime10 years–lifetimeAssumed
Toxicity of prophylaxis   
Probability of hepatitis with INH, %a0.2530.1–0.425([6, 24])
Probability of hospitalization in INH hepatitis patients, %a5.20.4–35([6, 24])
Probability of death in INH hepatitis patients, %a20.4–8([6, 24])
OR of developing hepatitis if age >65 years, relative to age 35 yearsa3.61–30([5])
OR of hepatitis in RA patients11–2([27])
RR of hepatitis with RIF compared to INH0.120.05–0.30([38])
Duration of mild hepatitis31–6Assumed
Duration of severe hepatitis63–9Assumed
Quality of life measures (utilities)   
Discount rate0.030–0.07Assumed
Rheumatoid arthritis0.680.5–0.9([33, 34])
Taking prophylaxis0.970.95–0.99([35])
Mild TB0.90.5–0.95([35])
Severe TB0.50.3–0.7([35])
Relapsed TB0.70.5–0.9Assumed
Mild hepatitis0.90.5–0.95Assumed
Severe hepatitis0.50.3–0.7Assumed

The RR of TB reactivation with RA and anti-TNF therapy has been recently reviewed (see Supplementary Table 2, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract) ([3, 8]). The RR with RA alone was estimated from the largest study that compared rates of TB in RA patients on DMARDs relative to the general population ([14]). The RR with anti-TNF therapy was derived from a weighted average of the 3 largest studies that compared rates of TB in RA patients on anti-TNF therapy relative to RA patients on traditional DMARDs ([2, 4, 14]). The range used for the sensitivity analyses captured the 95% confidence intervals (95% CIs) across all studies ([2, 4, 14-20]). The RR with anti-TNF therapy appears to decline over time, with divergent probabilities seen at 3–6 and 12–18 months ([21]). For the base case, we assumed a constant risk over time, but performed sensitivity analyses where the RR was increased for the initial 6 months or declined after 12 months (Table 1). The RR also appears to differ according to the agent used, with infliximab > adalimumab > etanercept ([17, 21]). The range used for the sensitivity analysis would capture differences between agents.

TB outcomes

We defined severe TB as disseminated TB, which appears to be more frequent with anti-TNF–associated TB (see Supplementary Table 3, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). The risk of death and treatment failure was derived from World Health Organization data for the selected low-risk countries (see Supplementary Table 1, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract) ([22]). The duration of active TB treatment was 1 year and fatal TB was assumed to occur after 6 months.

Risks with INH

The adverse event profile of INH is dominated by hepatotoxicity and was therefore the only adverse event outcome considered ([23]). The risk of INH hepatotoxicity has been recently reviewed ([6, 24]). We divided hepatitis into mild, severe (requiring hospitalization), and fatal. For the best estimate of the risk of hepatitis (symptomatic or transaminases >3 times the upper limit of normal [ULN]) and hospitalization or death due to hepatitis, we used pooled estimates of cases after 1984, when monitoring guidelines were published (see Supplementary Table 4, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). A pooled estimate over all the years was used as the upper bound for the sensitivity analysis. There appears to be an age-related increase in hepatotoxicity ([5, 25, 26]). For the best estimate we used data from a large prospective cohort with linear interpolation across all ages ([5]). The risk of hepatitis and severity of hepatitis with INH is theoretically higher in patients with RA taking concomitant medications with a potential for hepatotoxicity, although a study of 44 patients found no cases of elevated transaminases (>2 times ULN) using INH with methotrexate ([27]). We assumed no increased risk of hepatitis for the base case, but included a sensitivity analysis with an odds ratio (OR) of 1–2. The duration of mild and severe hepatitis was assumed to be 3 and 6 months, respectively, and fatal hepatitis was assumed to occur after 3 months.

Benefits of INH

The risk reduction with INH prophylaxis was 0.40 (95% CI 0.31–0.52) from a meta-analysis of clinical trials ([28]). We increased the bounds of the sensitivity analysis to reflect potential differences in compliance in a real-world setting. Patients who developed hepatitis were assumed to have received 3 months of protection of INH, with a risk reduction of 0.7 (range 0.5–0.95) ([29]).

All-cause mortality rate

The baseline death rate was determined by multiplying the death rate from all causes by the RR of death with RA ([30, 31]). The all-cause death rate was derived from Statistics Canada life tables, with sensitivity analyses to capture differences across countries ([32]).

Quality of life measures (utilities).

