The use of biologic response modifiers that inhibit tumor necrosis factor α (TNFα) have profoundly benefited patients with rheumatoid arthritis (RA) and other systemic inflammatory illnesses. With these great benefits come expected risks, since this proinflammatory cytokine also plays important host defense roles. Serious infections are perhaps the most recognized associated complication of anti-TNF agents, and opportunistic infections, such as tuberculosis (TB), have been reported at a higher frequency in most, but not all, studies comparing users to nonusers of anti-TNF therapies. Despite this concern, TB testing followed by effective TB prophylaxis can facilitate successful and safer use of these therapies. In this issue of Arthritis Care & Research, the study by Hazlewood and colleagues () addresses when to consider TB prophylaxis given the variable risks and benefits associated with the use of isoniazid (INH), the therapy typically given to treat latent TB to prevent reactivation and clinical TB infection.
Using state-of-the-art decision analysis methods, the authors do an excellent job of raising all the major questions, defining nearly all the parameters that frame the clinical conundrums, and comprehensively reviewing the expanding literature on this controversial topic. Like all well conducted decision analyses, we are now able to better “look under the hood” of the complex thought processes that influence clinical management and see exactly what we know and don't know about the factors that influence the fundamental decisions. As the authors note, for many of these questions there is a paucity of top-tier evidence. Thus, using the available evidence, the authors carefully include various assumptions in the decision model and test the veracity of these assumptions using traditional and probabilistic sensitivity analyses. Through these sensitivity analyses, we learn whether the model is sensitive to key permutations in input variables; namely, does the model generate different conclusions if key assumptions are changed? Further, the sensitivity analyses vary the parameters across a mostly reasonable range of other possible assumptions. From a quality of life standpoint, the conclusion of the overall decision analysis is that average patients who live in regions of low TB prevalence, such as Canada or the US, and who are age ≥65 years may not need TB prophylaxis if they have a minimally reactive purified protein derivative (PPD; 5–9 mm). However, the difference in quality-adjusted life expectancy between strategies of no TB prophylaxis (e.g., 9 months of INH) versus prophylaxis was a gain of only 1.1 quality-adjusted life-days, favoring the no prophylaxis strategy. Of note, the opposite conclusion was reached with respect to life expectancy, where there were 0.2 deaths per 1,000 patients from INH-associated hepatitis and 1.4 deaths per 1,000 patients from TB (1.6 deaths total) if prophylaxis was offered, compared to 3.3 deaths per 1,000 patients from TB with no prophylaxis. While no TB prophylaxis may be good advice for some patients, the model is not particularly robust to this conclusion given its sensitivity to some of the key assumptions that were included. Thus, clinicians are left wondering whether they can confidently make the decision to withhold prophylaxis in older adults with low positive tuberculin skin tests (TSTs; 5–9 mm) and an absence of TB risk factors. Importantly, the most appropriate cutoffs to consider a TB skin test positive for RA patients receiving immunosuppressive or immunomodulating drugs are not completely clear. The typical cutoff for a positive TB skin test in healthy individuals lacking risk factors for TB exposure is 15 mm of induration; however, for patients with risk factors for TB, the threshold is lowered (10 mm). For those with strong risk factors of recent infection or significant immunosuppression (e.g., those with human immunodeficiency virus infection, close contacts with TB-infected patients, or patients receiving prednisone >15 mg/day), a cutoff of ≥5 mm should be used (). The generalizability of these cutoffs to regions that use BCG vaccine is also unclear, and the positive predictive value of a TST induration diminishes with size such that the majority of 5-mm TST results in patients lacking risk factors and living in nonendemic regions will likely be false-positives (regardless of BCG history). Further, it is unclear if lumping 5 mm with 9 mm is necessarily appropriate, as their positive predictive values likely differ.