For the utility associated with RA, we averaged the estimates of 2 studies ([33, 34]). For outcomes associated with TB we used median Health Utility Index-3 measurements from the only available study that directly measured utilities in patients receiving treatment for latent and active TB ([35]). The utility associated with severe TB was derived from the utility of patients reporting severe TB symptoms ([35]). For relapsed TB, we averaged the estimates for mild and severe active TB. The utilities for hepatitis were less well-published. For mild and severe hepatitis, we assumed the same estimates as mild and severe TB with wide ranges for sensitivity analyses. Utilities for a given health state were multiplied by the baseline utility of RA to determine the final utility. Utilities were discounted at 3% per year.

Sensitivity analyses

We performed multiple sensitivity analyses. First, one-way deterministic (simple) sensitivity analyses were performed for each variable by varying the parameter across all possible values (e.g., 0–1 for probabilities) and reporting whether any threshold value (where the preferred decision changed) was within the plausible range. A two-way sensitivity analysis was performed for the RR of TB reactivation with RA and anti-TNF therapy and a three-way sensitivity analysis was performed for age, PPV of the screening test, and the utility associated with INH therapy. Finally, a second-order expected value cohort analysis (probabilistic sensitivity analysis [PSA]) was performed. In contrast to a deterministic sensitivity analysis, a PSA addresses the uncertainty across all key variables. We assigned a distribution for each variable based on its mean and plausible range (see Supplementary Table 5, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract) then drew 1,000 consecutive samples from these distributions and recalculated the expected value of the QALE for each of the 2 strategies. Beta distributions were used for all probabilities and utilities, and log-normal distributions were used for RR and OR ([36]).

Country-specific secondary analyses

The preferred treatment for patients from different countries with different TST results was determined (for model inputs, see Supplementary Table 1, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract) ([12]). We modeled the impact of prior BCG vaccine in patients from the UK, where prior BCG vaccination is common. The effect of BCG vaccine is dependent on the age when the vaccine was received, with vaccines received after age 2 years causing more false-positives on future TSTs. For 3 country-specific analyses, we performed analyses for patients ages 25 years, 50 years, and 75 years and overlaid the results on the three-way sensitivity analysis described above. The purpose was to highlight key threshold values for specific scenarios. Scenarios were chosen to capture high risk (Spain, TST 10–14 mm), moderate risk (US, TST 10–14 mm), and very low risk screening results (UK with prior BCG vaccine age ≥2 years, TST 5–9 mm).

Other secondary analyses

A secondary analysis was performed for patients with a positive interferon-γ–release assay (IGRA; QuantiFERON-TB Gold In-Tube [Cellestis] or T-SPOT [Oxford Immunotec] TB assay). The PPV was 0.98 (range 0.7–1.0), owing to the low rate of false-positive results from NTM or BCG vaccine, although the long-term annual risk of active disease in patients with a positive IGRA is not known ([12]). Secondary analyses were also performed for prophylaxis with 6 months of INH (6-INH) as recommended by some guidelines ([6, 37]) and 4 months of RIF (4-RIF). The monthly risk of hepatotoxicity with 6-INH was assumed equal to 9 months, but patients had less cumulative risk owing to a shorter duration of treatment. The RR of hepatotoxicity with RIF was derived from a systematic literature review of studies comparing 9-INH with 4-RIF ([38]). The efficacy of 6-INH and 4-RIF is uncertain, so we determined the risk reduction that would be required to favor prophylaxis ([10, 38]).

Assumptions

We made the following assumptions: 1) patients who develop hepatitis do not receive further prophylactic treatment but still start anti-TNF therapy, 2) patients remain on anti-TNF therapy for life, 3) the screening test was done prior to immunosuppressive therapy, 4) patients who survive active TB or hepatitis recover completely, and 5) no subsequent TB infection occurs after successful treatment for active TB. We did not consider any societal benefits from the prevention of TB transmission to others.

RESULTS

Base-case analysis

For the base-case scenario, no prophylaxis was preferred with a gain of 1.1 quality-adjusted life-days (Table 2). For unadjusted LE, prophylaxis was favored, with a gain of 2.6 days. The small differences in mean QALE and LE between the 2 decisions are expected given the rarity of the outcomes. For each 1,000 patients, there were 0.2 deaths from hepatitis and 1.4 deaths from TB with prophylaxis, versus 0 and 3.3 deaths with no prophylaxis. While severe or fatal disease from TB was more frequent than from hepatitis (Table 2), all hepatitis occurs within the first 9 months versus TB, which is an accrued risk over a patient's lifetime.

Table 2. Results for base-case scenario*
 ProphylaxisNo prophylaxisPreferred strategyDifference
  1. QALE = quality-adjusted life expectancy; LE = life expectancy.
Main analysis    
QALE, years8.38218.3850No prophylaxis+1.1 days
LE, years12.354912.3477Prophylaxis+2.6 days
Projected lifetime outcomes (cases/1,000 patients)    
Hepatitis    
Any90No prophylaxis9
Severe0.60No prophylaxis0.6
Fatal0.20No prophylaxis0.2
Active tuberculosis    
Any3582Prophylaxis47
Severe1126Prophylaxis15
Fatal1.43.3Prophylaxis2

Analyses of sensitivity

One-way and two-way sensitivity analyses

Prophylaxis was favored for patients ages <61 years and the decision was sensitive to several other variables (Table 3). The preferred decision was sensitive to small changes in the RR of TB reactivation with anti- TNF therapy and RA (see Supplementary Figure 1, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). Prophylaxis was preferred if the RR of TB reactivation associated with anti-TNF therapy was >5.8 and constant over time, or if the RR associated with RA was >2.5. If the RR with anti-TNF therapy was higher for the initial 6 months, prophylaxis would be preferred if the initial RR was >44, outside of the plausible range.

Table 3. Results of one-way sensitivity analyses*
VariableaValueRangeThresholdb> Threshold favors
  1. TB = tuberculosis; LTBI = latent TB infection; PPV = positive predictive value; TST = tuberculin skin test; RR = relative risk; RA = rheumatoid arthritis; anti-TNF = anti–tumor necrosis factor; INH = isoniazid.
  2. aSensitivity analysis performed across all variables listed in Table 1; only those variables with a threshold value are shown.
  3. bThreshold reflects the value where the quality-adjusted life expectancy is equivalent between the 2 decisions.
Age, years6520–8061No prophylaxis
Risk of developing active TB    
Annual probability of developing active TB in patients with LTBI (%)0.10.03–0.20.12Prophylaxis
PPV of TST 5–9 mm0.560.3–0.80.70Prophylaxis
RR of active TB with RA21.2–3.42.5Prophylaxis
RR of active TB with anti-TNF therapy    
Constant risk over time (base case)4.71.3–15.45.8Prophylaxis
Risk elevated for 6 months, then reverts to baseline4.7–3044 (out of range)Prophylaxis
Outcomes with active TB    
Probability of severe (disseminated) TB in patients with active TB (%)2913–5062 (out of range)Prophylaxis
Probability of death in patients with active TB (%)4.11–105.6Prophylaxis
Benefits of prophylaxis    
Risk ratio for developing active TB with 9 months INH0.400.2–0.70.26No prophylaxis
Quality of life measures (utilities)    
Taking prophylaxis0.970.95–0.990.98Prophylaxis
Mild TB0.90.5–0.950.71No prophylaxis

The decision was also sensitive to 2 utility measures; prophylaxis was favored if the utility associated with INH therapy was >0.98 or the utility associated with mild TB was <0.71. Thus, even for the low-risk base-case patient, prophylaxis would be preferred if the patient placed a low disutility on a 9-month course of INH.

Three-way sensitivity analysis

Prophylaxis was increasingly favored in younger patients and as the PPV of the screening test increased, as expected (Figure 2). The preferred decision was sensitive to the utility of taking INH across a wide range.

Figure 2.

Three-way sensitivity analysis for the positive predictive value (PPV) of the screening test, age, and the utility associated with isoniazid (INH) prophylaxis. The preferred treatment for a patient can be determined by plotting the patient's age versus PPV of the screening test, with prophylaxis preferred if the point falls within the green (upper) area. For patients within the hashed lines, the preferred decision was sensitive to the utility (preference) for a 9-month course of INH over the plausible range (utility 0.95–0.99). TBST = tuberculin skin test.

PSA

No prophylaxis was preferred in 75% of Monte Carlo simulations, with a mean gain of 1.5 quality-adjusted life-days.

Country-specific secondary analyses

Prophylaxis was preferred for patients with a TST >10 mm from all countries except the US and in patients with a TST >5 mm from higher-risk countries (Table 4). The differences between countries were driven largely by differences in the PPV of the TST, but also by small differences in TB mortality rates between countries (see Supplementary Table 1, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). For example, the TB mortality rate in France was 5.9%, which is above the threshold value of 5.6 (Table 3); thus, despite being a low prevalence country, prophylaxis was favored for a TST of 5–9 mm.