Like all decision models, there are numerous assumptions required and some that can be called into question largely due to the quality (and in some cases limited quantity) of the existing data. One assumption in particular that is potentially problematic is the rate of TB infections associated with anti-TNF therapy. The literature is not consistent on this issue. The estimate of a 5.8-fold increased risk is likely toward the lower end of the risk range, based on some of the larger and more comprehensive studies on this topic. The 3 most recent population-based studies conducted in North America and Western Europe (low prevalence TB regions) each suggested relative risks above this threshold and given the greater years of observation of these studies, presumably TB screening was widespread within these populations ([3-5]). Earlier studies conducted before widespread screening suggest even greater relative risks (). Intuitively it makes sense that the relative risk posed by anti-TNF therapy would be directly related to the a priori risk of the population and the prevalence of TB screening and treatment prior to anti-TNF therapy initiation. Even a slight relative increase in TB risk would prompt a strategy of more aggressive prophylaxis. Additionally, the range of utilities weights (patient quality of life measures, ranging from 1 = perfect health to 0 = the equivalent of death) considered for use of INH are very closely clustered, ranging from 0.95–0.99. A difference in only 0.05 utility units is hardly enough to be clinically meaningful, yet the model is quite sensitive to these assumptions. The major concern with INH, modeled in this analysis, is drug-induced hepatitis. However, perhaps the highest quality study (a prospective study with full denominator data) to evaluate the risk of clinically significant hepatitis suggests a rate much lower than that used in this decision analysis. The study by Nolan et al conducted in a public health TB clinic observed rates of clinically apparent hepatotoxicity of 2.8 per 1,000 in individuals ages ≥65 years (). Presumably, much of INH-related hepatotoxicity is preventable with appropriate clinical and laboratory monitoring, although the safety of INH may be impacted by concomitant medications taken by RA patients (e.g., methotrexate). Other factors to consider in this decision-making process relate to evolving TB testing and treatment strategies. Key issues related to TB testing and treatment that may confound such an analysis are the positive predictive value of the PPD test, newer laboratory-based TB testing approaches, and alternative treatment regimens to lower the hepatitis risk associated with INH. In secondary analyses, the authors examined the use of the newer interferon-γ–response assay (IGRA), although these data were not shown. If an IGRA was chosen over a PPD, given its higher predictive value related to the risk of developing active TB infection, the decision in a positive individual would be to use INH prophylaxis. In addition, the authors considered other TB treatment regimens, including 4 months of rifampin and 6 months of INH. Both of these strategies favored prophylaxis as well, and 4-month rifampin strategies have lower hepatitis risks (approximately 5-fold lower risk than 9 months of INH) and higher rates of completion and acceptance (2-fold more likely to complete therapy) ().
Importantly, the newest prophylaxis strategy of once-weekly INH with rifapentine for 3 months was not considered in the model. This strategy has been shown to have a similar or slightly better efficacy and safety profile than 9-month INH (at least when administered in directly observed fashion as it was studied by the US Centers for Disease Control and Prevention) and it may be gaining a foothold as a way to maximize the cost/benefit ratio of treatment risk, particularly for patients who have difficulty in completing a 9-month regimen of INH (). In addition, the decision analysis did not take into account other potential benefits of preventing TB, namely that of preventing transmission from new cases to other vulnerable individuals. Lastly, the authors are correct in being careful to vary TB risk assumptions to other parts of the world where TB is more common. This analysis may have limited generalizability to regions of higher TB endemicity, primarily due to the fact that the positive predictive value of TB screening tests will be much higher and the potential benefit of therapy accordingly increased.
We commend the investigators for this elegant decision analysis with careful delineations of the key inputs and decision points on a very timely topic. We believe, however, that the ultimate conclusion to this question, based on the currently available data in conjunction with the projected trends toward innovative testing and treatment regimens (both in relationship to TB and for hepatitis), is likely toward more rather than less prophylaxis, particularly with newer and potentially safer TB prophylaxis drug regimens. Regardless of the current conclusions to these particular questions, the approach taken in this analysis needs to become a standard one for an improved understanding of the factors involved in complex decision-making processes in rheumatology. Indeed, one might strongly argue that future clinical guidelines should better integrate formal decision science approaches in order to better elucidate and quantitate the impact of these variable assumptions.