Table 4. Gain or loss in quality-adjusted life expectancy (QALE) and (unadjusted) life-expectancy (LE) with 9 months of isoniazid prophylaxis for a 65-year-old patient using country-specific risk estimates*
Country of originOutcomeScreening test
TST, 5–9 mmTST, 10–14 mmTST, ≥15 mm
  1. A plus sign (+) indicates prophylaxis is preferred. For detailed country-specific risk estimates, see Supplementary Table 1 (available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.22063/abstract). TST = tuberculin skin test; QALD = quality-adjusted life-days.
CanadaΔ QALE (QALD)−1.5+0.9+2.3
 Δ LE (days)+2.3+4.0+5.0
USΔ QALE (QALD)−2.5−0.8+1.8
 Δ LE (days)+1.0+2.0+3.5
FranceΔ QALE (QALD)+0.2+3.0+4.6
 Δ LE (days)+4.6+7.1+8.5
UK (no BCG vaccine)Δ QALE (QALD)−0.6+1.5+2.6
 Δ LE (days)+3.2+4.8+5.6
UK (prior BCG vaccine age <2 years)Δ QALE (QALD)−1.7−0.8+0.7
 Δ LE (days)+2.4+3.0+4.2
UK (prior BCG vaccine age >2 years)Δ QALE (QALD)−3.8−1.7−1.2
 Δ LE (days)+0.7+2.4+2.7
SwedenΔ QALE (QALD)−1.3+0.4+1.2
 Δ LE (days)+2.3+3.5+4.1
SpainΔ QALE (QALD)+0.9+1.6+1.9
 Δ LE (days)+3.9+4.4+4.6
KoreaΔ QALE (QALD)+1.3+1.6+1.7
 Δ LE (days)+4.4+4.6+4.7
JapanΔ QALE (QALD)+0.9+2.9+4.0
 Δ LE (days)+4.0+5.5+6.3

For UK patients with a history of BCG vaccine, no prophylaxis was preferred for QALE, except in patients vaccinated at age <2 years with a TST of ≥15 mm (Table 4). However, even in patients with prior BCG vaccine and a TST 5–9 mm, the decision was generally still sensitive to patient preferences (Figure 2). Only in patients age 75 years with BCG vaccine at age ≥2 years was no prophylaxis favored across the plausible range for INH preference (Figure 2). For patients ages 25 years or 50 years from Spain and the US with a TST of 10–14 mm, prophylaxis was preferred and the decision was not sensitive to preference for INH (Figure 2).

Other secondary analyses

Prophylaxis was preferred for any patient with a positive IGRA, although the PPV threshold below which no prophylaxis would be preferred (0.70) fell just within the plausible range. Prophylaxis was preferred for 6-INH and 4-RIF if the risk ratio was <0.53 and <0.73, respectively (lowered risk by >47% and >27%).

DISCUSSION

The results demonstrated that for RA patients ages >65 years who are otherwise at low risk of TB reactivation, withholding prophylaxis (9 months of INH) prior to anti-TNF therapy may be reasonable. The absolute benefit was small and the decision was sensitive to several variables, including RR estimates and quality of life adjustment, emphasizing that the preferred decision is a close call and prophylaxis is also reasonable. For patients with a TST >10 mm or a positive IGRA and patients with a TST >5 mm from a higher risk country, prophylaxis was preferred. For UK patients with a history of BCG vaccine, no prophylaxis was generally preferred, although the decision was still sensitive to patient preferences. Prophylaxis was less favored as age increased, as older patients have a higher risk of hepatitis and less life-years to benefit from prophylaxis. Prophylaxis is likely preferred in low-risk patients with either 6-INH or 4-RIF, owing to the shorter duration of treatment and less hepatotoxicity. The long-term efficacy of 4-RIF is currently uncertain, but a trial is underway and would clarify the decision ([39]).

Our decision analysis has several advantages over a simple comparison of risks, which is commonly used when developing treatment recommendations. The British Thoracic Society recommended prophylaxis prior to anti-TNF therapy only if a patient's annual risk of TB reactivation exceeds the risk of hepatitis ([6]). This neglects several important considerations. First, the risk of hepatitis occurs initially versus the benefit from prophylaxis, which is accrued over a patient's lifetime. Second, hepatitis and TB are different outcomes with different mortality rates. Third, patients' preferences for the nonfatal health states may differ and, as demonstrated, are an important consideration when deciding to offer prophylaxis. Finally, through sensitivity analyses we were able to identify which variables are key drivers of the decision so that therapeutic choices can be personalized. With this in mind, we presented a three-way sensitivity analysis to demonstrate how 3 key variables (age, screening test result, and preference [utility] for INH) affected the preferred decision (Figure 2).

Limitations of the study include the uncertainty around the estimation of model inputs, although these are a reflection of the data available and not the model itself. The RR of TB with anti-TNF therapy is difficult to ascertain, as screening for LTBI is routine and most patients who screen positive are treated with prophylaxis. Early risk estimates from a US study (before LTBI screening was universal) were in line with our estimate, although this study identified TB cases through voluntary reporting ([2]). We may have underestimated the toxicity with prophylaxis in older patients. One study using administrative data found hospitalization rates for hepatic events in patients ages >65 years treated for LTBI of 2.6% versus 0.2% for untreated age-/sex-matched controls (adjusted OR 3.2 [95% CI 0.9–11.7]) and 6% versus 0.8% (adjusted OR 3.8 [95% CI 1.8–7.8]) for other toxicity outcomes ([40]). These data could not be incorporated in our model as using the adjusted OR as an estimate of risk would require a baseline risk of toxicity in patients not treated with prophylaxis. We addressed the uncertainty in model parameters by using broad ranges for the sensitivity analyses. We did not consider the possibility of TST anergy in RA patients receiving medications with significant immunosuppressive effects (≥15 mg prednisone), as the TST is not recommended in this setting ([6, 10]). Finally, we were unable to test the individual components of the PPV calculation in a sensitivity analysis, but instead used a broad range for the sensitivity analysis of PPV itself.

The discordant results for QALE and LE indicate that although prophylaxis provides a small gain in LE, this is offset by patients' preferences. Our results are in line with other decision analyses that have examined the benefit of prophylaxis in low-risk patients; prophylaxis is preferred for unadjusted LE ([41-43]), but QALE is sensitive to the utility associated with INH therapy ([41]). The key driver of this difference is the utility assigned to taking INH therapy, which is the patients' preference for LTBI treatment outside of any risk of hepatitis. Our best estimate (0.97) was derived from one published paper that measured utilities in patients being treated for latent TB, and has been used in other decision analyses ([35, 44]). Other groups have used higher values (0.99), but these were based on no direct evidence ([37, 45]). A value of 0.97 equates to a loss of 8 days over 9 months, suggesting patients would be willing to trade a reduction of 8 days in LE to avoid taking INH. In the sensitivity analysis, prophylaxis was preferred for a utility >0.98, again emphasizing that the decision is a close call.

There is considerable inconsistency in screening for LTBI in patients who are at increased risk of infection. Screening has become mandatory prior to anti-TNF therapy, whereas patients are not routinely screened prior to dialysis, which carries an RR of 10–25 ([10]), or prior to moderate doses of corticosteroids (≥15 mg), which likely confer a greater risk than anti-TNF therapy ([46]). Our results indicate that RA patients ages ≥65 years starting anti-TNF therapy are at a “risk threshold,” whereby withholding prophylaxis may be reasonable. Our results should not, however, be interpreted as demonstrating that these patients should not be offered prophylaxis. In fact, in all but the lowest risk patients, the decision favored prophylaxis and even in these patients the decision was sensitive to quality adjustment. The total number of cases of TB prevented was also greater than the number of cases of hepatitis. Our results support the use of 6 months of INH as recommended by some guidelines ([6, 37]), and likely 4 months of RIF. The decision should be individualized to the patient's age, TB risk factors, screening test result, and preferences after an informed discussion. Patients and clinicians should be aware of the “up front” risk associated with prophylaxis versus the lifetime benefit of reducing the risk of TB reactivation and that prophylaxis is not perfect. Ongoing surveillance of biologic agent–associated TB incidence and safety of prophylactic therapy through observational studies will help refine risk estimates and add clarity to this important therapeutic decision.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Hazlewood 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 conception and design. Hazlewood, Naimark, Bykerk, Bombardier.

Acquisition of data. Hazlewood, Gardam.

Analysis and interpretation of data. Hazlewood, Naimark, Gardam, Bombardier.

Ancillary